/talosii/ our mainboard

Attached: SASless with heatsinks.jpg (1920x1280, 582K)

Why should I go with this instead of a ARM ATX board for daily use?

You can actually control your firmware with OpenPOWER chips. Everything down to the DRAM training is opensource. You can even modify the OCC code (tiny on-chip core that regulates thermals)

Attached: architecture-comparison.jpg (749x452, 28K)

Here's a diagram of the different components on-chip for the POWER9 CPUs

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

>includes RISC-V in legend
>doesn't include it in chart

It's the darker region:
0 fast
1 open
0.5 efficient
0.5 inexpensive
1 owner controllable
1 secure

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

The SMT4 chiplets come in pairs because there is a chip variant for PowerVM (for AIX and IBM i) that has one SMT8 core in place of each pair of SMT4 cores.

The chips used with Talos II are Sforza modules of the Nimbus POWE9 chip.
Cumulus is the SMT8 chip.

Attached: sforza tray.jpg (1280x856, 255K)

Better picture

Attached: POWER.png (734x499, 293K)

I'll require citation for that chart please. Thanks.

Really cool. Hard to imagine a use case, though. All that stuff just for a server?

t. tinfoil pedo

No we aren't any better than them. We just have different reasons for overpaying for things.
They overpay because they like having devices that just work.
We overpay because we are all paranoid assholes.

Did you make that pic? A simple table would be much better.

What is the TPD per CPU?

It's from the initial crowdsupply for Talos I. Still fairly true to this day, although there was a moment where it looked like RISC-V vendor SiFive was going to be reliant on closed source initialization blobs.

Raptor presentation for coreboot, with their decision rationale:
youtube.com/watch?v=Bhwd5i6pGfg

Do those boards work on water?

4 core (16 thread) 90 W
8 core (32 thread) 160 W
18 core (72 thread) 190 W
22 core (88 thread) 190 W

from: raptorcs.com/TALOSII/

It's an IBM retention system, so any cooler mounting will have to be highly custom:
wiki.raptorcs.com/wiki/Talos_II/Mechanical_Specifications

I mean, IBM makes POWER9 systems that are liquid cooled, but those use a different chip module and socket. Pic related.

Attached: SUMMIT_NODE11.png (2048x1356, 1.89M)

Pic: 2U heatsink for Talos
You can sort of see how the retention system works here.

Attached: 2U heatsinks.jpg (1920x1280, 637K)

3U HeatSinkFan assembly for comparison

Attached: P9 heatsink.jpg (1600x1200, 446K)

There is some crazy detailed Processor Register documentation out there as well; looks like around 6000 pages total:
wiki.raptorcs.com/wiki/File:POWER9_Registers_vol1_version1.1_pub.pdf
wiki.raptorcs.com/wiki/File:POWER9_Registers_vol2_version1.2_pub.pdf
wiki.raptorcs.com/wiki/File:POWER9_Registers_vol3_version1.2_pub.pdf

What can I buy, TODAY?
Any laptop or SBC?

>Any laptop or SBC?
No. Mobile is a lost cause for open hardware at this point.

Don't tell me that, give me some hope. I just want something portable.

I mean, I'm sort of watching Purism's Librem 5; they at least sort of have their heart in the right place.

Raptor actually threw a bit of a fit when they heard that Purism was using proprietary DRAM training code, and using a loophole to still call it RYF compliant:
twitter.com/RaptorEng/status/1009163155694669824

That said, if you need a phone or a modern laptop, they might be the place to look. Their laptops are at least me_cleaned; which is better than most x86 vendors.

If you are willing to go with older machines, the FSF does list some Libreboot-ed Thinkpad vendors, or you could buy one and flash it yourself.

Ironically though, these are still reliant on a proprietary Embedded Controller. The Libreboot website even specifically tells you to use the non-free Lenovo BIOS to update it.
libreboot.org/docs/hardware/#ec-update-on-i945-x60-t60-and-gm45-x200-t400-t500-r400-w500

I'll check those, thanks.

QubesOS on Talos II/Power9 would be great imo

There was been some discussion about that on Twitter, but I also found this thread from earlier:
groups.google.com/forum/?nomobile=true#!msg/qubes-devel/IFLyyCUbLmQ/XRAyAt8_BQAJ

Very interesting. And great to see that Raptor themselves are interested. Hopefully something happens on that front soon(tm).
Thanks for the link.

Here's the thread I saw on Twitter,
twitter.com/justinrwlynn/status/999099825265115136

Notable posts:

Rapor
> Absolutely. Problem is, qubes only receives funding for development on backdoored/insecure x86 systems. They're not willing to work on other machines without payment. Perhaps it's time to do a Qubes-style OS built around Debian and targeting only blob-free architectures?

JSharp
> Well, there's been a heap of great work that's gone into qubes and, if it were possible, I'd use it. However, some of its core architectural assumptions don't work well with our OpenPOWER systems, so we'll have to scratch our own itch, so to speak.

Sounds like part of the problem is Qubes reliance on x86. Also at this time, KVM is better supported on Power than Xen.

JSharp's page on the Raptor wiki doesn't mention anything there yet, but I'll do some more digging.
wiki.raptorcs.com/wiki/User:JSharp

Why is Xen even needed? What does it offer over KVM? To me it just seems like a security hole. I get reports of new major CVEs in it every month or so.

I think there was a mention of making Qubes compatible with KVM and Xen; apparently at the start Xen was somehow a better fit for Qubes.

Quoting from the last post in s link:
>From upthread, sounds like Qubes on KVM on Power might be most realistic.
>Have there been any improvements to KVM since
>qubes-os.org/attachment/wiki/QubesArchitecture/arch-spec-0.3.pdf
>was written?
>Specifically, this conclusion:
>
>"We believe that the Xen hypervisor architecture better suits the needs of
>our project. Xen hypervisor is very small comparing to Linux kernel, which
>makes it substantially easier to audit for security problems. Xen allows
>to move most of the “world-facing” code out of Dom0, including the I/O
>emulator, networking code and many drivers, leaving very slim interface
>between other VMs and Dom0. Xenʼs support for driver domain is crucial in
>Qubes OS architecture.
>KVM relies on the Linux kernel to provide isolation, e.g. for the I/O
>emulator process, which we believe is not as secure as Xenʼs isolation
>based on virtualization enforced by thin hypervisor. KVM also doesnʼt
>support driver domains."
>
>Eight years later, what would need to be done in KVM to close the gap with
>Xen? All of the above?

If they are looking to change the kernel entirely they might as well use a different OS, then focus on the hypervisor. Is more logical to use one of the microkernels out there and apply an hypervisor like they want instead of reinventing the wheel yet again.

I know it's not going to happen, but it would be hilarious if someone decided to use the Hurd microkernel for that.

The only reason the Hurd is not used is because the developers didn't want to make drivers for basic things, not because the OS itself is bad, is actually pretty good.

so to sum up, we won't be getting Qubes on the Talos because monolithic kernels a shit?

Attached: OS-Structure.png (1280x683, 56K)

Will this motherboard protect me from (((them)))?
I really wish there was a complete diy mobo schematic eveneracion if it mentions soldering everything.

Meh, they can keep on trying with Xen. Not sure how far they'll go with that strategy though.

It will protect you from everyone except yourself, and everything except a rubber hose.

>Will this motherboard protect me from (((them)))?
yep. or at least way better than anything else on the market.

Are there any talos mainboards that support the SPARC architecture?
Just wondering, I think SPARC is more expensive than POWER anyway.

JSharp's comments on Twitter seem to suggest that even with Xen ported to power, there are some x86-centric design choices in Qubes:
>> Well, there's been a heap of great work that's gone into qubes and, if it were possible, I'd use it. However, some of its core architectural assumptions don't work well with our OpenPOWER systems, so we'll have to scratch our own itch, so to speak.

Um. No.
Talos I = POWER8
Talos II = POWER9

I know there was some old SPARC design that was opened up, but the whole point of Talos is to use modern hardware, and modern SPARC is controlled by Oracle and Fujitsu IIRC. I don't see either of them making an OpenSPARCFoundation any time soon.

In all honesty somebody should take over the Hurd from the hands of those autists, is a legit good OS that is suffering from the autists not wanting to do basic drivers and only "improving the design". Is a good microkernel architecture and last time I heard you can use seL4.

No. Talos only does POWER9.
Also I don't know if everything is really open in SPARC. I know they have OpenFirmware or OpenBoot or whatever back in the Sun days, but now that Oracle and Fujitsu run the show, idk if they would've thrown in any hidden botnets into newer implementations

I was under the impression they were still focused around open source software and hardware. I'll look into it.
Thanks for the answers.

well there's this: sparc.org/
but Oracle can be rather kikeish at times and I know nothing about Fujitsu, so that's why I'm unsure.

Hurd is dead. Accept it. People keep claiming it's good or something, but I see nothing special about it other than that it would have been the OS of choice had Linux not gotten big. But that's a different timeline. And no, you can't use seL4 with it (they were planning on it, but it's never gonna get done). It's stuck with mach and always will be.
Consider looking at the Genode project or Redox. Those are microkernel operating systems with some kind of a future.

Why the fuck is there no open source x86 motherboard? Patents preventing it? I just don't understand why no one makes one.

That's definitely true. And not to be a hipster or anything but I think the more esoteric an architecture is, the more secure it will be because fewer people know how to use it, generally speaking.
>x86
>CISC
>ever

Attached: comeon.jpg (222x200, 6K)

Why would that matter? How would the motherboard give a shit about what instruction set the CPU has, it just connects everything to it.

The only reason it doesn't come out of the dark is thanks to lazy devs but it can do process isolation and capabilities like Genode. Genode is definitely a good choice too but not my point.

Oh for fucks sake, you want pedos? Fine look at the blue checkmarks on twitter jesus fucking christ.
It's not like you need encryption for that shit you just need to not be suspected.
Nobody is actively searching every computer and every post and every DM this is why there are rumors of child porn sharing through twitter DM. All you need is a small active private circle(look at the fucking bulbapedia page)
What we want is to mess with shit that intel and AMD and every other company does not want us to mess with. Gtfo with your shit.

Find a way to make an intel CPU to work without all the crap with blobs.(most famous examplr intelME)

I was thinking of OpenSPARC, which only involved the UltraSPARC/OpenSPARC T1 and OpenSPARC T2

en.wikipedia.org/wiki/OpenSPARC
oracle.com/technetwork/systems/opensparc/index.html

>well there's this: sparc.org/
That is interesting, I just assumed it was an unwanted cross-licensing operation between Fujitsu and Oracle, like what exists between Intel and AMD.

> Executive Level Members:
> Oracle
> Fujitsu
>
> Affiliate Level Members:
> Cobham Gaisler AB
> Kilpatrick Townsend & Stockton LLP
> Zhuhai Orbita Aerospace Science & Technology Co., Ltd.

Huh, what is Kilpatrick Townsend doing as a SPARC member?

Um. WTF?

I don't mean the firmware blobs for the CPU, I just mean the actual motherboard. The BIOS, UEFI, the circuit schematics. I understand that AMD and intel have a duopoly on x86 and will never open source anything on their CPUs, but what's stopping someone from creating a motherboard for these CPUs that's open source?

Can I run HarveyOS on it?

Attached: HarveyOS.jpg (281x268, 13K)

I'm a mechanical engineer casually passing by Jow Forums and this STILL gets me hard. Wanna run CFDs on this bad boy until it screams for mercy.

Attached: Cumming4.jpg (225x225, 22K)

>expensive = overpriced

I think pretty much all x86 chips still need a chipset, so you end up buy additional chips from AMD or Intel.

Regardless, a lot of hardware init code is locked down, and some init is done by ME or ASP/PSP.

Also, if you're not American Megatrends, you basically get treated second class just in terms of getting those init blobs. AMI gets early access, while everyone else has to wait until the FSP is released.

Look at the trouble they have porting coreboot to new boards (Raptor actually does some of that kind of work on contract)

Sadly, for the full effect of that, I'm guessing you need the datacentre Tesla drivers. Noveau doesn't have NVLink support as far as I know.

Slower hardware without any additional benefit. You are retarded. There is your you.

Tbh idk if there is ANY fluid dynamics software for the common college graduate that even runs on a GPU, let alone a server rack and the costs that entails.

Most just pay subscription fees and run their test on somebody else's mainframe.

It's a bit of a pity that so many of the PCIe slots are unusable in single-socket configuration. No, I don't have a good idea how they'd fix that, but it's still a bit of a pity.

Are those your own pictures, user? Do you own one?

>You can even modify the OCC code (tiny on-chip core that regulates thermals)
I don't understand why Intel/AMD don't allow this. It would an absolute goldmine for overclockers especially, but also for all kinds of enthusiasts and hackers. What do they have to lose? I can't imagine that code contains any sauce so special that the other party would just runaway win the competitive race with it (especially given architectural differences making it impossible to apply directly anyway). It's so weird.

Qubes strikes me as an incredibly ugly an inelegant idea. There should be no need to run several instances of an operating system just to properly isolate application processes from each other. If that can't be done on a single OS image, then there's something wrong with that OS.

Attached: a.jpg (933x720, 55K)

Look in the bottom right corner and ye shall see.

I wish I did though.

>>x86
>>CISC
>>ever
x86 still CISC. The chips they run on are RISC since P6 though.

Depends what binary firmware blobs all the controllers on the motherboard use. They run on the architecture of the host CPU.

>several grand for a computer that even a few year old Intel can beat
>Talostoddlers will defend this

This was the idea behind containers, and it led to a bunch of new and interesting security problems.

Oh. That means they have articles on Talos II that I haven't read though, so that's nice.
>I wish I did though.
I'm very strongly considering one for my home server. The main reason I'm holding off isn't even the price, but worries about the availability of spare parts in case something breaks. With my current x86 system, even if it breaks down completely, I can just use spare parts that I have at home or even go buy the cheapest stuff available and get it the same day, and just plug in the drives and have it work again. If a Talos system breaks, I'm afraid I might be without everything for possibly weeks.

That's what I was making fun of.
One of the biggest reasons I am against x86, despite using it in all of my computers so far, is that it's needlessly complex. On top of having few security issues, RISC and MIPS are usually open source and are much more simple than CISC.

I'm not too worried about Raptor; they've been around for a while and survived both MiniFree refusing to pay them for contract work, and the failed Talos I crowdfund campaign.

As far as the hardware itself, the slots are standard PCIe, and the memory is just registered DDR4 ECC RAM.
wiki.raptorcs.com/wiki/Hardware_Compatibility_List

They don't have the budget for exotic custom ICs on the mainboard, so my only concern would be the chips. As far as I know, the these are the systems which use the three different Nimbus (SMT4) POWER9 modules:
Sforza - Talos II
Monza - AC922
LaGrange - Zaius/Barreleye

Although I did come across this tidbit from the OpenPOWER Summit:
>Gigabyte showed its OpenPOWER server based on P9 Sforza
ibm.com/developerworks/community/blogs/572f1638-121d-4788-8bbb-c4529577ba7d/entry/March_6_2018_at_10_54_54_AM
so obviously Sforza isn't custom to Raptor.
In a worst case scenario, maybe you could buy Sforza chips direct from IBM?

Found this on Twitter, I think this was some sort of video encoding (VP9 maybe?) that someone started to optimise to better use the VSX SIMD instructions on POWER9

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

You answered you own question. They don't want consumers having any more control than they already do

Glad to see people shilling Talos II finally, this is the real deal. If it's out of your price range then just grab a libreboot compatible system, or at least coreboot + me_cleaner.

This. Anything mobile is botnet. Librem 5 would be nice, but I'm not spending $600 on a tablet with botnet + killswitch. Just buy a burner phone for $20 and don't keep anything on it.

>They don't want consumers having any more control than they already do
Yeah, but that's exactly what I don't get. What do they even stand to lose from it? I get that Intel wouldn't want people to go tweaking their carefully (((binned))) SKUs, but for K models, surely?

Brainlet here, what would this system actually be used for, what applications would it suit?

How is this better than Intel/AMD?

I'm not worried about Raptor going out of business or anything, it's just that I can't go to a local store and get a replacement the same day.

>Brainlet here, what would this system actually be used for, what applications would it suit?
Pretty much anything that doesn't specifically require x86. Pretty much the whole of Debian is ported to it, which covers quite a lot.
>How is this better than Intel/AMD?
Primarily by being completely open. No IME, no hidden NSA backdoors, no SMM or other ring-minus-whatever modes. As a strict upgrade, it also has PCI-E 4.0.
Also, the higher core-count chips are actually pretty reasonably priced, at $2,625 for a 22-core chip, or $1,425 for 18 cores. Not a totally game-changing price-drop compared to Xeon, but pretty good nonetheless.
Another point could be that the POWER ISA is open for alternative implementations, so in case it takes off, it's not going to be a duopoly forever, which x86 will definitely be.

From what it sounds like it would be absolute heresy to run any Microsoft shit on it. I'm definitely interested.

Great. Where can I buy a blob free graphics card with HBR2 DP?

If you're talking about phones, then wait for Librem5 and see what they'll end up with, best option so far.
As for laptops, either you get an old ThinkPad and Libreboot it - but you'll have to live with old hardware + x86 + proprietary EC.
Or get an Asus C201 Chromebook and Libreboot that, still old-ish hardware, but not as old, plus it is an ARM-based machine and the EC is free software (this is what I have).

>absolute heresy
What Microsoft shit would you run on it even if you wanted to? NT 4.0 was the last Windows version that supported Power.