TalosII ppc64el computers started shipping here's an unboxing by tenfourfox: https://tenfourfox.blogspot.com/2018/04/unboxing-talos-ii-its-here.html?m=1
Anyone here got one yet? Mine's still waiting to be shipped.
Ye, we're getting closer to summer, I guess.
He also forgot the archive.
Autism. 99.9 percent of users will never need this. If it proves to be significantly better than the alternatives in terms of performance, price, and compatibility then I'm all for it but as of right now its not
The RaptorCS guys told me a few weeks ago mine was shipping early this month, still need to buy everything else for it since I only got the mobo and twin 8 core CPUs.
This >>907242 mostly, but I also have code I need to port to Power9 and developing highly optimised ISA specific code is basically impossible if you don't have a native system to test it on and fuck using cloud infrastructure.
Hope that these buyers will make it possible for a cheaper machine to exist after that.
and then we'll finally get the ultimate power laptop
It will be difficult to say if the price will drop in the future, we probably wont see a new workstation design until Power10 comes out in a few years and I don't think the demand exists for a cheaper desktop-oriented model although they could make one by taking the existing dual socket design and removing the second socket and associated hardware.
no-one is exempt from botnet in $CURRENT_YEAR
>Management (BMC) interface
let's just take the botnet and make it open source botnet and it's okay now.
File: f0b0fe838c84a67⋯.jpg (Spoiler Image, 130.88 KB, 937x937, 1:1, 8e4b6c3f708d8fdc33f060dd98….jpg)
>TalosII ppc64el computers started shipping
>shilling dead-end PPC
>the goyim don't want the botnet and want open source
<make them pay for the price of their data ahead of time.
its cheaper than a budget xeon and a middle end gpu lol.
Also you have no idea how much big companies depend on obscure ISA's to run their hacked together 30 year old software.
>hacked together 30 year old software
More like 50+ year old software, the Z80 is still in production for a reason...
Because it's a proven design that's cheap to make and capable for its price?
I wish I had the money for this thing..
> still in production for a reason...
And PICs too.
There a still a few thousands of people in corps around the world who know how to fucking exploit hardware, when you make a blender you don't need a 8 core arm cpu to make it works even if it as fucking IoS networking.
Me too my friend me too, shill it for freedom and maybe at some point it will be affordable for poorfags like us.
I really can't wait till there's affordable RISC-V stuff, even just a pi-like SoC, with support from a distro. I want to make a NAS or something out of it.
Sadly on thing available now is a devkit meant for debian porters and things like that.
4000 dollars what a bargain. I rather get fucked by the botnet then pay 4000
>I rather get fucked by the botnet then pay
>4000 dollars what a bargain
Have you ever bought a workstation?
4k isn't much at all.
If I had $4K to spend, I'd just get an Amiga 3000 or two with all the goodies instead. It's not like I'm running an Internet server or whatever that needs latest workstation. Plus, I'm not even that big a fan of Linux/BSD anyway.
>a $4k workstation that's unable to run most professional softwares because they're only available on Windows.
Its like buying a racing boat to sail in your backyard swimming pool.
you autists need to face the truth, linux is NOT a productivity OS
because the price of this thing is extreme jewry.
for this price you could get the newest jewtel processors and risk burning up 10 motherboards trying to get me_cleaner to work with it.
Linux is a productivity OS for developers, easily.
Enjoy your proprietary platform firmware
enjoy burning up 5 grand on something can can be solved with me_cleaner and a $20 router with a firewall.
i'm not opposed to this but the pricetag is bullshit.
>can be solved with me_cleaner
That's not a complete fix. A BlackHat presentation explained that the ME can still be exploited even if you me_cleaner it or use the HAP bit thing
You can buy just the mobo and CPU for $2500.
>for this price you could get the newest jewtel processors
the point is not to do that
For 5 grand I can buy the TALOS mobo with two 18-core POWER9 CPUs. What x86 system can you spec in this price range?
It's only the mobo that is expensive due to being a low-volume custom part. The CPUs are surprisingly cheap. As a result, the higher you spec your TALOS system, the more cost-effective it becomes.
I don't care much for security, I only want the autistic feeling of being in control and owning my machines.
You're just like NSA, anon. They also want the feeling of being in control and owning your machines.
Agreed. Linux is a kernel.
Much like the situation at ski resort where young women are looking for husbands, and husbands are looking for young women. But the relationship is not as symetrical as it may seem at first glance.
No, because there is a fucking huge codebase for them, for many applications its easier to just use a chip from the 70s than it is to re-write the code you have for a newer chip.
>And PICs too.
I forgot about them, I miss working with PICs, simpler times.
>for this price you could get the newest jewtel processors
A single Xeon Platinum 8160 costs $1000 more than the entire dual 8 core Talos system, has less threads, less cache, lower base and turbo clocks, and less memory channels. Even an EPYC CPU comparable to what two 8 core Power9 CPUs provide is more than the entire Talos system.
And even after you buy the newest x86 CPU you still are forced to run binary blobs on undocumented and highly privileged processors.
I assume you have some evidence to corroborate your assertion
>undocumented and highly privileged processors
where's the documentation on these processors?
I wasn't meaning to imply that the Z80 wasn't a proven design, just that the main reason why they are still around today is because of the codebase which exists for them.
>where's the documentation on these processors?
Undocumented can also mean that no documentation is publicly available, Intel has internal documentation on those parts of their CPUs but don't release it for obvious reasons.
and where are the cpu schematics for this power 9 processor?
that's the firmware, where are the cpu schematics?
they're behind a 6 figure paywall which requires you don't release them
>Undocumented can also mean no documentation publicly available
all you gullible faggots dropped $5k on this system because it was "open" yet none of you are considered privileged enough to view these "open" documents.
so basically, they felt for the "open" meme and got jew'd?
they lost $5000
You just click the links in the openpower page, the IBM portal needs a user account but openpower page is fine, at least on all the links I've tried.
Everyone semi-literate understands that you can just git the firmware source code and download the spec sheets, this is an open software machine, not yet open silicon.
The FSF has endorsed this machine.
I would like to have a machine where even the silicon itself is libre but no one is going to get there if the steps taken to get there are attacked for not being perfect.
Shitpost on your botnet machine because the alternatives don't reach the nirvana fallacy which you don't even uphold.
what's the point of "open" firmware if you have no idea what's on the chip? for all anyone knows the firmware goes on one piece and there's a completely seperate botnet somewhere else.
The sentiment is great but it's meaningless if the hardware isn't open and it definitely doesn't justify the price tag.
>PowerPC 64-bit EABI little-endian
IBM just calls it an ABI (not "embedded") in their docs, why the fuck does debian want to call it "el"?
I can't wait to buy one of these second hand on fleabay with my yearly salary. Kinda jealous.
If anything, this serves as a testament of why closed source software is shit; you can't get it to run on another architecture and are wholly dependent on the vendors making their applications software available for it.
What's the point of open anything unless you can scan each layer of every single CPU?
Reducing the unknowns is better, otherwise we'd all stick to windows and solaris.
This machine has potentially an infinite support cycle by the community, even when IBM stops supporting POWER9 you can still patch vulnerabilities.
Spanish: endian little
English: little endian
Nothing you can do about it gringo.
>linux is NOT a productivity OS
What are the top 100 OS used in supercomputers ?
>what's the point of "open" firmware if you have no idea what's on the chip? for all anyone knows the firmware goes on one piece and there's a completely seperate botnet somewhere else.
>The sentiment is great but it's meaningless if the hardware isn't open and it definitely doesn't justify the price tag.
Do you remember how history went ?
Before the mid 80s everything when software was shared there was no license, blueprints were available, and then progressively each years some hardware and software was closed it was made more obscure.
To be honest I think it's a miracle that we are slowly going backwards on this, 100% libre software/hardware will only be achieved by slow upgrades with the years passing by and I hope that when we achieve that we won't go back to the nightmare we actually are living.
I just hope that people won't make the mistakes we made I don't want people to suffer like today all these suspicions all these abuses are damaging morally and mentally to everyone.
>A single Xeon Platinum 8160 costs $1000 more than the entire dual 8 core Talos system, has less threads, less cache, lower base and turbo clocks, and less memory channels.
It can run in 8-socket configs and has full-width AVX512 execution units though.
to distinguish it from older, obsolete ppc64 ABIs
>It can run in 8-socket configs
In theory so can the larger socket Power9 CPUs, the ones in the Talos system are the equivalent of the W-Series Xeon while IBM has also has a larger version which is for servers. Also the Xeons in the 8 socket config still are only limited to being directly connected to 3 other CPUs since each only has 3 UPI links.
>full-width AVX512 execution units
True, but to be fair if you are at the point where a significant portion of your code is 512 intrinsics then you may as well just port it to CUDA or OpenCL.
If you are willing to take a step back in terms of processing power you can get most of the way to 100% libre software and hardware. Most microcontrollers have massive manuals which describe in great detail each register and how each module operates, some have display hardware and many have USB. A computer made with these could only get more libre if the VHDL of the silicon was released.
POWER9 has it's own unique SIMD, apparently it is the only processor to support AltiVec 3 instructions. I was able to find some of these instruction's names but I cant find ANY kind of complete or useful documentation on POWER assembly, let alone AltiVec 3. So if anyone has any details about it, I hope someone can share a link in the thread.
Every vector instruction in the OpenPower ISA for the Power9 is in this document, page 223 is where the vector instructions start
Instruction set doesn't tell you anything on its own about performance. You can have 512-bit wide vectors and implement them with 128-bit EUs by repeating each insn 4 times, for example. POWER9 is known to have weak vector units, since its main target market (high-end servers) doesn't have much use for them.
Well you don't need strong CPU vector performance when you have 8 P100 GPUs connected to the CPUs via NVLink
In many workloads, yes. However you still suffer sync latencies if your workload requires constant back-and-forth data shuffling between sequential and vector algorithms. There IS still a market for CPUs with heavy vector units.
Thanks a lot for this, I will have to go through it.
I understand, I just wanted to point out that POWER9 has it's own SIMD since
>you should be grateful to be able to pay the IBM "open" goy fee.
this is essentially what this is. IBM charges $100,000 + for the priviledge of being able to look at the data sheets, which talos did, and then passes the cost on to the customer.
Would you support this if Intel or AMD came out and said alright, you can look at PSP and ME, but you have to pay us $2000 first for every chip sheet you look at and you have to sign a NDA ensuring you don't tell anyone else about what you find, ensuring that they also pay the $2000 goy fee.
It's better than nothing I guess, but I'm not paying the goy fee.
also the cost get's passed on to you, so talos can look at the datasheets, but you can't.
Wait, we can't look at the datasheets for the talos? Then its fucking botnet.
>using proprietary software
>I would like to have a machine where even the silicon itself is libre but no one is going to get there if the steps taken to get there are attacked for not being perfect.
The only solution is running everything on open-toolchain FPGAs.
To free yourself from Wangbozos 10
Is there any libre FPGA silicon out there?
>libre FPGA silicon
oh my fucking god don't post
How do you think those FPGAs are implemented? With fairy powder and unicorn farts? How do you know they're not botnetted behind your back?
>>908292 just shifted trust from the CPU manufacturer to the FPGA manufacturer. It doesn't make anything better.
no, you are not allowed to look at the datasheets, not unless you pay $100,000 and then sign an NDA. You have to take IBM's word for it, just like you have to take Intel's word for it or AMD's word for it, only their processors are half the price.
>only their processors are half the price.
<8C/32T POWER9: $595
>12C/24T Threadripper: $799
>16C/32T Threadripper: $999
<18C/72T POWER9: $1375
>24C/48T EPYC: $1850
<22C/88T POWER9: $2575
>32C/64T EPYC: $3400
Intelavivshit costs even more, so I left it out from the comparison.
Save for the somewhat overpriced 22C part, the POWER9 CPUs in Raptor's offer are priced very competitively. Especially given that they trade punches with x86 parts that have 50% more cores.
you have a point these aren't any more expensive at this level. xeon's are more expensive than this list.
What kind of case is this? What are these blue things?
4 threads per core? WTF? Apple is fucking gay for dropping the PPC arch.
POWER9 actually has 8-way SMT, not 4-way SMT, see:
But the CPU's availible for Talos II don't seem to have it enabled.
This is bullshit. POWER9 has 4-way SMT.
There exists a special version of this processor that "fuses" every two 4-way cores into a vertically clustered 8-way "core", giving half the core count and double per-core thread count. There is absolutely no technical advantage to doing that, as every thread can access only resources of half the fused "core". However it lets the buyer jew software vendors who jewishly price licenses per core of the target machine, which is a common practice in big-iron corporate software. Jewing jews is the sole reason this version exists.
The 4 and 8 way SMT is due to differing requirements among VM and bare metal applications. The VM guys wanted less cores with more threads each while the bare metal guys wanted more cores with less threads each. The solution was to divide the Power9 die into sub units called slices which have two 4-way SMT cores each or one 8-way SMT core each depending on if the application. The biggest 4-way SMT CPU is 24 cores, the biggest 8-way SMT is 12 cores.
>4 threads per core? WTF? Apple is fucking gay for dropping the PPC arch.
PPC was a lot different back then, its only through IBM slaving away at it that its gotten good.
But then why do the benchmarks on that site say that it makes it faster? I've been reading a little bit about this stuff but I don't really understand POWER9 that well.
Can you install Windows on it?
>But then why do the benchmarks on that site say that it makes it faster?
Thats for a very specific workload involving databases. Database performance is often very sensitive to changes in system memory access patterns since most of the time the software is reading and writing to large amounts of data in RAM, due to the way modern CPUs handle things like cache coherency and system memory accesses its more efficient in the case of Db2 running on a Power9 CPU to have a single core with more threads than it is two cores with the same number of total threads.
Find a Windows NT for PPC and install that.
You can install Windows10 on an x86 system emulated inside QEMU.
That will be big-endian.
Why? Do you not understand that Windows is proprietary software that disrespects your freedom? What's the point of buying this computer for the sake of "security" and then installing Windows on top of that?
Windows NT for PPC exists. They didn't change the arch endianess between that period.
PPC/POWER can do big or little endian. It doesn't care. You aren't going to get Windows NT 4.0 to boot on this irregardless. It would be better to use QEMU, and probably faster even though emulation will be slow as shit.
>PPC/POWER can do big or little endian.
Yes I am aware of this, my point was that some programs won't work correctly however it just occurred to me that basically none will work because its PPC and not x86.
Apple had a bit of trouble when they switched from PPC to x86, they had an emulation mode where programs developed for PPC would be able to be run on the new x86 Macs but a surprising amount of programs didn't work correctly because they had endian-aware code and the switch from BE to LE broke them.
point and laugh
>what's the point of "open" firmware if you have no idea what's on the chip? for all anyone knows the firmware goes on one piece and there's a completely seperate botnet somewhere else.
>The sentiment is great but it's meaningless if the hardware isn't open and it definitely doesn't justify the price tag.
Does that say PowerPC?
>it definitely doesn't justify the price tag
Most people who brought a Talos system did so because a P9 server costs an order of magnitude more... security autists are only a minority of people who brought them. Of the two people in this thread who actually purchased a Talos system one did it because of the first reason and the other didn't say.
>what's the point of "open" firmware if you have no idea what's on the chip?
The micros on the Power9 CPU which handle things like booting and power management are all PPEs, which are basically just stripped down PowerPC 405 CPUs. The Talos wiki has info on the various subsystems and the code they run including links to source code. Also you don't have to pay to get access to all the OpenPower documentation if you are just an individual for academic/personal use if you want more info.
>for all anyone knows the firmware goes on one piece and there's a completely seperate botnet somewhere else.
Even if they released the VHDL for the CPU there is no guarantee that its actually correct or that the fab didn't add a small microcontroller without IBM knowing. The only way to be sure that you aren't running any botnet is to literally design and fab a CPU yourself in your garage, so get too it.
So, what all has been ported to POWER9? I'm specifically interested in kdenlive, darktable, GIMP, libreoffice (calc and writer) and Firefox. What is the state of x86 emulation on it, as well.
Are you purposely being retarded? You couldn't install that on PPC Macs and you wouldn't be able to install it on current year x86_64 PCs either.
All have PPC64le packages available in the Debian repo, they also have ports for ARM/MIPS/etc.
Can't find a PPC64le package but considering there are 64 bit ARM packages available it should compile for PPC64le just fine.
>What is the state of x86 emulation on it, as well.
Slow, just like all emulation.
Having the vhdl would be very satisfying imo not for trust purpose (it would help tho) but helping solving or understand Errata would be simpler.
That's fine. I can do with GIMP-only, or else unironically use Gentoo.
You might wanna try literally anything else in a vm instead
>but muh performance / slow emulation
I'm running Vegas in a Windows vm (x64 host and guest, no IOMMU) and despite the vm having fewer resources (obviously), Vegas still outperforms the natively running kdenlive by factor 3. While rendering videos at a higher bit rate. And not crashing every 20 minutes.
If you want to use a floss alternative, maybe try blender, it's supposed to be a good video editor, once you learn to use it. All I'm saying is, don't use kdenlive, it's one of the worst editors I ever used.
One of the things I've noticed in recent years is that many projects which once had Altivec optimizations have completely removed them due to lack of interest and unfixed bugs. Thankfully it looks like GIMP didn't gut support. It's absolutely embarrassing to see NEON get preferential treatment, but I guess this is the bed IBM made for themselves.
The downside to blender is that you'll need to render stuff out multiple times if you are doing complicated stuff.
kdenlive has worked well for me in the past. It has everything you need unless you want to be super fancy.
>It's absolutely embarrassing to see NEON get preferential treatment
It makes sense, there are lots of cheap ARM single board computers available and as of AARCH64 NEON is standard. The number of people who want to run GIMP on their RPi or Beaglebone is greater than the number of people who want to run GIMP on their PPC64 system,
TLDR IT'S NOT HiFive+HDMI, NOT AN OPEN CHIP, FUCK OFF AND DIE NIGGER
Wasted trips, SiFive's chips are less open than Power9
>go to SiFive webside
>Freedom U500 Dev Kit
>its a fucking Virtex-7 dev kit from Xilinx
>they don't provide you with the full RTL, just a bitstream and a document less than 100 pages in length
>"trust us you dumb fucks, its open, just buy this $3600 dev kit and flash this bitstream to it, what more do you need?"
>[hand rubbing intensifies]
The only RTL they supply is a stripped down evaluation package (after you sign up to their website and agree to their terms and conditions of course), not the full one because they want you to pay for it after you sign an inch thick stack of NDAs and other legal bullshit.
>no RTL, just PCB design files and their shitty Eclipse clone
>barely 300 pages of manual for the chip and dev kit, a fraction of the amount of documentation thats available for the ebul freedom hating ARM chips used in the Raspberry Pi
>$1000, only 8GB RAM, no display output (not that it would matter since the RISC-V ISA has no accepted vector instructions yet)
>"limited quantity, buy now! dont you want freedom?"
>[more hand rubbing]
The worst part is that because the RISC-V ISA is still 'in development', there is essentially no guarantee of forward or backward compatibility for code. Your $1000 inferior RPi clone may not run some RISC-V code developed in a years time for the latest ISA because of minor (or major) changes to the instruction set or core architecture.
There is a link in this thread to a 1200+ page manual detailing just the instructions of the Power9 CPU, there is barely 150 pages of documentation for the U540 CPU.
>pay through the nose for a botnet-free computer
>install botnet on it
I for the most part agree with you, currently POWER is more open in practice.
>The worst part is that because the RISC-V ISA is still 'in development', there is essentially no guarantee of forward or backward compatibility for code.
I don't think this is fair, the people who made the ISA weren't students, they've been steadfast against featuritis and already have Nvidia and Western Digital using it, I doubt they'd break it now.
>famous last words
You can compile it without networking support, Raptor even provide instructions on how to do this. There's even a coreboot port in the works for fuck knows why.
>Nvidia and Western Digital using it
They don't suffer from the same problem
>Nvidia chooses V1.XX of the ISA
>uses it for a generation or two
>Nvidia is the only one who will ever write code for this chip
>a few years later they choose to upgrade to V2.YY
>company A makes a RISC-V chip using the latest version of the official ISA
>six months later official ISA gets updated
>company B makes a RISC-V chip using the latest version of the official ISA
>six months later official ISA gets updated
>company C makes a RISC-V chip using the latest version of the official ISA
>six months later official ISA gets updated
>you now have a fractured instruction set, with people on different and potentially incompatible versions of the ISA
RISC-V is still years away from being mature enough for large chip vendors to adopt it, those that are doing it now are small startups just trying to cash in on hype.
>they've been steadfast against featuritis
They are constantly making small changes to the way things currently work as well as adding new features. There are literally vector extensions being proposed/prototyped and if it gets in (which it most likely will) its going to have to be standard to avoid the fiasco that occured as a result of NEON being optional in pre 64bit ARM chips. You don't understand the politics that has evolved around RISC-V since Softbank brought ARM.
I am jelly as fuck, mine hasn't shipped yet.
POWER is not POWERPC. PPC did outperform x86 for a long time - but this performance had a linear relationship with heat/power usage; they were total housefires (to be fair, it was made for mainframes). Look at the heatsink design of the G5, it's a monster. They also had problems with vendors not meeting their demand. Oh, and Steve Jobs was an idiot who didn't want fans in his computers.
>Look at the heatsink design of the G5, it's a monster.
Compared to modern high-end coolers, it's pretty tame. Ever seen a Noctua NH-D15?
>RISC-V is still years away from being mature enough for large chip vendors to adopt it, those that are doing it now are small startups just trying to cash in on hype.
I just named some of the biggest players in the industry that have adopted it.
>They are constantly making small changes to the way things currently work as well as adding new features.
RISC-V is built with a modular approach, what most people think of as RISC-V is actually RV64GC which is RV64IMAFDC, each letter proceeding 64 is just an extension, all of those I named are frozen, the fact that they're working on other modules like V isn't surprising and doesn't encroach on the core of the other extensions.
>You don't understand the politics that has evolved around RISC-V since Softbank brought ARM.
You're right, I don't.
>Ever seen a Noctua NH-D15?
Not exactly a fair comparison since those produce 20dB of noise.
The ISA bus lives?
Instruction Set Architecture. And yes, the ISA bus does live.
(09:32:43 PM) qweo: Is there any hope at all of Talos becoming less expensive in foreseeable future? Is it largely a problem of volume?
(09:33:11 PM) tpearson: qweo: We're working something big, so there might be hope, yes :)
(09:35:32 PM) duncan^: a big announcement or a big box? or a big surprise?
(09:36:30 PM) tpearson: duncan^: Announcement, for people that can't afford a full dual socket Talos box
(09:37:17 PM) duncan^: That's good. I guess this is what was being discussed previously?
(09:38:35 PM) tpearson: well, let's wait for the official announce :)
(09:40:16 PM) qweo: Is it far away?
(09:41:13 PM) tpearson: shouldn't be, let's say under a week
(09:41:34 PM) tpearson: otherwise I'd be in trouble for leaking hints here ;-)
(09:41:44 PM) jn__: wow, that's pretty soon!
(09:42:03 PM) qweo: Wow!
(09:43:00 PM) tpearson: mind you, it's not going to be ultra cheap, but let's just say more affordable if you don't need a second CPU
(09:43:42 PM) tpearson: the simple fact is that libre systems are always going to be somewhat more expensive due to the extra work required to develop and maintain them
(09:43:55 PM) tpearson: but we know the Talos II itself is out of reach for people
(09:46:29 PM) RobbieAB: tpearson: Well, surely once you get past the initial pain, the higher-cost factor should fade a little?
(09:47:38 PM) tpearson: RobbieAB: Of course, yes, but at the same time there are people that literally can't afford the Talos II mainboards right now, they want them but they are out of reach (being a relatively high end server / workstation board)
(09:49:46 PM) RobbieAB: tpearson: Oh, I get that side of it, I was more thinking that, assuming this is successful longterm, the major pains will slowly fade as A) what we have get's more robust and stabilised (beyond anything in the proprietry space!) and also we will slowly assimilate more component options?
(09:50:01 PM) tpearson: yes, that's the goal!
From their website
>Additional Sforza CPU options are planned for Q2 2018
I'm guessing they're going to start binning CPUs so the price doesn't start at $375.
The real problem is their $2,325 motherboard, hardly anyone needs dual sockets or 16 DIMM slots. They've already made a variant with the SAS ports removed, so they'll probably get rid of a socket and a few PCI slots next.
Also the guy who did the unboxing in the OP has an update.
POWER9 has both variations, it depends on your workload, the Talos II has sforza which is SMT-4 way.
I doubt the CPUs will change, those options refer to 18 and 22 core options I bet.
The CPU prices aren't a problem, they're cheaper than intel when it comes to price per dollar.
It's the motherboard that's the killer, but even then, people expect the world for a penny, there will be people complaining that these aren't raspi prices no matter how much he drops it.
mITX POWER9 gaymer RGB wifi board when?
>I just named some of the biggest players in the industry that have adopted it.
I was meaning companies like Microchip, Texas Instruments, Nordic Semiconductor, etc. In other words companies who sell chips which are going to be programmed out in the wild by a variety of people with a variety of toolchains and other variables, those companies can't have an ISA which isn't mature enough yet to get an industry agreed upon "this is what a XX-bit RISC-V processor is". Its getting close though.
It doesn't matter if WD is using RISC-V because only WD engineers will be programming those chips, it doesn't matter if Nvidia is using RISC-V because only Nvidia engineers will be programming those chips.
>RISC-V is built with a modular approach, what most people think of as RISC-V is actually RV64GC which is RV64IMAFDC, each letter proceeding 64 is just an extension
The issue is that its optional, as an end user of a RISC-V micro I want guarantees that I am getting a certain set of extensions when I buy a 32-bit micro for my embedded product. When I buy a microcontroller with a Cortex-M4 I know what I am getting in terms of the processor.
>You're right, I don't.
So basically a group of big companies gave $98 billion to Softbank to invest in whatever they thought was the future, one of the biggest contributors was Apple. The first big thing Softbank did was buy ARM Holdings outright. This is a problem for Google since Androids biggest competitor now indirectly owns the company which every chip vendor relies on for IP, placing them in a potentially compromised position, not to mention its not good for the IoT goals of Google and many other large tech companies.
It was around the time ARM Holdings was brought by Softbank that Google began quietly stepping up their involvement in RISC-V.
What will happen is in maybe a year a handful of versions will become standardised
>a low power 64-bit version targeted at phone applications as a replacement for the ARM "LITTLE" processors
>a high performance 64-bit version targeted at phone applications as a replacement for the ARM "big" processors
>a SIMD version of the ISA intended to be used as the GPU in conjunction with the above in phones
>a low power 32-bit version targeted at IoT and other embedded applications
>an ultra low power 16-bit version targeted at IoT and other embedded applications
Google and one or two chip vendors (hard to say who but it will be companies like Broadcom or Microchip) will be the ones leading this effort and will announce that products are in development around the same time, Google will announce that the next Pixel will feature a RISC-V processor. Following this other chip vendors will announce RISC-V products now that there is essentially a narrow set of standardised versions to pick from amd to avoid being left behind by their competition.
The Pixel will be the biggest driving force behind RISC-V adoption, Google basically makes the Pixel as a reference design to keep the Android phone ecosystem somewhat focused, its Google telling Samsung and co "an Android phone should look something like this". RISC-V will then be quickly adopted by other phone manufacturers as they know ARM support for Android is now second priority to RISC-V and Google will put a lot of effort into making the transition as smooth as possible for end users.
TL;DR RISC-V has become a weapon in big game of corporate warfare
Raptors CPUs are already binned. There are only 12 and 24 core CPUs.
the mobo was designed as a workstation mobo, plenty of high-end commercial customers want that.
If they released a single-processer version, that would have helped a ton, but apparently power9 processers have somewhat limited PCIE lanes, and the board lacks the ability to reroute them between processors, so the primary processer handles all the peripherals on the board and the top two sockets, while the 2nd processer deals with the other 3. They are not fake x16 sockets, all three have full pcie4 throughput(given you have 2 processers and cards all using pcie4)
>manual pg 10
>The T2P9D01 has been specifically designed for applications requiring high security, high I/O
capability, and large amounts of processing capability. PCIe slots and peripherals are directly
connected to each CPU, therefore two CPUs must be installed to use all PCIe slots
Also I'm pretty sure the 'additional options' are/were the 18 and 22 core processors. iirc they weren't offering them a couple months ago when i checked last.
I want to know if i can mix cpu models on this board, as in using a 22 core in the main socket and an 8 core in the second to 'unlock' the other three pcie sockets and the oculink adapter(basically a pcie x4 header for cables).
Never. the cpus simply aren't designed for that.
But i could definitely see a higher-volume single-socket ATX board targeting companies and programmers that need the security/open source firmware but don't need all the performance. Hopefully for ~400$, though i doubt they will make/sell enough volume for that.
Does anyone here have the Asus D16 mobo? How's the libreboot support for that board all these years later?
If enough autists are still keeping it real, i might actually get a talos II. It'll probably get similar or much more support for years to come, and the manufacturers might even help a little.
>There are only 12 and 24 core CPUs.
Really there are only 24 core CPUs since the differences between the SMT-8 and SMT-4 models is basically just post-fab configuration. It just occurred to me how aggressively binned the SMT-8 CPUs must be since an entire super-slide has to be functional to be considered a working SMT-8 single core whereas only half of one has to be functional to be considered a working SMT-4 core.
>power9 processers have somewhat limited PCIE lanes
The Sforza modules actually have 48 lanes (the most amongst the current Power9 CPUs) but only 40 and 44 are broken out on the Talos system, however this still puts it on par with current Xeon workstation CPUs which its targeted as a competitor to. The EPYC CPUs have more but they lack dedicated Infinity Fabric pins which means in multi socket configs half the PCIe lanes are used for CPU-CPU communication so you don't gain any more available lanes.
>I want to know if i can mix cpu models on this board, as in using a 22 core in the main socket and an 8 core in the second to 'unlock' the other three pcie sockets and the oculink adapter(basically a pcie x4 header for cables).
Given how multi socket systems work, that you can do it with Intel Xeon CPUs (provided certain things are the same between the two CPUs), and what the various boot stages on the Talos system do, there shouldn't be any reason why it wouldn't work. If you email RCS they should provide you with an answer quickly.
supercomputers dont have anything to do with the kind of productivity hes talking about
>Never. the cpus simply aren't designed for that.
LGA2011 Xeons weren't either. Yet this thing exists:
>If you email RCS they should provide you with an answer quickly.
pls share it if you get it
>apparently power9 processers have somewhat limited PCIE lanes
>48 lanes per socket
nigger, what? That's double of what a typical desktop CPU has!
Since the two cores share L3 cache and memory interface, they need to be enabled as a unit even on the normal SMT4 variant.
>as in using a 22 core in the main socket and an 8 core in the second to 'unlock' the other three pcie sockets
Two 18-corers will cost you less tbh. The 22-core SKU is ridiculously overpriced right now. Yield issues, probably.
Yeah i didnt actually bother to count them. the issue is actually that there's no way to switch which slot any lane is dedicated to.
It still won't happen with power9. Maybe with a later iteration, 10 o 11. But i think we will see another e-atx or atx power9 mobo.
Also, they're offering other sas controllers than the proprietary microsemi one. Are these new ones soldered on like the microsemi?
>It still won't happen with power9
I know, I know. It's still a funny idea :)
>Most people who brought a Talos system did so because a P9 server costs an order of magnitude more... security autists are only a minority of people who brought them. Of the two people in this thread who actually purchased a Talos system one did it because of the first reason and the other didn't say.
I'm the security autist.
What config did you order?
Rather than making a single socket board for a tiny minority wouldn't it make more sense to depopulate the already existing board further?
Red is what would be removed on a single socket board, blue is the proprietary SAS controller and ports that are already removed and I'm not sure what is in the green circle as it's not listed in their manual.
With a cheaper single socket board you can start off with 4 cores and 8 gigs of ram to get a decent system for maybe around $2,000 now and work your way up to 22 cores and 1 terabyte once prices go down in a decade. Having an EATX board that's more than half empty might seem odd but I'd get one if it's a few hundred dollar cheaper.
Also their μPCIe connector seems odd:
>μPCIe connector (J10108) is incorrectly identified as OCuLink (SFF-8621) due to errors that were discovered post-production in IBM OpenPOWER reference designs. While the physical connector is the same as OCuLink ports, the wiring is different, notably in the presence detect / clocking pins.
>Assuch, while an OCuLink peripheral is unlikely to be damaged by being plugged in to the μPCIe port, it will not function while attached.
So what can it be used for exactly?
Absolute madmen... And to think I was the only one who cared enough to get one back in august.
Because of massive, gobstoppingly fucktarded incompetence at the time:
>Along with other Mac clones, Apple killed Starmax clones made by one of their two main CPU vendors, Motorola. This greatly angered Motorola, who responded by internally deemphasizing PPC efforts.
>Vector coprocessor, AltiVec/VE/VMX, outperformed any other ISA's, but only because it was massively overengineered to the point it's about 1/3rd the transistors of the entire rest of the CPU. Mandatory on-die instead of an optional off-die upgrade like FPUs used to be, crippling yields.
>G4/G5 was the result of going with the gimped laptop (603-derived) branch for all G3s, killing the "Habanero" desktop (604-derived) branch. Thus when such features were added back to the G4, it resulted in a jack-of-all-trades-master-of-none chip, which by the time of the G5 had been left without the laptop friendly G3 arch.
>Apple's motherboards were rarely updated, and their custom chipsets were updated even rarer. This caused Apple to miss out on AGP, PCI-X, PCIe, DDR, USB 2, SATA, etc., for multiple generations, in addition to ignoring half or more of the FSB throughput in Motorola's latest CPUs.
>Early in the G5's run, a startup named P.A. Semi designed a G5-compatible PPC that could meet or beat contemporary x86 in laptop use. Apple deceptively strung them along, ultimately turned them down at the last second before the x86 switch, and bought them out afterward.
How can one be certain these are indeed 100% botnet-free? What if it's just a honeypot to catch those who are willing to shell out much money for supposedly botnet-free computers? Anyone decapped and analyzed all the chips the system has?
If you're that paranoid your only real option is to build your own computer from discreet logic or buy a 70s era minicomputer like a PDP-11. Even then, you'd need to put it in a faraday cage because they generate enough RF to be intercepted from a few blocks away. (still applies to modern equipment though tbh fam)
I thought this machine was a lot more money than it is. Did it go down in price or something? Might end up getting one in a few months if the price is this low. I think it's worth it.
The full workstation is kinda expensive because it is built of heavy-duty workstation/server parts (rackable case with dual redundant PSUs, FirePro GPU and so on...), but if you buy just the mobo+CPU(s) and put them in a normal PC case with a normal gayming Radeon, it's not *that* expensive (for what it is).
You're probably thinking of the first Talos computer from 2016 that used POWER8 and cost $3,700 just for the motherboard. The cheapest POWER8 CPU option was $1,135 and their cheapest prebuilt was $7,100.
You'd also need to get a heatsink for $55. The most basic build if you've already got a PSU, RAM, case and storage would be 55+375+2325 = $2755. It's still way too much for someone that just wants an x86 alternative desktop for use as a daily driver.
>55+375+2325 = $2755.
There is a mobo+CPU+heatsink bundle that comes at a discount. $2475 for the config you described:
I agree it's expensive, but it's not bad for what it is. Also, since the CPUs are very competitively priced, the more powerfully you spec your system the more bang-per-buck you get.
You may be correct about that.
>there are at least 2 or 3 other people on /tech/ who brought a Talos system
Feels good man, I only ordered mine back in March this year though.
>I'm the security autist.
I would be lying if the security aspect had no influence on my purchase, while I brought the system to do Power9 dev work on the fact that its entirely owner controlled is definitely a bonus in my eyes.
>Mobo + X2 4-core CPU bundle.
Fair enough, are you going to do anything particular with it or is it just for desktop use?
That's actually not as much of a cost saving as you would think, the majority of the extra cost between a single and dual socket mobo is actually in the PCB needing to be twice the size rather than there needing to be an extra socket and set of DIMM slots.
>You're probably thinking of the first Talos computer from 2016 that used POWER8 and cost $3,700 just for the motherboard.
One of the reasons why the mobo in the Talos I was so expensive is because the Power8 CPUs didn't come with memory controllers on the chip (they used discrete memory controllers, the 'scale out' Power9 CPUs designed for servers require these as well), requiring two Centaur chips on the board which are the under the heatsinks between the DIMM slots in your pic.
>You'd also need to get a heatsink for $55.
There isn't actually a Mobo+CPU(s) combo which doesn't come with the heatsinks, the only way is to order the parts individually.
>Also, since the CPUs are very competitively priced, the more powerfully you spec your system the more bang-per-buck you get.
The x86 systems Phoronix benchmarked the Talos II against had CPUs which cost almost as much as the entire Talos II system, the ARM server they benchmarked it against cost quite a bit more as well.
If I can't get reliable wifi signals from the next room, what makes you think that CPU radio signals will be easy to interpret from the next room?
>Fair enough, are you going to do anything particular with it or is it just for desktop use?
Desktop and server use, I was gifted a Vega 56 so I might try some x86 games with QEMU + GPU passthrough. I might also run it as a multiseat system so I can share it with the rest of my family.
Only the 4-core option was available at the time, so I might upgrade later when it comes down in price. I'm aiming to get 15+ years out of this thing after all.
Because I've worked with TEMPEST systems and have seen demos of various attacks on non-TEMPEST hardware. Besides, older CPUs running at a few MHz are in the 60 and 80-meter ham bands, so you can tune into this sort of stuff and demodulate it with the right equipment.
Here's a PDP-8/e playing music on an AM radio with nothing but it's own RF emissions: https://hooktube.com/watch?v=akvSE5Z474c
Even if you can't figure out what the CPU is doing, you can intercept the terminal much more easily.
Even if the chips are a bit hacked together and the fact that most corporate system admins will shy away from switching to it, in the end its still a reliable platform which has no backdoors (atleast intentional ones). This feature alone is enough to convince countries like Iran, Russia, Turkey, NK, Taiwan, N. Cyprus, and Pakistan to buy them and migrate their system to it. While it is true that most companies and government's run their system critical stuff on ancient hardware because of reliability and shared knowledge, they also care a lot about corporate espionage and foreign spying. Im pretty sure a country like Iran would migrate their power grid system to this platform to avoid another sabotage, even if that means half the country will be out of power for a month. While this example is a bit extreme since Iran relies on unofficial channels to buy their hardware and countries like Russia and China actually has their own custom chips, there are still a lot of people who would prefer to run their software on open hardware.
The moment the likes of Iran start buying this shit is the moment cianiggers start bugging and backdooring it to hell and back at the factory.
Did you mention ordering one before, or are we now up to 6 Talos owners in this thread?
Well its still better than having official backdoors. CIA couldn't intercept the Siemens machines they acquired but they still came close to making Iran a nuclear wasteland because one engineer took their work laptop (windows, intel) to his home and plugged a compromised usb drive lol.
Just how slow would x86 emulation be?
>CIA couldn't intercept the Siemens machines they acquired
Because Siemens machines in question aren't produced in the US. This thing is.
I believe there's a video somewhere of Raptor guys playing Unreal Tournament semi-fluidly under Qemu on the previous gen POWER8 Talos prototype. It can only get better from that.
tl;dr: fast enough for normal desktop apps and 10+ year old PC gaymes.
>Did you mention ordering one before
Yes, I am >>907254
>I believe there's a video somewhere of Raptor guys playing Unreal Tournament semi-fluidly under Qemu on the previous gen POWER8 Talos prototype.
That wasn't under QEMU, the Raptor Engineering guys got Unreal Engine 4 running on the original Talos system
Do you not understand the hypocrisy of wanting to buy a Talos computer for the sake of security only to install Windows on it?
but its in a VM, silly! the main OS that u use for everything else would be running a Free as in Freedom GNU/Linux installation ^.^
That's a different video. There was also one where they compared performance of natively running UT with emulated x86 build of UT.
So games from nowadays would suffer quite a bit, right?
A VM isn't going to increase your security. If it did, the only tool needed for security is a VM and everybody would run all their software in a VM.
Is POWER9 vulnerable to Spectre you gasworthy namefag?
It *does* increase security, though it's obviously not a silver bullet. And yes, it is common these days to run every web service in a separate VM for security reasons. I believe it's a matter of time when this trend reaches desktops too.
That was a rhetorical question, dunderhead.
>And yes, it is common these days to run every web service in a separate VM for security reasons
So you're implying rhetorical questions are not questions?
>I believe it's a matter of time when this trend reaches desktops too.
Install Qubes OS bruh
>I want to know if i can mix cpu models on this board, as in using a 22 core in the main socket and an 8 core in the second
Note that there is no 20-core POWER9, but there are 18-core and 22-core ones.
How is IBM selling these things so cheap? I get that the lower end options are just binned parts with defective cores, but even the normal chips' prices are incredible. If Raptor gets enough volume to make mobos competitively priced with TYAN/SuperMicro, this thing might make some serious inroads.
Only reason I haven't gone KVM w/ PCIe passthrough yet is sheer laziness. Even if I never ran Windows and didn't care about security, it would mean never having to reboot, and much easier diagnostics for segfaults.
>How is IBM selling these things so cheap?
Because their business model is closer to Intel/AMD than their mainframe division. They can simply sell more to a few OEMs then they'd ever sell if they were using them in their own servers or mainframes.
It's not like IBM was gonna use their binned 4 and 8 core CPUs.
You'd be surprised. Some enterprise workloads are memory bandwidth limited. It then makes sense to spread the few required cores across as many sockets as possible to maximise bandwidth.
Even Intel produces processors like the $7000 quad-core Xeon Platinum 8156 for such scenarios.
lol looks like it'll run q3 engine games and unreal 2004 at 15fps.
clearly this is not an option for your proprietary gaymes.
some of the open source shooters that are out now, you might be able to compile for power and be good.
Have you read what it said under the video? Stuttering is caused by Qemu translating vector SSE instructions into arrays of scalar ops instead of VSX equivalents. This shit needs to be implemented in Qemu for a massive speedup in gaymes as it's mostly gaymes that use vector ops extensively - and with TALOS out there somebody might be finally motivated to do that.
This, and that video was made on the much slower POWER8. POWER9 is almost twice as fast in lowly threaded workloads.
>How is IBM selling these things so cheap?
IBM made some clever design decisions which resulted in even defective chips being able to be sold, AMD chose to use multiple dies for their Threadripper and Epyc CPUs (Intel did the same with the Xeon Scalable series) which increases the yield for the higher end chips but raises the cost of the lower end ones since the CPUs need to be made from four dies with matching properties (number of functional cores, maximum clock speed, etc). Also there is a lot of license fuckery with x86.
>Also there is a lot of license fuckery with x86.
External interfaces should be unlicensable, unpatentable, uncopyrightable and otherwise unjewable
Seriously, who the fuck thought it a good idea to forbid compatible competing implementations and enforce by law a de facto monopoly?
Have you considered what would have happened if every chip vendor could make x86 chips? its not good.
I would love it if every other chip vendor made x86 chips. Then they would be cheaper and we would be able to buy versions without botnet.
Sort the list by year of introduction, if every company could make x86 processors freely then every architecture after x86 started to gain dominance would have likely not been developed. Linux as we know it today likely wouldn't exist since one of the main reasons why it took off in the early days was that it could be run on everything while Windows only ran on x86.
x86 is so awful that Intel themselves tried to kill it with the Itanium ISA in the early 2000's, and if every chip vendor could make x86 then x86 would be the only ISA made, because why the fuck would you make anything else?
It's a shame they forgot about PPC when they ported Quake3 to MIPS, SPARC, ARM, Alpha, and AMD64.
The first thing I'm going to do when I get my Talos II is emulate Quake 3.
>wants to emulate Quake 3
>literally the post above his is a link to the source code for Quake 3 Arena, which would allow him to compile it for PPC64le
QIIIA literally had a official native release for PPC processors along with most other games that used its engine.
Old commercial wintel emulators for PPC Macs like Virtual PC & RealPC ran at about 1/3rd native speed for pure x86 assembler (in other words, about like a , though sometimes you can reduce or nearly eliminate emulation penalty using cached dynarec/JIT and various HLE techniques (like boilerplate WinAPI or Linux libs that are precompiled to native), depending on how repetitive or standardized the code you're running is, and whether or not your emulator (NOT Bochs or QEMU!) offers such features. For gayman, this means heavily handcoded CPU-bound titles will play slowest, while standard engine/library-based GPU/IO-bound titles will play fastest.
The only reason nobody makes x86 chips anymore is simply a lack of interest. As late as the Pentium III era, there were numerous others like Centaur, Cyrix, and NexGen. Honestly, it would be easier than ever to do, since the recipe for modern x86 CPUs is literally "RISC core+x86 decoder".
Yeah, the retail PPC port lacked a lot of PPC optimisations contemporary ports outsourced to dedicated Mac houses (Westlake, Macsoft, Feral, etc.) had. That changed after id handed the Mac port of Quake 3 to The Omni Group for later official patches, and then after the engine was open sourced, the ioquake3 fork soon became one of the best optimized pieces of PPC software ever written.
massive kvetching from the current monopolists, for sure. I can't see this as anything but a good thing.
>this means heavily handcoded CPU-bound titles will play slowest
This really depends on the quality of the dynarec. Theoretically it can get as close to bare metal performance as its cleverness permits, while there is an upper bound on what you can do to accelerate context switches and other hard to emulate OS-level hardware features. It might be that the titles bound by lots of small IO or GPU requests get bogged down the most not to mention syscall-heavy DRM schemes.
>The only reason nobody makes x86 chips anymore is simply a lack of interest
No, it's the army of lawyers Intel will send up your ass if you try to implement any ISA features newer than the original Pentium. VIA tried, and got sued to oblivion.
>As late as the Pentium III era, there were numerous others like Centaur, Cyrix, and NexGen.
NexGen and Cyrix were out of the game before Pentium II came out. Centaur produced ridiculously slow shitchips with a 486-level feature set, which admittedly immunised them from jewry. Not a viable route today though tbh.
>Honestly, it would be easier than ever to do, since the recipe for modern x86 CPUs is literally "RISC core+x86 decoder".
...+a tower defence against a jewyer zerg rush. Even nVidia backed down from that and went ARM with their APUs.
>This really depends on the quality of the dynarec.
I'm speaking from personal experience with SoftWindows (which supported HLE for GLIDE) and Rosetta (which did HLE for OpenGL), in addition to large amounts of (DOS/Windoze & OSuX, respectively) WINE-style precompiled native OS libraries.
>VIA tried, and got sued to oblivion.
Wait, what? I thought VIA has a patent-sharing agreement with Intel, and is still making new Nano architecture x86 chips.
>Wait, what? I thought VIA has a patent-sharing agreement with Intel, and is still making new Nano architecture x86 chips.
For many years already they haven't made a new one. The court battle ended up settled out of court under undisclosed terms, but judging from VIA's actions afterwards, they most likely are allowed to sell and market their x86 chips only in SE Asia.
Judging from the demo being played on a 500 mhz G3, the port is decent enough. The only thing really holding it back seems to be the Rage 128 Pro and the implementation of OpenGL on MacOS 9.
Anyone got theirs yet? Mine's arriving today.
Lucky fucker, I won't be able to get mine until early July due to contract work.
Better post pics and show what OS you put on it when you get it up and running.
Who cares. Anons want a
workricestation at their homes, not a server.
At least they dropped CDIMM
You'd hope they'd keep it free, although I'd bet it would vary on vendor.
It would be interesting to see if they make one for under $7,000 and not require some ridiculous minimum order.
>implying I won't be migrating my employer's servers to POWER9 during our next upgrade cycle if they're cheap enough
I've already convinced him that modern x86_64 is pozzed beyond repair, which is why we're still running old Opterons.
Not all of us are NEETs, anon.
Now that is how a laptop should look!
Show me a NEET who can afford a Talos ricestation.
>Before the mid 80s everything when software was shared there was no license, blueprints were available
This is not true. There have been hardware and software licensing programs since the 60's.
How bad is it that I initially thought you meant the heat output would be enough to cook rice with, not rice as in SPINNAN CUBEZ.
>cooking dinner while compiling at the same time
Can that boring SuperMicro case really be considered rice?
You can buy just the innards and put them in your own case.
Good luck trying to find an eatx case
That won't be a challenge for a s3ri@z ricer.
What is the guarantee the aren't intercepted en route and tampered with by glowindarks (like was the case with Cisco gear)?
None, but you can grab the firmware, look at it, compile it, and flash the motherboard again.
WHY DID I BLOW MY LOAD ON SPARC SERVERS AND A SHITBOX CAR????? I'M GOING TO KMS NOW
WHY AM I SUCH A LOW TIME PREFERENCE NIGGER?????????????? REEEEEEEEEEEEEE
What are you on about? Most riceboxes can do eatx.
Are you the livestreamer that bought 3 T3s?
Serious question: why buy this and not an openSPARC PC? how is this less botnet?
Mohamed Mahmoud Mohamed Farag, the Talos II actually exists.
IBM is still in the CPU business for the foreseeable future, Oracle isn't. Without somebody big pushing the ISA forward, it will stagnate as a technologically leading-edge product.
The largest developer and manufacturer of SPARC processors these days is not Oracle but Fujitsu. You don't hear about them much because they focus on *very* big iron and supercomputers, but they're there and plan to stay there for foreseeable future. SPARC's gonna move forward, even over Oracle's dead body high time for that.
Has Fujitsu actually made a commitment to push the ISA forward with something along the lines of a future M9, or is it just cranking out new silicon with minor bugfixes and extensions like Renesas did with SuperH?
Fujitsu will join the ranks of Broadcom, Qualcomm, Calxeda, and AMD to piss away money on that anemic Brit architecture known as ARM. If the Chinese or Russians don't do anything with SPARC and MIPS, they are dead in the water.
>how is this less botnet?
Due to open RTL openSPARC is less botnet, but the motherboard for SPARC might be more botnet.
>why buy this?
I am going to assume you mean POWER vs SPARC
14nm vs 65nm
10MB L3 cache/core vs no L3 cache
48 PCIe 4.0 lanes vs 8 PCIe 1.0 lanes
DDR4 DIMM vs DDR2 FB-DIMM
I think raptor engineering have some slogan "most free you can get this decade", I think that works.
Given how they aim this shit at HPC, I guess they won't let the ISA rot, however it's hard to tell given how secretive Fujitsu is regarding their silicon. Any commitments will be known to their business partners and nobody else. It's not like they publish their docs and roadmaps on their website, unfortunately.
One doesn't exclude the other, anon. Even Intel used to make ARM chips not so long ago.
And I do not expect ARM to become capable of replacing SPARC or POWER or x86 in supercomputers for the next decade at the very least.
Alright so sparc is out, what about riscV? how far are we from getting usable silicon?
You can get something worse than an ARM for the same price of the Talos II bundle right now!
Intellectual "property" shouldn't exist.
high* time preference.
If you invent a gun better than maxim or gattling ever did, or even kalishnakov, wouldn't you want some kind of patent protection to forbid people from manufacturing it?
Wouldn't you want to give the idea to the working people for in defence of their soviet?
No. I would rather allow people to work, trade for, or manufacture their own guns of their own design to protect their family and neighbors from invading armies and enemies. However, I believe that people should own the rights to their means of production including ideas.
>Intellectual "property" shouldn't exist.
Indeed. However, even within this fucked up system, copyrighting interfaces is a whole next level of fucked up, negating the purported benefits of those laws.
I would want to release a lot of Zyklon B over that soviet. For free.
So, what's the poorfag option for getting a PPC processor? Do I still have to buy old Macs or """Amigas"""?
There's new benchmarks posted the phrononix article isn't out yet but the results are pretty disappointing. It's also using an older GCC rather than 8 which has POWER optimizations to port x86 vector intrinsics switched on.
This sort of 'out of the box' performance benchmarking is pretty worthless as most of the time developers just make sure it works on PPC by doing tests in QEMU, it needs a proper port to take advantage of the hardware. If you're going to spend 5-7 grand on a computer that isn't x86 you should also be porting software over to it properly.
There aren't many options but those newer Amigas make the Talos 2 look like a bargain.
For macs the G4 powerbooks are pretty decent if you max out the ram and put in an IDE SSD. They're 15 years old now so they need new thermal paste and a refurbished battery. I'd recommend the G4 12 inch powerbook since they're cheap if not in pristine condition, it uses the same display as the thinkpad X61 so you can put in a 1400x1050 panel.
The G5 desktop isn't too expensive but it's very hot and power hungry for what it is. The other options are the PS3 and the Wii but support is limited and development has stalled.
I guess I'll have figure out how to make that iMac with a dodgy CD burner boot off USB 1.1 through firmware commands until I can either fix it or get the money for a more capable PPC system. Though I think a relative may have a iBook G4 laying around.
I love the z80 tbh
I have a dual G5 PowerMac with MorphOS and OSX 10.5 installed. with 8GB it's very useable. Editing HD video, a few older games, some coding.
I miss renoise, qt creator, and many other applications though. It's hard to escape.
A Raspberry Pi 3 is most assuredly faster than first generation PowerMac G5s. Unless you want a historic monolith heater, there is no poorfag PPC/POWER option.
I am going to call your bullshit.
A G3? sure. A G5? Fuck no.
Yeah, no. The PPC970 series CPUs are nearly twice as fast per MHz than Cortex-A53 at integer code, and much, much more than that in FP. Not to mention you can fit a usable amount of RAM in there, unlike a SJWPi.
>faster than powermac g5
just kys fam
Unfortunately, I could only find dual core 2GHz G5 on benchmarks. FFT is really nice. Everything else, not so much.
Core per core, you're right. The first gen G5's were 1x1.6GHz 1x1.8GHz 2x1.8GHz. A 4xCortex-A53 vs 1xPPC970 would be close. You guys are also forgetting Altivec code is being striped out rather than fixing bugs because of lack of interest. Meanwhile Neon optimizations are everywhere.
NEON units in A53s are pretty anemic. Even running purely scalar code, a 970 will easily outperform them.
Watt for Watt, absolutely not even close
>shifting the goalposts
The rpi 2 we have in our robotics (and embedded crap) club in uni croaks with xfce. And apparently the difference between that and rpi 3 isnt too great.
I will be shitposting from a PPC wii with debian on it very soon. preperation is almost complete and a webm shall come for those who are patient.
What can a multi processor g5 do that a Beowulf cluster made out of ARM SBCs can't?
Run single-threaded software at an acceptable speed.
Yeah no, I don't think you realize how piss weak those are by modern standards
Still thrice as fast as an RPi.
Good enough for typical desktop use.
So is any random ARM SBC
Someone photoshop a giant cock that touches the ground and keeps going on that pic.
I'm eager for the day there's a freedom respecting architecture with a whole ecosystem like x86.
i.e you buy a case, a motherboard that fits into it, a standardized heatsink and fan standard (can just reuse intel's ATX I guess) and so on.
RPi isn't. Most of its clones ain't either.
So for you guys whose primary motivation is escaping the x86 botnet, why not just get a MIPS laptop? I mean,for the vast majority of you, all this is going to be is a dedicated shitposting machine, and you really don't need the full power of Talos just to post anime traps on /tech/. Why not get a cheaper x86 alternative?
Nigger, I have over two and a half thousand tabs open right now in my browser. The browser process (palemoon) takes ~13GB of memory, not counting shared pages. It's hard to find an x86 laptop that would not bog down under that, and you're telling me to get some MIPS shitbox?
What the fuck is wrong with you? Close your fucking tabs.
You can screencap this thread too.
I gobble cocks for breakfast and dinner. And I'm getting hungry.
>removed the tripfag
Why the fuck should I close my precious tabs if I can instead buy myself a freedom-respecting shitposting machine that can easily handle them all for the measly price of 1$ per tab? I expect it to shave a few minutes off my browser's startup time, too.
I don't have my G5 anymore, but I do have my 1.5GHz G4 Powerbook. I'm compiling Gentoo on this bitch and will pit it against a Raspberry Pi 3. Unlike the SPARC guy, I'm not a complete moron. I will be back in exactly 3 days with my findings.
Well they listened to you, are you happy now?
memory leaks and instability are going to own you, my friend. Also a quick reminder that palemoon once fucked up my hard drive, causing a shit ton of unreadable sectors.GREAT software.
That's not something a piece of software can do, retard. You got a defective drive. Also, Pale Moon is pretty stable once it finishes loading all the tabs. I only need to restart it once every two weeks or so.
>(you) are about to close 1174 tabs. Are you sure you want to continue?
The problem is that there still isn't a really good exension for unified tab/session/bookmark management. There's treestyletab, there's session manager, and then there's plain old bookmarks, but every concept is separate and clunky on its own. What is needed is an extension that blurs the line between active tab, loaded-but-inactive tab, and "dormant" (like bookmarks or session manager contents) tab, being able to shuffle them around and change status freely and group them etc.
Is 2018 the year of the POWER desktop?
>That's not something a piece of software can do, retard.
Of course it can, I deal with HDD destroying software everyday.
web browsers do massive amounts of r/w to HDDs, if there's a bug that ups that it could absolutely destroy sectors.
Terrible pics if you have anything better than dial-up. Even so, looks like a nice clean setup. Not planning on jumping in right away but this has my attention.
Weapons should be used on soviets, not given to them.
>Weapons should be used on soviets, not given to them.
This. Nukes especially.
>blew 7k on it and the audio and amdgpu doesn't even work
this is the linux equivalent of hardware lmao
Raptor are going to be releasing a cost-reduced single-socket version of the Talos II, with prices starting at $1,399.99 for a desktop system.
It's based on the same board as the Talos II, but it all the components for the 2nd socket and SAS controller are unpopulated.
A motherboard-only option is planned.
SAS will still be an option
>with prices starting at $1,399.99 for a desktop system.
>CPU and memory sold separately
>complete desktop system
This is what you get for being an early adopter. Audio output comes from HDMI on the GPU, does it not? I would imagine if this is the card they're shilling, it will get fixed soon.
>audio doesn't even work
Well duh. There is none, so it doesn't work. You're supposed to buy your own soundcard or use HDMI audio if you need it.
Did I say complete?
I called it earlier, those sockets probably cost $2-300 each. This model is ideal for those that want a liberated computer with modern specs but don't need a monster PC.
This is what an unpopulated board ought to look like >>909662 so there's only 2 PCIE slots available. That onboard proprietary SAS connector in the corner takes up one lane being connected to CPU1 so that's kind of a waste for a libre computer. If I had one of these I'd only be using NVMe and video card in those slots anyways so two is enough.
As I predicted for under 2 grand you'll be able to get a system with a decent 4 core CPU, a few gigs of ram and storage plus a video card. Upgrade that to 18 or 22 cores and half a terabyte of ram or more while taking advantage of the newer PCIE4 cards whenever it's cheap enough to do so.
"desktop system" kind of implies that it's a complete, working, you know, system.
>If I had one of these I'd only be using NVMe and video card in those slots anyways so two is enough.
What about a soundcard? There is none on the mobo.
Most new video cards have audio on board. You could get some HDMI/Displayport adapter that breaks out a 3.5mm audio jack.
maybe an usb dac would work
There's always HDMI sound. If you really need the extra slots there's OCuLink.
I'd be using a half decent USB DAC, even this $2 thing plugged into a USB hub or wired into the case would be good enough for most people and would work fine. I've been into computers almost my entire life and can count on one hand the times I needed more than 2 PCI slots in my computer.
The PCI lane for that is connected to CPU2 and it isn't a real OCuLink port. Still don't know if the issue of incorrect wiring can be dealt with by using an adapter or when/if they'll fix it.
Looks like the cost of freedom is $571.30 or $352.02 without the newegg special.
They have OCuLink breakout adapters on their site now.
by 'unpopulated' do they mean they didn't solder on ram/cpu sockets and associated components?
I'm not sure my OCD would be able to cope with that. empty, unsoldered space doing nothing in on my still very expensive mainboard... I'm either getting the talos II full or the next thing they release, a dedicated single cpu board. (hopefully in m-atx), not truly enough hardware to justify a full atx board unless they find a suitable libre sound card and dedicated more cpu lanes to pci slots.
There's no way those shitty magnesium sockets cost more than 50$ to produce. Unless IBM kiked them that hard on manufacturing licenses. There's no way the socket magically costs 100x as much as an intel socket just because of the beefy lever and relatively low volume. The cpus and sockets are the same as on the commercial servers IBM normally makes, right?
>imblying /tech/ doesn't install gentoo all day, every day.
So, any yurofags getting one? If so, how long did it take to ship and what kind of shit you had to go through with the customs? Did it come with any suspicious glowing in the dark chips soldered in that weren't there in the schematics?
I was on holiday in the US when I got mine, took it home on the plane expecting to get slapped with customs charges but they didn't care. The whole holiday probably cost me less than VAT should have done.
>Did it come with any suspicious glowing in the dark chips soldered in that weren't there in the schematics?
>weren't in the schematics
power customers to not rate to view those schematics.
Checked or carry on? I'd be this little dog seeing that luggage get knocked around on the carousel.
You don't want to know what the couriers do with parcels delivered to you. It's packaged to withstand all of that, you dolt.
<spoiler>Also this >>915915</spoiler>
>Unless IBM kiked them that hard on manufacturing licenses
I'm guessing what they'd cost in the Talos but they do mention on their wiki that the way the CPU is held in is licensed
>The Talos II uses IBM's proprietary high-pressure CPU retention mechanism. This design is available for OpenPOWER members to use, but remains licensed.
So much for my plans of making a water block for this CPU and selling off a few extra.
>The cpus and sockets are the same as on the commercial servers IBM normally makes, right?
Of course not the sockets used in most of the hardware IBM sells have 3899 pins. The socket on the Talos II has 2601 pins and as far as I can tell the only other thing that uses it are the Power LC922 and LC921. Their 4 socket Zeppelin system that will come out soon should use them as well.
Looking at their roadmap for POWER9 there will be newer chips in 2019 with a 67% increase in memory bandwidth. Is there a chance that the Talos II board would support them? Were POWERn+ CPUs in the past backwards compatible or did they require newer boards?
>making a water block for this CPU and selling off a few extra
Don't worry. That'll be perfectly fine. As long as "a few" is a two figure number and buyers have to contact you personally to get them. Consider partnering with a license holder, if you want to make a business out of it.
Maybe we could ask Raptor to partner with Koolance or whoever to make an official P9 mounting plate for their flagship CPU block? I see there would a small market for those (count me in) and I doubt it would cost much to develop.
>Sforza socket waterblocks
IBM already make them
Do they sell them in small quantities to small OEMs like Raptor? If so, this should be added to their store yesterday.
I don't know, the only place I have seen them used is on IBMs Power9 GPU server and even then it was only as an option the system is designed to have water fed in at very high pressures, the water lines are rated for like 200psi if I remember correctly.
Welp, maybe adapting a COTS block is a better option for a desktop system.
>Features In Progress
>HTML5 Java Script Web User Interface
You could get away with a lot less pressure for a Talos system, the IBM server needs high pressure to reach the required flow rate since there is a CPU and 3 P100/V100 GPUs in series for each feed line to cool.
The question is whether the block will allow appropriate flow with normal coolant pressure and whether the fittings are compatible with normal tubes.
>starting at $1,399.99
Loonixfags have become macfags, its official.
>pay thousands of dollars for a (((x86))) lappy that overheats and throttles all the time, has a bloated botnet OS, drops support for opengl in favor of proprietary "metal", and has a shit keeb with an emoji bar
>pay thousands of dollars for a based RISC POWER9 desktop workstation with a /minimal/ libre OS, opengl support, and will probably be paired with a comfy mechanical keeb.
yeah sure, same thing.
And yet MacOS still works much better than any hobbyist OS such as looniks
Sure thing my dear homosexual cuck.
Enjoy your botnet.
Macfags have POWER cpus too.
macfags had (in the past) PowerPC arch you retarded nigger.
POWER != PPC
Wow you're late. Macs have had intel CPUs for probably a decade now
Wrong, they have MOTOROLA CPUs
Enjoy your soldered RAM and shitty G4 ibook
>works much better
a completely empty statement that tells the reader nothing
He probably meant the GUI
somebody gas the tripfag already
Heys guys, ditch this place and come over to /radb/, it's like /b/, but rad
Why has nobody posted pictures of their Talos II systems?
At least 6 people in this thread ordered them.
I want /tech/ review before I decide to take the plunge.
It was not to my liking.
Everyone was larping. I want to take the plunge, but when you're reduced to using a single board computer because of electricity bills you have to use your imagination.
I posted a picture of my motherboard in the PPC thread when my order arrived, this thread was kinda dead then so I didn't bother posting it here.
>but when you're reduced to using a single board computer because of electricity bills you have to use your imagination.
Do you mind if I ask what life choices led you to this?
>At least 6 people in this thread ordered them.
SPOILER: it's a single shill and you're retarded.
Cool. What do you use it for? I assume you do more than shitpost with such a beefy workstation?
I have only had it running a few days and in a sense I am still setting it up so I am still shitposting from my small Ryzen system.
I brought it for a combination of reasons
>its a power9 system and I am planning on doing some development work which will need to support PPC64 so having native hardware is nice
>its pretty power efficient for the compute so mining crypto like Monero is actually profitable
>its botnet free
The power efficiency is surprisingly good, I was running some basic stress tests on it and despite all 64 threads being at 100% for the better part of an hour the temperature of CPU1 (the back CPU) sat at about ~65 degrees while CPU2 (the front CPU) was at ~45 degrees with both fans almost idling.
Whats really interesting about the P9 CPUs is that the lower end core count models still have a fuckton of L3 cache. The way the CPUs are binned means that for anything 12 cores and below each core has its own 10MB of L3 cache, its only at core counts above 12 that cores start having to share L3 caches. So the 8 core models in my system have 80MB of L3 cache each.
Grossly overpriced motherboard
>its botnet free
Comes with an embedded uboot/linux/systemd/webserver with a full network access.
>Grossly overpriced motherboard
You clearly don't understand economies of scale, you not buying one means you are part of the reason why they are expensive. The low production quantities of the unique components such as the CPU sockets as well as the low production quantities of the motherboard itself means that the cost is going to be higher. A lot of the cost is in the development and production setup, if only a small number of units are going to be produced then those costs end up being significant per unit. My order was in the low 2000's and placed it comparatively late, and given that many people would have only ordered a single system I suspect less than 2500 of the two socket motherboards have been made. The single socket version is less than half the price because the development and production setup costs have already been largely covered by the two socket version.
For an equivalent quality Intel motherboard you are still in the range of $500. When companies like Supermicro design a motherboard they are expecting to sell 250,000+ units, 100x more in terms of production numbers is the difference between a $2500 motherboard and a $500 motherboard.
>Comes with an embedded uboot/linux/systemd/webserver with a full network access.
BMC is hardly botnet, and not only is it libre but it you can reflash it at any time and the flash memory that it boots from is socketed so you can physically remove the chip from the board if you want. Also the BMC only has network access through the first network port so if you want to disable it just connect the network cable to the second port. Its not anything like Intel ME, it doesn't have full access to the system.
Holy shit! That's a lot of L3 cache!
The 8 core Power9 CPUs have more L3 cache than the $4600 Epyc 7601, the 4 core Power9 CPUs have more L3 cache than the $10,000 Xeon Platinum 8180. Memory IO is the biggest bottleneck in quite a few enterprise applications so having massive caches can provide an easy improvement in performance.
I'm surprised there aren't any videos on youtube as their previous attempts at POWER8 based computer has 3 videos about it.
If I were them I'd send out a decent system to several FLOSS or Linux focused tech channels for a week each to do videos on, even smaller tech sites like Phoronix could get in on the action. It's been out for several months already and there's barely been any attention paid to it asides for a few benchmark articles and 2 blog posts.
You should be able to get by using a Threadripper waterblock since it's 569x638mm and the Sforza is 500x500mm. There's enough room around the socket on the motherboard for it to jut out a bit towards the ram.
Preordered the lite version. Looks good.
>Someone here might buy that
powerpc is gay and also botnet
How much watts does the dual CPU system draw in Linux idle?
Is it possible to disable that BMC shit (at least the bloated "features") or use a simpler distribution (I guess not)?
I would pay thousands for an off-the-grid CPU but I never audit the millions of lines of code that run on my computer anyway, so why bother.
I think Michael from Phoronix got one because he put up benchmarks for it not too long ago.
It's POWER, not ppc.
What I just said. Why would you buy a POWER rig to support PPC? I know PPC is derived from POWER, but there haven't been any recent derivations (the chips in the G5 were the last I believe). I doubt modern POWER instruction sets are compatible.
Let us know how it performs in Cryptonight. If it's really good, then mining Monero when not using it for other things could help pay it off and make the investment less daunting.
Raptor gave him SSH access to one, I don't think he physically has one
You'll never amount to anything so why bother living?
Because I don't have to audit my DNA to enjoy my life
Yes someone earlier in the thread said you can remove the chip easily or just flash something else in it. Its not necessary for the board to function unlike IME or PSP
Since POWER4 they've used the same instruction set as PPC. Just look at the list of compatible OSes on their wiki page
If this computer only ran AIX V7.2 nobody would care about it.
I don't have a power meter on hand to check my system but the Talos wiki has some numbers.
>Why would you buy a POWER rig to support PPC?
PPC64LE is the ISA of the POWER processors
>Let us know how it performs in Cryptonight.
That's one of the things I was planning on looking at. The POWER9 CPUs have enough L3 cache to run a Cryptonight thread for every physical thread so it should be quite fast, there are benchmarks for POWER8 CPUs which are pretty impressive.
The BMC does very little outside of acting as a remote console once the system starts to boot. When power is initially applied to the system the BMC boots and checks all components on the system once this is done the system can be properly powered on, after this the BMC basically acts as a remote console like SSH except it runs all the time unlike SSH which only runs after the main OS has booted. You can disable the BMC by not using the top networking port and instead using the bottom one, both are 1GbE and the main system has access to both but the BMC only has access to the top port.
You could hack up your own replacement for OpenBMC and flash it to the motherboard should you choose to.
Trying out the Power9 optimized fork of xmr-miner (https://git.raptorengineering.com/git/xmr-stak/), getting combined ~3100H/s across 64 threads which is just short of two Epyc 7601 running the same number of threads. Epyc is limited to only running 32 threads per CPU due to its smaller cache cache while the Power9 is using all available threads.
Also I just noticed that sensors actually shows power draw for the two CPUs, at idle in Linux each is drawing ~45W and at 100% on all cores in xmr-miner the power draw is ~115W each. This puts Power9 quite a bit ahead of Epyc in terms of power efficiency.
With this is mind...
8 Core Power9
Whats really funny is the cost of the two Power9 CPUs and dual socket motherboard is ~$1000 less than the cost of a single Epyc 7601 by itself.
>first he posts thumb nails
>theres a pube on one of the stands in first pic
God dammit Anon.
>looking for pubes on people's rig pics
I'm not sure what to think about your fetishes, anon.
>everyone buying things I don't like is a shill
boy it's sure cuckchan in here
>I assume you do more than shitpost with such a beefy workstation?
why would you assume that
This is precisely why it's such a monster at Monero mining.
Does the BMC act as a sort of rescue system too? How would an embedded GNU/Linux distro like ProteanOS do there?
t. pissraeli intel shill
Hmm. This makes it even more tempting. Is there a list of video cards it is compatible with? I've heard not all drivers will work with POWER.
I was thinking this rig with two Vega GPUs would be a great mining rig. One GPU could mine 24/7, while the other and the CPUs could mine whenever the system is idle.
>buy totally libre hardware
>to use a GPU needing a proprietary firmware to be usable
He could just use nvidia with nouveau supported hardware if he wanted libre software for his propietary hardware GPU.
>Hmm. This makes it even more tempting. Is there a list of video cards it is compatible with?
Any nouveau supported card is compatible provided you compile everything yourself. AMD cards are compatible if you pay for (((enterprise class))) GPU's as they need a propietary blob uploaded from the kernel that is CPU arcitecture dependent. AMD card's using the R300 series driver might be able to work without blobs provided you compile everything yourself, but those cards are ancient. Just get two 700 series nvidia GPU's and use nouveau for best mining performance with libre software.
Like RetroBSD and LiteBSD on dsPIC32's?
>You should be able to get by using a Threadripper waterblock
Duh. The problem is mounting it. I'm a programmer - I don't have the skills to DIY a mounting kit.
I just remembered that all HDDs and especially SSDs are botnetted now. What do?
Wrap them with a double layer of tinfoil. Triple if you're paranoid.
Keep in mind that 32-core 2nd gen Threadrippers are coming in few months. For ~$1500 you will get 32C at at least 50% higher clocks than that EPYC. More if you're willing to overclock. Counting in mobo price you might get more H/s/$ that way (though likely less H/s/W).
Yes. I think BSD 2.11 was the first version without AT&T code, so you could run BSD without buying a license from AT&T.
It was Weird modified ARM ISA called StrongARM.
Where are you getting that $1500 from? I would expect it to be at least $2000.
There's a single libre and foss ssd firmware chipset you can make or buy and set on an existing SSD, jewgle it. But for HDD's you are screwed weirdly enough. If you don't want to get that single openfirmware SSD, then you need to buy a HDD without the S.M.A.R.T support as that is the botnet/backdoor of the firmware. But those drives without S.M.A.R.T are very small capacity and very old.
>buy the botnet and insecure x86 architecture goy
What about using a NAS with super strict firewall that only allows encrypted SSH and NFS4 communication?
NVMe SSDs run circles around SATA interfacing SSDs; and it's an 'Open Sores' standard -- do you know if they have documentation for their firmware available?
So? What's it gonna do? Store hashes of 0.00001% of the encrypted streams it the controller handles?
Inject malware into your BIOS?
>Interface compatibility (SATA/SAS/NVMe)
Yes, it appears to support up to 1TB NAND modules in the interface of your liking. The only downside to the project is that the firmware board is based on the ARM arcitecture. There's a FPGA design that you could probably convert to use RISC-V or POWER's ISA instead for maximum security. I haven't researched this very much, do your own research.
>what is ZPO
>what is a persistent memory for virus
>what is sending a special code encoded in the bitstream passing through the SSD/HDD firmware that triggers the S.M.A.R.T backdoor.
>what is not having control
It's fucking botnet.
This doesn't solve the inherent insecurity of S.M.A.R.T enabled SSD/HDD's. Just take a look at the openssd project above for an example, they use ARM proccesors based boards which are vulnerable to specter like vulnerabilities. Now imagine what the harddrive manufacturers use to save money, its like a entire OS ready on a HDD/SDD. But the OS is like a minix lite OS specific to the HDD.
I wonder how many HDD's you would need over SATA to hack in order to use their proccessing power for something useful?
>Now imagine what the harddrive manufacturers use to save money
For regular HDDs, they probably just use microcontrollers, and so are safe. For SSDs where they need CPUs for maintaining data integrity and fetching bytes scattered throughout the drive, their OS probably has no userland and runs everything ring0, since it is not a user OS. Definitely no networking code, so the only danger would be built in malware or malware injected firmware blobs from your kernel or BIOS, in which case you're already hosed.
>Does the BMC act as a sort of rescue system too?
If the main OS halts or otherwise shits itself then the BMC acts as a sort of backup SSH so power cycle the system and recover it.
>Is there a list of video cards it is compatible with? I've heard not all drivers will work with POWER.
I briefly tried installing the OpenCL/RoCM drivers for the WX7100 but the AMD repos only have x86 builds (Ubuntu only comes with the AMD display drivers), I should be able to just build them form source though. Nvidia has Power9 builds of CUDA available but those are proprietary.
>He could just use nvidia with nouveau supported hardware if he wanted libre software for his propietary hardware GPU.
nouveau is awful, also it doesn't allow the GPU to be used for compute which is what he is interested in
>Keep in mind that 32-core 2nd gen Threadrippers are coming in few months.
They still only have 64MB of cache so its not like you can run any more mining threads.
>at least 50% higher clocks than that EPYC
The 7601 is clocked at 3.2GHz, the 32 core Threadrippers are only anounced up to 3.4GHz.
>More if you're willing to overclock.
I can overclock as well.
>Counting in mobo price you might get more H/s/$ that way
The Talos Lite motherboard with an 8 core CPU costs $1600 and can run the same amount of mining threads as the 32 core Threadripper.
AMD's pricing habits and leaked listings.
I still don't fully understand what ROCm is. It seems to be related to heterogeneous computing, but how does it work. Is it a bunch of libraries, compilers?
>supports up to 2TB RAM
While I'd love to have a Libre software supported GPU with full 3D acceleration, it probably isn't possible for any card yet.
Nvidia is still woefully incompetent and malicious with supporting Nouveau and care only about the proprietary driver at this point.
AMD is at least slightly better supported, but there are still huge problems with proprietary firmware in each card.
I've only ever seen Intel have good free software support without proprietary dependency.
They specifically advertise that on the 8 and four core you get double the cache per core because none of the enabled cores are paired.
>Good luck trying to find an eatx case
That's what this is for:
Pretty sure it was 4.4 BSD lite
I come from 4chan
your thread seems to be missing some images
Actual Talos II Lite board