TALOS II secure workstation is back!

I'd love to see alef ressurected for a plan9 system on Talos :)

...

Oh yes, here's the defeatist. Tell us about how everything is futile. Fuck earth, amirite? xD

The webdev is probably a victim of "best practices TM!", try and learn modern node and weep.

Are you saying I could turn this into a lisp machine?

You could in theory tweak the microcode to run Lisp at bare metal speed. Or legacy x86/ARM code.

spit out the RMS pill and take the ted-pill.

the romantisization of lisp machies is cute.
the special purpose mechanisms for reducing dispatch, tagged pointers and garbage collection overhead have all been surpassed by modern general purpose cpu tech.
A more realistic system would be a "libos kernel" combining a microkernel with the racket run-time.

It would be cool to go back and see what can be done with todays tech, I'm sure that you could create toy kernels using forth, lisp, ada, etc.

It's just amazing that this devices makes such an idea possible.

How weighty is the racket runtime? POWER9 L3 cache is 120MB, you could have the whole thing running in CPU.

The thought of running a whole gentoo install inside the cpu cache amuses me. That thing is fucking expensive though. And technically the hardware is not secure it's just backdoor free. You can still do things to it while it is powered on to attack it. The POWER arcitecture was never meant to be secure only fast. This project was not meant to be secure, but again, fast and backdoor free.

What do you mean by secure hardware? Built in privilege levels like risc-v? An ME or PSP hypervisor but open source?

Is there some type of security feature present in x86 that can't be fixed in microcode?

Try holding a magenet up to a proccessor while it is running. If you can fuck with it that way you can get data off of it in other ways i.e it is insecure from a hardware standpoint. Also if your derp enough to try this then use a computer you'r willing to wreck.

Yes it is called a hardware debugger to the physical proccessor by a dedicated as fuck(think elite tier of the nok set allocation agency) adversary. Though I couldn't imagine what would get (((them))) so anally ravaged at you to be that dedicated. Maybe things related to pizzagate? Or things related to nukes being fake? idk.

If they offer the 24 core 8 threaded versions that's 384 threads on a single system.

The closest you can get to being sure of no backdoors is to implement everything critical on FPGAs. The Power chips are open though so you can feel quite a bit safer than x86.

System management mode. The original ring -1.

The only thing this would be useful for is a few edge cases and lots of VMs.

I can see the server world liking this.
Or the AI world.

Not all of them;
but those used for the talos are.

Why is everyone assuming you need two POWER CPUs?
You can just use one CPU.
The 12 core 4-SMT variant should be a good match against something like Threadripper considering they're both 14nm.

Price has me worried if it's anything like the $5000-$6000 of the first attempt. I'm a dirty Australian so it'll be even higher.

Only if their software stack runs on it, that's the critical problem trying to compete with Intel.

Yeah OpenPOWER is a new-ish IBM thing. Funny how they were once one of the worst enemies to FOSS.

uhh why there isn't a i7 option for gaming???

Rumors on the TALOS irc is that IBM are helping out with the board to make it cheaper.
I too would get destroyed by best beaches tax, it's what killed my pledge for my first one.

>(((ibm))), mother of copy+paste A.I in cy+2, helping make motherboard parts
Welp it is backdoored already if true, FUCK.

Your making software wrong then. What everyone does and best hardware practice do not line up. Because paralellism and FLOPS per second do not line up on current proccessors. Devs use a single line of instructions. This is because of heat issues of using multiple cores on x86/CISC architectures. But on RISC architectures this is backwards. You use less heat on many more cores which is paralellism. But with all software being single threaded it is difficult to learn/port to other architectures and get good performance without software dedicated to such a job. This is yet another reason why POWER was used on supercomputers. A better question is how to easily split single threaded proccesses into mutli threaded proccesses and maintain execution order for easy porting.

They are still claiming open firmware, if you don't trust IBM then why trust their CPUs?

He's retarded and doesn't know what cause he's shilling for, so he just attacks all sides equally.

Putting le epik pol maymay lithp parens around a company that supplied the nazis and maintains a dress code is a dead giveaway that he's criminally ignorant.

Two POWER 9 CPUs probably cost $5k USD alone.

The POWER8 CPU prices on the original TALOS ranged from $1135 to $3350 each. The newer POWER9 CPUs are probably going to be comparable in price, and there will probably also be single CPU options available. Expensive for sure, but what really killed the project last time was the sticker shock from the $4000 motherboard, and the minimum price of $8000 for an assembled system. If they can bring the price down to $1000 for a motherboard or $3000 for a complete system, they will probably bring in a lot more customers.
crowdsupply.com/raptor-computing-systems/power8-cpus

The price was indeed heavy.
But the worst kill boner was Francis Rowe who went full retard in the middle of the campaign.

The three criteria I'd say they have to hit are:
- price around $2000
- power consumption at desktop levels
- performance around ivy bridge levels even after whatever underclock/undervolt they have to pull to get the power consumption reasonable
Or in other words, they have to take server parts and force them into a convincing enough impression of desktop parts. If they can manage that I'm sure there will be tons of people all over it.

Had literally nothing to do with the campaign, it decided to promote TALOS on the libreboot website which got people thinking there was some involvement.
TALOS is purely a product of Raptor Engineering, specifically, Pearson of the coreboot project.

I could deal with a 2k or even 2.5k motherboard, There are workstation and gaming motherboards that reach up to 1k after all.
There will be a price for freedom, but to be 4 times the price just made it too hard for me.
Probably a little higher than threadripper levels, in a way AMDs core hysteria has made POWER9 look a lot more reasonable

I looked at some (((benchmarks))) that show POWER8 is roughly comparable core-for-core and clock-for-clock with Broadwell-EP Xeons. POWER9 is too new to find any data on it.
anandtech.com/show/10539/assessing-ibms-power8-part-2-server-applications/8
anandtech.com/show/10539/assessing-ibms-power8-part-2-server-applications/9
POWER8's power consumption in load situations isn't that different from high end x86 desktop hardware. There are 130W 8-10 core models, or 190W 10-12 core models. The 130W models have almost the same TDP as the FX-8000 series at 125W, while the 190W 10/12-core POWER8s are comparable to Threadripper at 180W, the FX-9000 series at 220W, or the bigger Core i9 CPUs at 165W. Obviously, add a second CPU and these numbers double.
I realize that you want a 65W 4/6/8-core PowerPC instead of a 130W 8/10-core POWER, though. I would suggest capping the maximum frequency in software, but idle power consumption appears to be about 85% of full load power consumption.
anandtech.com/show/10539/assessing-ibms-power8-part-2-server-applications/10

One thing does concern me about POWER9.
en.wikipedia.org/wiki/POWER9
>The specification for Power ISA v.3.0[28][35] was released in December 2015. It is the first to come out after the founding of the OpenPOWER Foundation and includes enhancements for a broad spectrum of workloads and removes the server and embedded categories while retaining backwards compatibility and adds support for VSX-3 instructions. New functions include 128-bit quad-precision floating point operations, a random number generator, hardware assisted garbage collection and hardware enforced trusted computing.
en.wikipedia.org/wiki/Power_ISA#Power_ISA_v.3.0
No citation, of course. I found a PDF of Power ISA 3.0 and searched through it. The only thing of interest I found after a few searches would be Logical Partitioning (LPAR) described in Chapter 2 of Book III on page 945 of the PDF below. LPAR doesn't seem to be automatically malicious. I think it's just IBM's implementation of hardware virtualization, similar to what you would find on newer SPARC hardware. Hardwarefags should download the PDF and see if I missed anything.
docdroid.net/tWT7hjD/powerisa-v30.pdf.html

Trusted Computing isn't necessarily a downside but it all comes down to who has control over it. The user should. I believe the original TALOS had some trust-enforcement stuff but you could introduce your own root keys and such.

Would anyone here like to crowdfund one of these? Tiers could be based on money contributed and hours usage per day. The biggest contributer can pay the power bill too. I'd really only like one for shitposting in screenfetch threads.

I wonder, how long until someone licenses the Amiga brand and sells one of these for 40%, just so some nostalgic fanboys can run (((AmigaOS))).

I guess you haven't heard of the Amiga X5000 yet?
amigaos.net/hardware/133/amigaone-x5000
arstechnica.com/gadgets/2017/05/the-a-eon-amiga-x5000-reviewed-the-beloved-amiga-meets-2017/

I was actually mildly interested, but fuck that price. I'd be better off using an actual Amiga with prices lime that.

It does look like POWER LPAR is their version of SPARCs LDOMS.
Hardware enforced trusted computing is not a problem if you have control over what the hardware enforces.
It becomes a problem when Intel and Steam get together to create unbreakable DRM.

POWER8 Benchmarks might be useful to extrapolate from, remember POWER8 was 22nm whereas POWER9 is 14nm.

...

I am aware of that.
Haha
I see the grudge here
Indeed and that's unfortunate.
It had an inevitable correlation at that time since the libreboot project was promoted on for the Talos.
I hope the campaign will succeed.

Listen Francis, you should be intelligent enough to see that trying to out left the super lefty FSF by causing a shitstorm is not a good thing, and then vaguely attaching your name to a project during said shit storm would be more harmful than helpful.

...

See what I found on cc.bingj.com/cache. It seems like the Microsoft botnet is really fast when it comes to this. But there's still no price to be found. We just have to wait a few more days I guess.

racket-minimal is about 10mb or so iirc.

That's very nice...

Like the ability to generate proper random numbers, don't allow memory leaks from L3 cache, properly isolate execution of software, implement AES and other hardware crypto accelerations properly, etc.

Noice

Is there any first party or even third party open source project similar to this?
en.wikipedia.org/wiki/PowerVM_Lx86

Would Qemu be the only option?

Currently impossible.
Fair point.
Is this really a hardware feature? Providing you could secure memory the OS kernel can do the rest.
Literally disabled in BSDs due to hardware crypto either being backdoored or badly designed.

Fuck yeah. If the price is reasonable (and if I had money...) I'd be all over that.

It's technically possible, but I don't think anyone has gotten desperate enough to put a radioactive isotope in amongst microelectronics yet.

...

this is a freaking ad
MODS

this mods delete this faggot shit

...

Shit NSA, you're sounding desperate, please don't Clinton tpearson

This is nice.

This is still SMP just like with any other CPU; you get diminishing returns per-core every time you add a thread. At eight virtual cores, you're probably looking at somewhere in the neighborhood of 1/6 the performance per virtual core, depending on how the physical core is built. It's still incredibly powerful either way.

Even mainstream video games are beginning to go parallel, but in reality, it's up to the end user to use a system appropriately.

Needs way more usb ports

holy fuck

They weren't kidding when they put "high pressure" on the spec sheet, you might need an impact tool to affix the HS

youtube.com/watch?v=8eULB8uMIuc
This video is about Intels SGX extensions which will essentially lock you out of owning your CPU good and proper.

While only tangentially related I think it deserves to be in this thread as it is one of my main reasons for considering the TALOS II.

I want something that is actually owner controllable.

Use something older.

I have actually had the chance to play on a Minsky server and its no way near that bad. The Power8 CPU cores have so many execution pipelines that in most cases even running 8 threads on a single core you still get very good performance.

wew

Fuck if it's $3000 I'll buy one...if they keep they the integrated FPGA's.

Looking at the board shows only one chip that might be it, under the heatsink on the top right.
What did the talos claim to achieve with the fpgas anyway?

don't forget you can purchase unique sets of random numbers from institutes that maintain those isotopes

though, of course, that means trusting an outside source

Do yourself one better and make a lab with an particle accelerator to seperate atoms for the isotope yourself.

www.fourmilab.ch/hotbits/hardware3.html
www.aw-el.com/price.htm
www.unitednuclear.com/index.php?main_page=index&cPath=2_5

So a Smalltalk machine like the old Xerox Alto?

There is no such version. There are 24 4-way threaded cores on the die, in pairs that share L1 cache and memory unit. It can be factory-configured with each core pair combined into a vertically-clustered 8-way "core", where every thread can really only access resources from one core half and is no faster than on the separated core config, even if it's the only one running.

So you either have 24 4-way threaded cores or 12 8-way threaded 2-way clustered "cores" config, with the latter being utter nonsense bullshit that exists solely to jew software vendors that charge license fees per CPU core.

You can listen to cosmic rays with a radio like random.org.

fourmilab.ch/hotbits/
You could do the same with some americium from a smoke detector.

That's actually a good idea but you have to take from multiple source to make a good randomness.

So a thermonuclear weapon?

They used it to implement some chipset functionality. No ASIC available on the market for this purpose was documented well enough to satisfy their openness autism, so they reimplemented that shit themselves on an FPGA.

archive.is/XelN0

I'm pretty sure they did their homework and chose an FPGA that's as open as possible.

Nobody's stopping you from buying just one and leaving the second socket empty for a future upgrade. You don't need to buy the top-of-the-line 24-core beast either. I'm sure there will be cheaper SKUs with some cores disabled available.

The previous-gen POWER8 processors were actually pretty cheap for performance they provided (at least the lower-end models). It's the TALOS mobo that was ridiculously expensive and blew the complete system cost through the roof.

For the record, the top-of-the-line Intel Xeon 8180M goes for $13k per pop.

I call bullshit. Around $500 is the ceiling.

[laughs in Australian]

Here in Brazil a shitty chinese computer fan that will kill itself in 1 day costs 10 dollars.
Non-shit fans are 30-50 dollars a pop.
My gtx 780 was 1200 dollars in late 2013.
COMMUNISM WORKS

But that price is in kangaroo bucks, right? If so, that makes it around 700USD at current exchange rate, which is not that much of an import premium.

Also:
That's not a bare mobo. That's a mobo+cooler bundle.

seriously, niggers?
SERIOUSLY, NIGGERS?!?!

Fug. I had hoped they would use some PC-standard heatsink mounting bracket, so that customers could use their own coolers of choice. That bundled HSF looks noisy as fuck.

Qemu has a user-mode emulation mode that is pretty similar to what you linked. In fact, the Raptor guys already demonstrated some older commercial x86 games running smoothly on the TALOS prototype under Qemu.

I looks like the brackets are screwed to the board, you should be able to replace them with whatever fits your preferred cooling solution.

...if I can get alternative brackets fitting both the cooler AND the screw positions on the board. I seriously doubt any manufacturer will waste time supporting a fringe autist system nearly nobody uses.

BASE BUNDLE = LOWEST CPU
That means the bare minimum you can get is a 16 thread CPU.

EVGA SR-2 dual socket mobo equipped with 2 hex-core xeon's is the botnet free gaymen beast.

My current rig is an i7-930 @ 4000 mega power

The last time the bundled CPU was the only CPU they offered. And when you spend $$$$ on an expensive platform, it's not to run an anemic quad-core on top of it.

Also, a dual-socket NUMA config for a measly eight cores doesn't make any fucking sense when you can get more cores in a single socket without any NUMA bullshit. Which leads me to fear you can, in fact, not get them. It might be that IBM threw at this desktop experiment some quad-core binning refuse nobody would buy otherwise, while reserving all more powerful (and interesting) parts for their usual large corporate customers.

user, the botnet was there since well before the i# branding began.

be specific. Intel cpu botnet begins with sandybridge, afaik.

The Intel Management Engine botnet was present in Core 2, but you could easily remove it without consequence. Everything beyond Core 2 has built in botnet that has only recently been neutered.

That must be some new definition of anemic you're using. Remember that the challenge with Talos has always been taking top end server parts and fitting them to the price/power envelope of desktop parts.

Sandy Bridge started exposing botnet features for the (corporate) user/admin. However botnet itself was there many generations before.

The last time they offered a bunch of CPUs, it's link is in this very thread .

I agree that it doesn't make much sense to buy the 2 CPU base option but offering a single base is good, it shows what the lowest barrier of entry is.

user, the last attempt they made was with an octacore POWER8, and the CPU ended up being the cheap part compared to the mobo. Also, while the entire system would be expensive, it would have enough grunt to warrant the price tag. It was expected to be within the performance range of similarly-priced Xeon systems.

Now retain the expensive part (mobo) and slash the cheap part (CPU) in half. What do you get? A computer that's still expensive as hell and doesn't measure to its price. Sure, it'll kick butt of an average "high-end" gaymer shitbox under well-threaded loads, but that's not what a multikilobuck machine should be compared to.

A system in this price range would go against toppest-end professional workstations. The kind built last year with Core i7X or Xeons with four-digit sticker prices for the CPU(s) alone. And Threadripper bringing this performance tier's price down to high three digits doesn't help TALOS's value proposition.

Price? By mobo's pricing alone the highest-end of the market is the only place it could dream of competing. And there's only so much you can do to bring its price down when the production volume is so small.
Power draw? Nobody cares about that in high-end parts. Remember the 225W FX-9xxx gayming shitchips? They weren't even high-end, and people bought them. Overclocked monster builds can reach 300W on the CPU and high-end PC mobos and coolers are designed to supply and remove this kind of power. Heck, GPUs routinely exceed power ratings of the hottest CPU. The tech to deal with it in desktops is there. That's not a problem.

IIRC they intended to offer multiple CPU choices in the beginning, but quickly scrapped that idea to the "would be nice in the future" pile. The base (and quickly only) model was a 130W octacore, which was the bare minimum expected from a system in this price range.

If you don't provide enough performance to justify the price sticker, your audience will be limited to big corpos looking for devkits to write software for their big-iron POWER servers on. And that's not a big market. If you want to attract private customers and smaller companies, the value proposition needs to be there. The bang-per-buck need not match the x86 competition, but it should be close enough to justify paying extra for the stallman factor.
I don't expect enough bang to be there without at least eight cores per socket, like in the first TALOS. Hopefully for a reasonable price.

You are wrong.
What you say is true for their prebuilds but they never removed CPU option for those buying the bare mobo + CPU, in fact they increased the amount later in the campaign as an effort to draw more funds.

If they ONLY offer this option then I will agree with you but our only datapoint goes against what you are worried about.

I hope you're right.

Are australian cities named as homosexually as possible on purpose?

It is you who is homosexually inclined for you find strong homosexual connotations in everything you see.

Bent-lee
Coh-burn
Wan-ga-rah

Brazil was communist?

If we define socialism as "following the communist manifesto but not fully", Brazil is heavily socialist right now. But Brazil was never entirely communist, and the one time it almost happened the US planned a military invasion of Brazil but the problem solved itself with a capitalist military coup.

thats a meme-tier definition, "communism" isnt "following the communist manifesto exactly". The manifesto was a party program of immediate goals for its time, not the definition of communism and what it entails to be 'followed' by future socialists. government regulations and welfare doesn't make a state 'heavily socialist.'

Damn user, I would have expected far, far worse than that in brazil from what ive heard about vidya.

It's basically a rule of thumb that anything computers at the very least doubles in price in Brazil.
It's also an impossible market to compete in because companies have tiny profit margins so people can actually buy things, which means only big companies that can mass produce things have a chance of surviving, and they don't profit much by doing it.

How much of this today has to do with Brazil's attempt at "self-sufficiency" during the military dictatorship, that banned most imported consumer goods, including the (almost all American-invented) PC platforms that sprang into existance?

The reasoning is the same but I wouldn't say it's because of the dictatorship.
The dictatorship didn't hurt the economy, it brought something we call "the economical miracle", our economy progressed like never before.
Basically Brazil proved a tyrannical government is by a large margin better than communism.

So you 'strayans suck coh? Interesting.

TALOS IS LIVE
Basic Bundle = $2,300
PCIE 4.0
Fucking hype

NIIIIIIIIIICE

Its interesting that they have an optional Microsemi SAS controller.

Just as I feared, only quad-cores available. However, $310 for the CPU is a nice price. Dat $2150 mobo though...

Yeah you were right, tpearson on IRC says that better CPUs should be available later on.

I am tempted to get the Basic bundle and get the CPU upgrades later as the mobo + cooler sans CPU is 2260.
Might as well spend 40 bucks more for CPU.

The cooler is expensive too, and no off-the-shelf alternatives available due to the wacky mounting system.

I wonder if the mobo supports asymmetrical CPU loading. Could it work with one quad-core and one octa-core put in?

ALL OF THE BONERS

quad-core boners are pretty soft tbh

Depends on what hardware you actually have.
I still work in LGA775 platform.
I could upgrade to anything it would still be better.
But I was saving for the past decade for complete libre hardware so now I know where to put it.

When you compare it to a 16-core Threadripper you will be able to get for roughly half the price (CPU+mobo), it does deflate quite a bit... An octa-core Ryzen 7 1700 + mobo costs ~$500. The stallman tax is steep on this one.

The key thing is the idea this might have a future, an even slightly price and performance competitive future, with even moderate volume. As opposed to, say, the final Amiga and RiscOS machines built off the dregs of embedded PPC/ARM CPU offerings.

not a very solid justification to throw $2300

True. If it were within 20% of a top-end Xeon workstation, that wouldn't be such a tall order. The part that gives me hope is that most of the price premium is the mobo, rather than the CPUs. It's a shame nobody in wintel land shares POWER9's bus, like certain AMD Athlons & DEC Alphas did, for instance.

Top-end Xeons cost >$10k per pop though? I'd be happy enough if it traded blows with a ~$1500 x86 system. In fact, everything depends on the pricing and availability of higher POWER9 SKUs. If they manage to sell octacores for under $1000, bang-per-buck will probably be acceptable to me.

How much better is this than the d16 workstation. (2x sixteen core AMD opteron, D16 Motherboard, GTX 660 ti)

For starters, it's supported and new replacement parts are being produced. It's open down to the firmware and microcode level (the Opterons still use microcode blobs).

The POWER9 will smoke the Opteron in single-thread workloads. (easily >3× faster, if not much more). Under heavy multithreading, the quad-core base config will likely lag a bit behind (except in memory-heavy workloads), while the 2×quad-core will be clearly superior. Of course, the mobo supports higher POWER9 SKUs (up to 24-cores per socket, if you can get your hands on one) so there is room for upgrade in the future.

You can't use nVidia's GPUs, since nVidia doesn't release public driver builds compiled for POWER and the open nouveau driver sucks massive cocks. AMD GPUs will work well with open drivers though (Raptor already demonstrated them in action on the first-gen TALOS prototype).

That's the current definition in the year of $329 consumer octa-cores.

Are you insane? Port nouveau over to use it on openbsd. They have made leaps and strides in usability at 3/4th's the performance of nvidia's driver.
AMD's

Because CISC is not fully retarded when we're talking GPUs.

Yes on cisc you need dedicated proccesors for specific tasks.
But I am confused. What are the data proccessing benefits of a dedicated graphics proccessor(GPU) on a RISC system? Just simply add more powerpc proccesssors at that point so you aren't coding/porting for two different architectures?

Unless your going to use legacy opengl and future vulkan based software(i.e games). But who in their fucking right mind would use powerpc for gaming? You lose on all the benefits of the vast library of x86 games and are limited to open source emulators to compile for powerpc. This is also not the target audience of the talos workstation.

OpenBSD doesn't support any Nvidia cards
There is no ppc64le port for OpenBSD.

Hence why I said ported you faggot. It would be a decent amount of work.

A lot of work and would most likely be not accepted by theo, bluetooth got nixed because it wasn't good enough.
Nouveau, while a noble effort, is fighting an uphill battle against Nvidia and as such doesn't have the code quality that OpenBSD would want.

80x86 sucks, but it doesn't suck that much.

gnoo
please do the needful.

To be fair, the cheapest dual Socket P mobos are ~$600, and most hover around $1k. This thing is expensive, but not that expensive.

You mean the raptor tax.
Imo I prefer the money to go to raptor instead of monopolistic, "diversity" loving companies.

>>>/cuckchan/

Yea they do... Nvidia have Linux releases for both x86 and ppc64le as part of the CUDA toolkit, there are several OEMs which manufacture ppc64 servers which are usually configured with lots of Tesla GPUs and the IBM Minsky machines are the only systems with NVLink between the CPUs and P100s.

>>>/suicide/

The difference is production numbers and how static costs get spread out, the development costs and setup fees for manufacturing runs are significant for complex pieces such as motherboards.

...except the said CUDA drivers are:
a) not public, and:
b) don't support OpenGL or OpenCL, only CUDA compute. That's utterly useless for a desktop computer.

Wasn't Vulkan/DX12 supposed to render this distinction irrelevant? When is The Dream™ of full abstraction going to happen

Well, Vulkan ain't supported either. These are drivers for big number-crunching clusters, not a desktop computer.

And don't think for a moment nVidia will ever willfully resign from locking in customers to their proprietary CUDA ecosystem as a jewing strategy.

Not ever at any point.

Yes. Do it.

Uhm, yes? These are server boards.

*usually for server boards

I mean, Talos I and II are literally the only desktop Power systems in existence, why would Nvidea make an X-driver?

That was my point, faggot.

The "why" is irrelevant, the point is that their GPUs are useless for TALOS desktop users. Unlike AMD's ones with open-sauce drivers that can be compiled fine on architectures other than x86.

There's already a PPC/Nvidia X11 driver, albeit in a near abandoned state. I doubt Nvidia went to the trouble of making NVLink with IBM only to not supply drivers for the newer hardware.

Nah AMD/ATI opensource drivers have issues on PPC too. Even ARM GPU driver support is terrible and they're not out of reach or deprecated like MIPS/SPARC/PPC.

Although its really that *any* closed source drivers will be useless unless Talos really takes off and makes a market for desktop POWER9.

You're going to need to use Nouveau or AMD-GPU - and since AMD's Vulkan stuff is still closed source in userspace, use RADV instead. Nouveau might have Vulkan support in a couple of years at best.

Any more info on this driver? That seems really interesting.

see: raptorengineering.com/TALOS/op_qemu_gl.php

PPC ≠ POWER
Also, that driver supports only ancient GPU hardware.
This was already mentioned:

These drivers are CUDA-only. They do not support usual APIs for graphics rendering.

Actually POWER 3 through 8 including PPC 970, CELL, and Xenon are covered under Linux PPC64 arch. It wasn't until IBM decided to change endianess that a new branch PPC64el was made. Some POWER 8 and all of POWER 9 are PPC64el.

I'd like to correct my bullshit statement. POWER is bi-endian. x86 is the sole reason they've shifted to little endian and PPC64el arch is only for newer POWER processors. This doesn't change the fact that PPC 970 and POWER 3 through 7 ran the same PPC64 arch.

Is this a mac driver? Or are you talking about an opensource X11 driver?

There were proprietary Nvidia and Nouveau drivers which worked fine on PPC32/64 Linux a decade ago. After Apple dropped PPC, Linux development stalled except on IBM servers. The same is true for ATI/AMD drivers as well. Currently, 3d is either bugged to hell or not supported. There are PPC KMS issues on both sides.

Can you give a link for more info?

Neat, Freescale P5020 (2.0GHz,64-bit, 2-core, e5500 core); odd that it calls it a PowerPC CPU, even though it's Power ISA 2.06

lists.x.org/archives/xorg-announce/2010-March/001286.html
I'm a bit confused because I know I had hardware 3d acceleration working on Debian Squeeze, but I thought it was the nv driver I was using.

It worked well-enough for Raptor to do that demo. Given their track record, I expect they'll spend some time on fixing those snags once TALOS is out and shipping.

What demo are you referring to? CUDA already exists for IBM/Nvidia platform. If I were them, OpenCL would be my priority right now. There's no need for them to work on GPU drivers when basic 2d is good enough. I doubt many people will even bother connecting a monitor.

...

This looks interesting, but it's emulation.

At the end of the video they've shown some of the games running natively, for comparison. Anyway, emulation or not, 3D is working.

Ahaha, get fucked monkey niggers. I'll run you over with my car like I did in the 90s. I hate you. You glow in the dark.

And nearly two decades ago, nvidia actually had an open 3D driver. Funny how the timing coincided with all that Microsoft-bribing-Dell lawsuit stuff.

ut2004 there looks worse than pixomatic mode on a pentium 2. someone needs to tell ryan gordon to get off his ass and re-run make

...

...

I would. Any game that is open source can be compiled to run. This is a great way to force you to mostly only use free open source games, of which there are plenty out there

As long as you have a real GPU there's a ton of shit to play

...

Each core is 4-way SMT, think Cell SPUs but useful

GLES is fucking awesome. What ARM devices do you have? And which is best for glorious GLES gaming?

On Android maybe. OpenGL is the standard for 3d on GNU/Linux.

"server-grade" motherboard + FPGA and $200 cpu.
Great Value.

I don't you understand basic economic principles such as economy of scale. It's quite obvious to me that this Raptor company doesn't have the same scale as the big ODM have. Well guess what, they have to start somewhere. That starting point is usually the most expensive.

As far as Im concerned, GLES is the standard for 3D on *ARM*. And it works amazingly well. I have an old Toshiba AC100, which is a lowly ARM dual-core 1GHz with 512MB shared RAM. What that thing can do with such low specs on GLES amazes me, all while being completely fan-less and cool running. And BTW, it runs Lubuntu

This better still be around by the time I get money to buy it.

Ancient, unsupported version of Lubuntu with an even more ancient nVidia-provided kernel. The GPU driver is buggy as fuck too, with 2D rendering artifacts and random hangs and crashes when you start watching video or running multiple GLES apps.
You can update to a newer community-ported kernel and userspace based on some current distro, but then you lose any GPU acceleration and are left with a dumb framebuffer display. Which is no loss really, since nearly no Linux app uses GLES anyway.
I have one too.
It runs Arch so I get modern software.

I never suggested otherwise. It does make things difficult when GNU/Linux is the primary OS to be used on this hardware. So difficult that people are writing dirty hacks to get OpenGL working on hardware that doesn't support it.

All ARM devices are garbage
dolphin-emu.org/blog/2013/09/26/dolphin-emulator-and-opengl-drivers-hall-fameshame/

The Tegra boards aren't too bad. It makes a world of difference when your hardware supports OpenGL out of the gate.
youtube.com/watch?v=-68fDxQAFfk

Wrong. I have 16.04 on it, but 16.10 is available, and I wouldnt be surprised if 17.04 was available as well.

FYI, 16.04 is a LTS and is still fully supported

How loud will this gay thing be?

16.10 is the last one, they ditched AC100 afterwards.
Is the ARM port supported on the LTS schedule too?

Likely loud as fuck. It looks like a garden variety server cooler, and nobody cares about acoustics on those.

except you know... your operating system's process pool itself, with hundreds of jobs to dispatch that could do well running in parallel.

this isn't a fucking multi-core nintendo, it's an application processor with multi-user, multi-tasking operating systems running on them.

There are no specs on the fan, appears that Dell use them for their OEM shite.
That being said it looks like it's just a clipped in 120mm fan, surely you could replace it with something better no?

Yeah, the fan should be easy to replace. Not the radiator though, the mountings are completely non-standard (for a PC at least).

Well yeah the radiator block looks like it's made for POWER so good luck replacing that.
But the fan itself has standard holes, shouldn't be too hard, it's just the 4 wire that makes it somewhat unique.

An advanced video coding fan?

Probably a PWM fan.
Those have 4 pin connectors.

Look at the connector, it has 5 slots (one unconnected) and a clip. You'll need an adapter to connect a standard fan. Not hard to make tho.

As long as the slots are .1" apart, any standard fan plug should work.

the empty slot is weird, though.

PWM fans have 3 pins. This is a pwm fan that also senses rpm. Pins are 12v, gnd, pwm, rpm, for the faggots here who've never wired a circuit.

...assuming the pinout is the same.

...

...

It's literally fucking botnet. Not joking.

They run Linux-based open firmware on it.

Fans are plugged into the motherboard you worthless troll.

No, that's not unique for a CPU cooler fan. Every modern CPU cooler fan has that. That's PWM so your fan won't be blasting at full speed the whole time. The fourth pin allows the motherboard to set the speed lower when there's less load. That's all. And if you have a motherboard that has more four pin plugs you can hook up more PWM fans to cool your case and use software to control them instead of needing a fan controller (hardware) to control three pin fans.

There's a lot of people who get confused by fans that have a fourth pinhole and think it needs to be plugged into a four pin plug but that's not true. You could also plug it into a three pin but it will run at full speed if you do. You'd need a fan controller to turn it down then. Btw there are actually people who look everywhere to buy a four pin fan controller, don't be one of them.

Sage because offtopic.

The problem is that the plug has a fifth pinhole for some reason. It might not be compatible with standard connectors (different dimensions, different pinout, etc.).

No there isn't a fifth hole.

look closely at the fucking photo:

Ah. Yes that fan is for Dell shit only. According to pinoutguide the first and third pin are switched because you know Dell loves to release shit you can't use on other shit. I guess you can still try and see if it works, or maybe switch the pins yourself.

That isn't the same fan has the one in and
It's another type of connector that isn't a PWM connector.

see

I guess you are correct. The cooler pic: seems to have a normal 4-pin connector.

Gas the Dell tech war now.