/talosii/ the board we need

It is, but you'll be using the BMC's 2D GPU, so no 3D acceleration

Attached: awaiting heatsinks.jpg (1920x1280, 366K)

nice

Mo cores! whoo!

Attached: full usage.jpg (1616x1730, 616K)

Attached: 1515886847749.jpg (657x527, 31K)

Attached: 8ch TalosII.jpg (1400x1050, 706K)

Chromium porting:
wiki.raptorcs.com/wiki/Porting/Chromium

Attached: chromium p9.jpg (3116x1452, 471K)

Jow Forums's favorite plus:
Adélie
Debian
Fedora
Whonix

wiki.raptorcs.com/wiki/Operating_System_Compatibility_List

Attached: gentoo.jpg (1200x1079, 246K)

SAS module does need to load a blob, but it's optional

BCM reverse engineering is underway
wiki.raptorcs.com/wiki/Project_Ortega

Attached: drive backplane.jpg (1600x1200, 363K)

at least everything that runs on your CPU is open, and it has an IOMMU, so BCM and SAS aren't a big deal.
Besides, your hard drive or SSD has proprietary controller firmware, so you should be encrypting anything that leaves the CPU anyway.

Attached: Everything is opensource.png (522x310, 37K)

SAS free board looks more like this

Attached: T2P9D01-no-SAS.png (1182x862, 1017K)

Needs more BSD

Attached: freebsd.jpg (750x537, 77K)

Attached: no proprietary firmware here.jpg (1200x1600, 522K)

Attached: Yo, Cores in Cores.png (1081x774, 171K)

I actually really, really want one of these POWER9 machines. It'd be an excellent workstation and I could retire my other machine for and use it exclusively for video games.

Too much of a poorfag :^(

how much did it cost and how well does it run pretty much anything?

I bought the mobo + dual 8 core CPU bundle for around 2k and bought the rest of the parts from amazon. It came out to around 7k.
As for the software, it runs pretty much all major FOSS and what it doesn't run I just port.

they also just announced a new cheaper single socket micro-ATX board (pic related)

Attached: 10adc87317c0242a1644ec0d0c490696b94cac70.jpg (3090x2757, 811K)

damn man thats pretty expensive. i get it but i think ill just stick with amd until its closer to the 2k mark

Very nice. I cheaped out and only got the Lite one with 16 threads. It's still nice.

It's also SAS free, but I run a LSI SAS card (for SATA).

The base model is a nice little pre-built for around $2200 that comes with the case, single socket board, a quad core processor, 8GB RAM, and a 128GB SSD. All you'd really need to add to that to make it a great desktop for home use is a nice lower end video card and maybe a couple TB of HDDs, which don't really cost anything. You can get a decent display for like $100 these days, and mice and keyboards are just as cheap if you don't have those already.

I'd say the cost for a fully functional base model would end up around $2500 if you cheap out on the video card.

This would literally be perfect for me. The only problem is that their 8-core CPU doesn't seem worth it and the 18-core one is far too expensive.

How does it perform? Got any benchmarks for us?

You have to remember that it's SMT4, which means for every physical core, the CPU exposes 4 virtual threads. That means the 8 core has 32 threads, which is a pretty good amount imo.

I have kept that in mind, but I also know that POWER9 doesn't quite compete in some workloads with x86. For example, 2x 8-core POWER9 chips are barely able to beat my 1700X in timed Linux compilation.

phoronix.com/scan.php?page=article&item=power9-epyc-xeon&num=3

SMT4 doesn't give you 4 times the performance, and scales well only when you're not simultaneously using the same units in a core on all threads. That said, I would still *really* like one of these POWER9 chips, because I think it'd be a lot of fun to hack on software and porting it.

>2000 euro overpriced shit
No, thanks. I use blob free laptop for 40 bucks.

me again

Keep in mind that the benchmarks I linked to are sort of unfair. Not only does multimedia encoding make use of SIMD instructions on x86, but it's also hand-optimised for it. Furthermore, the EPYC and Xeon systems are massively advantaged in terms of real cores, where the Xeon system has 40 cores and the EPYC systems have 32 cores. The POWER9 only has 16 real cores, leaving it at a massive disadvantage.

Then clearly they're not for you, yet you felt the need to come in here and shitpost about it. Why?

I believe that was the phoronix article that compared a talos with full spectre mitigations to an x86 box with none. POWER9 ships by default with full spectre mitigations for user-mode processes which carries a significant performance impact. X86 vendors don't ship the patch for that reason.
Still though, it's true that POWER9 single thread performance isn't on par with ryzen. It's possible to dial back the spectre protections in firmware to gain some performance back, though.

Do you have any details on to what extent POWER9 is vulnerable to Spectre? What variants is it vulnerable to, or does it attempt to mitigate variant 1?

The hardware is vulnerable to Spectre 1, 2, Meltdown, and L1TF, but they're all mitigated in firmware (with varying performance impacts).
Raptor has them documented: wiki.raptorcs.com/wiki/Speculative_Execution_Vulnerabilities_of_2018

I see, that's a shame. I think I could live with it though.

Do you know about in-hardware fixes for Meltdown and L1TF? Those are the most serious problems for me. Wiki says Nimbus DD2.2 chips should have it, are those available?

Yes, they only sell DD2.2. DD2.1 was the pre-production test silicon.

Now make one that will fit a 10" rack and also create a 10" case.
Perfect breeding grounds for a new generation of open-source home networking environments built with affordable (due to not being enterprise-grade) hardware.

>POWER9 will not ship with vulnerability to Meltdown or any loss in performance compared with the current prototype chips (DD2.1).
Looks like they're already shipping DD2.2. There shouldn't be any performance hits then, nice.

Yep, I read further into the wiki.

Hey, I actually found a pretty cool post on exactly these Phoronix benchmarks, improving the performance significantly in some cases. Looks to be mostly a software optimisation thing.

sthbrx.github.io/blog/2018/08/15/improving-performance-of-phoronix-benchmarks-on-power9/

why cant you mount these on a rack? i dont see why not

Also turns out that I was wrong. It was dual 4-core, not dual 8-core systems that was benched.

In that case, this is extremely competitive with my current Ryzen 1700X, and I'm more than willing to pay a premium for it. I'm gonna wait for the Blackbird motherboard to see what kind of I/O and PCI-E it offers though.

I might actually get one of these machines.

Never mind, it *was* 2x 8-core, but I'm still getting it an 8-core chip. Just need the Blackbird motherboard to be released. It's going to be a lot of fun.

Is it even possible to run a Windows VM on this? If so, I'd assume a major performance hit.

It's a different architecture, but you *can* use qemu x86. It would be emulated x86 however, and probably extremely slow.

You almost certainly wouldn't be able to do anything computationally heavy in an x86 VM.

Because 10" is kinda small.
They would obviously fit a 19" rack, but 19" entirely intended for enterprise and thus in an entirely different price class. In addition, 19" gear is built on the assumption that you got a dedicated server room where cooling is more important than low noise levels, which is the exact opposite of what you'd want at home.

I want 10" to become the go-to equipment for home server usage. And the one thing you cannot yet get for 10" is servers and appropriate cases.

i get you. i think though its not worth fragmenting the market introducing 10" as a valid standard

>power architecture
>ibm

Attached: cat-looking-disgusted.webm (275x154, 240K)

7
7
7
7

>x86 architecture
>intel

Attached: tips auschwitz.jpg (700x609, 72K)

>tips auschwitz.jpg
en.wikipedia.org/wiki/IBM_and_the_Holocaust

>If so, I'd assume a major performance hit.
Except for the old versions of Windows NT for PowerPC!

Blackbird specs are available, just not price ATM.
It's pretty much like a cheaper version of the Talos II Lite, but with integrated audio out, and an open firmware SATA controller rather than the proprietary firmware SAS chip
talospace.com/2018/10/initial-blackbird-specifications.html
servethehome.com/raptor-computing-systems-blackbird-supports-up-to-8-core-power9/

Some optimization has not been done at all yet
Even just an initial pass adding VSX (vectorizing) instructions makes a huge difference. Pic attached is webm VP9 optimization.
src: twitter.com/trudluc/status/1011714979505065986

Attached: optimization is important.jpg (600x371, 20K)

Also, power delivery is intended for 8 and 4 core on Blackbird
>we should in fact be able to support the full 8 core device at listed clock speeds. When you go higher the board becomes power limited so, for instance, the 22 core device won't be able to go to full WoF frequencies.

Also down to 2 RAM slots
>even the 2D space is a bit constrained but the biggest constraint is actually in 3D. Any more channels would mean more layers, more regulators, and overall more expense. We are trying to go as low as possible in price
twitter.com/RaptorCompSys/status/1048490695714131968
>with the 4 and 8 core chips, it is harder to reach a point where DRAM bandwidth is the primary bottleneck, at least in real world applications. This machine is true dual channel, so should still be quite adequate for most use cases.

Attached: blackbirdOpenPOWERSummitAmsterdam.jpg (1536x2048, 409K)

Because the boot firmware and BMC is all open source, you can even do some *horrible* things, like make your boot status show up in Comic Sans (pic related)

Attached: openbmc.jpg (2048x1536, 528K)

Attached: Talos_ii_rev_1.00_bottom.jpg (4096x3700, 1.77M)

Attached: Talos_ii_rev_1.00_top.jpg (4158x3714, 1.26M)

There is also a crazy guy on the wiki building 4x Talos II cluster:
wiki.raptorcs.com/wiki/User:JSharp#T2.2Fx86_FSC_Heterogeneous_Cluster_.28veritates.40kraftwerk.29
>8x IBM POWER9 CPU (18-Core)

Attached: Veritates.jpg (1140x1905, 198K)

Attached: cable management desired.jpg (1600x1200, 506K)

imagine all the chrome tabs I could have open

>K8s development
>Real-time distributed raytracing, nonvolatile ram enabled persistent world simulation
WTF is this guy working on?

wiki.raptorcs.com/wiki/BCM5719

Its not fully open-source so don't call it that.

That's why there was a link earlier to Project Ortega

So, what does this magically give me over my Threadripper? Worse performance? All I can come up with.
Also no, "open source" is not an answer, I don't care as long as things work and they work. I'm not doing secret projects or related that I'd need this any security improvement from this (neither do you guys though).

>what
Security.

Distributed and high-performance computing, obviously.

Defeating the NSA boogeyman

more like the board we need, not the one we can afford

Control. Intel and AMD collaborate with hollywood to implement hardware DRM and lock you out of your own computer at the firmware level. With POWER9, you're not going to be watching 4K Netflix, but you can inspect and modify every component of the system firmware.

how the fuck you have 176 threads

two of these
raptorcs.com/content/CP9M08/intro.html

>POWER
>mATX
>green PCB mask in 2018
looks like perfection

3... maybe 4

You'd need to emulate it, so it will be slow.
I wish there were PCIe cards with x86 CPUs for this exact purpose, there was this one old mac that had an expansion card with a 486 CPU to run windows and whatnot on.
Either way, x86 is legacy shit that doesn't have much time left, RISC-V will take over.

Yeah, we'll see how much the Blackbird boards are, I'm hearing ~700 USD for the board, but I hope they sell a bundle

The current Talos II single cpu bundles throw in a 4core + HeatSinkFan for free

could be nice seeing a new mainstream CPU market that's not a duopoly
I don't see us non-enterprise plebeians being allowed to use any sort of CPUs that aren't backdoored despite RISC-V being a thing
it would probably come with a small closed source chip that's basically another Intel ME

No, it won't - the end result of RISC-V will probably be like ARM, only big companies will be able to make the fast chips, and their designs won't be open

Remember why RISC-V is taking off in the first place, it's BSD licensed. If it was GPL, no large corp would touch it, and no-one would be interested in licensing it, since they couldn't add proprietary extensions.

Hell, even SiFive was intending to keep initialization code proprietary, and they've just barely started.

NO! NO! STOP BUYING POWER9 CHIPS! YOU NEED THE GAMES AND BACKWARDS COMPATIBILITY! BUY INCEL CORE XTREME DUO QUAD POOPYLAKE REHASHED SHIT, GOY! IBM SINGLEHANDEDLY KILLED 20 GORILLION JEWS! REEEEE!!!

Attached: 7c2.jpg (267x297, 24K)

RISC-V could eventually make a good replacement BMC instead of ARM though...

Yeah really.
I'm just waiting for someone to outright say it:
>Nooo, these chips aren't meant for you; their for the ENTERPRISE only, the CLOUD, BACKEND, COMPLICATED stuff, you wouldn't understand what to do with it
>Now go be a good pleb and buy more tablets and phones

It's really funny when you look at who is the new market for these, Google & Rackspace. The hyperscalers know what's going on, they want control, and to have *their* equipment opensource, but consumers are too stupid to care.
Look at who's also funding Coreboot and LinuxBoot and Open Compute Project; Facebook and Google keep showing up.

But user, what about the 40 gorillion who were brutally murdered by punchcard machines? Think of the children!

ARM isn't any more free than x86. RISC-V on the other hand is free for the core ISA.

And if you're developing a new core, with RISC-V you not only get a well thought out ISA for free, but a large library of software and tools designed to work with it too.

The only downside is that you're helping novidya for free since they're gonna use open source compilers for their future RISC-V based closed-source GPUs, but it's a compromise worth making.

Dubs and I place the order

Attached: 1438337081770.png (249x281, 141K)

I'll play

Oh COME ON

Having a free ISA is nearly worthless for the end user. There's no guarantee of an open source implementation, no guarantee of any open source software for it, and even if people do port it the manufacturers could just build a locked down SoC that runs purely proprietary and closed source software like Apple does. And they will do this because normies don't care.

RISC-V supporters are cucks. They're giving away all of their shit for free and just begging big companies to bend them over and fuck them. RISC-V is a stupid redditor pipe dream with no future. I sure as shit won't be porting over the free software I maintain, because RISC-V is how you get burned. It'll be exactly like MIPS where you spend countless hours porting a free toolchain and making everything nice and easy to use for developers only to get slapped in the face by corporate jews.

The answer to RISC-V is "no". Fuck off.

And you lost to an "actually it's GNU/Linux" post

Roll

basically, no open source CPUs without seizing the means of production

Personally, because how will you trust whoever is assigned to "seize the means of production"

quints are not dubs

>muh commie shit
The answer is simply government oversight and regulation, which includes breaking up monopolies and tearing up bogus intellectual property rights claims. Communism has failed over and over again. We're not trying your specific flavor of it again, numb nuts.

there are already FPGA implementations of RISC-V

it's as close as we've gotten to an open source core as it gets except maybe for opensparc, where everything was released except for the silicon layout, so they're actually more or less at the same level of implementation

>all those electrolytic caps in the top left corner
>total of 2 DIMM slots
That better just be prototype things

explain what's wrong with the caps

>my eff pea gee ayy iz as faster as a power9!!
Yeah, call me back when you can use your shitty little 1MHz tinker board to run anything bigger than a Hello World program.

Attached: vO7lRZ7.jpg.png (621x702, 56K)

>REV:1.00
Board might be a prototype
2 RAM slots is the design though