It Looks Like Raptor Is Gearing Up To Release A New Open-Source POWER System

[WolfpackN64]

QEMU isn't difficult at all, and there is libvirt to make it even easier. Mind you, if you want to run x86 binaries (ie proprietary games) you really need an x86 CPU. QEMU emulation is excellent, but you only get near metal performance through hardware virtualisation - which means your virtualised hardware is the same architecture as your actual hardware. Emulating x86 on POWER would probably be acceptable for compatibility with the odd legacy application, but totally inadequate for AAA gaming. On the other hand, if POWER actually starts to become popular with Linux users, maybe Valve will begin to compile their binaries for POWER and really make things interesting...

Well, I'm willing to take a performance hit (currently on an Athlon X4 860K). But if it's too great I might indeed be better of with an x86 CPU. Shame though.

[ssokolow]

and I have no idea how it would deal with clients, like Steam, that install more programs, all x86 binaries. If this were easy enough to set up, I would seriously consider an affordable-ish POWER9 system.

The Debian wiki has instructions on how to use the kernel's binfmt_misc support to set up transparent qemu-user indirection so that binaries for non-native processor arches function the same as native ones... just slower.

As for performance, you can get an idea for the overhead by setting that up and then installing both a native and non-native build of the same piece of open-source software on your current hardware.

[kpedersen]

Only problem is that QEMU User Mode looks difficult to set up and I have no idea how it would deal with clients, like Steam, that install more programs, all x86 binaries. If this were easy enough to set up, I would seriously consider an affordable-ish POWER9 system.

In the end this is actually quite simple to achieve for incredible functionality: Check out https://forums.freebsd.org/threads/a...64-host.65130/

In the end, you instruct the kernel (via binmiscctl on FreeBSD, Linux will have similar) to execute a certain file format using qemu-user-static (in a similar way you can get the kernel to use wine to launch .exe implicitly rather than manually launching them with wine "wine test.exe" for example)

Then you just need an appropriate architecture chroot or Jail (Debian's debootstrap or FreeBSD Jails make this easy)

And then you pretty much just run the binary as normal from within the chroot and it works. It feels too good to be true. It is also oddly fast too because it only needs to dynamically translate the program and not the entire kernel and hardware (unlike standard qemu). Most executables do little more than automating the kernel and hardware anyway so actually have relatively little actual instructions to process. It is easily enough to run Quake III on any modern hardware. I am quite sure Raptor hardware will suffice.

For testing I have a FreeBSD system where if I SSH in on the following ports, I get different architectures:

port 22 it is native amd64
port 23 it is armv6
port 24 it is aarch (arm64)
port 25 it is i386
port 26 it is emulated amd64

I couldn't get it working for emulating sparc64 or POWER though. aarch64 and amd64 are the best supported.

[s_j_newbury]

In the end this is actually quite simple to achieve for incredible functionality: Check out https://forums.freebsd.org/threads/a...64-host.65130/

In the end, you instruct the kernel (via binmiscctl on FreeBSD, Linux will have similar) to execute a certain file format using qemu-user-static (in a similar way you can get the kernel to use wine to launch .exe implicitly rather than manually launching them with wine "wine test.exe" for example)

Then you just need an appropriate architecture chroot or Jail (Debian's debootstrap or FreeBSD Jails make this easy)

And then you pretty much just run the binary as normal from within the chroot and it works. It feels too good to be true. It is also oddly fast too because it only needs to dynamically translate the program and not the entire kernel and hardware (unlike standard qemu). Most executables do little more than automating the kernel and hardware anyway so actually have relatively little actual instructions to process. It is easily enough to run Quake III on any modern hardware. I am quite sure Raptor hardware will suffice.

Maybe Michael can run some benchmarks? I'd love to see how it performs native vs x86 emulation.

[nomadewolf]

Which part of this is "open" ?

The firmware. Which makes it much harder for you to be spied on.

[s_j_newbury]

I'm going to see if I can hack Gentoo to support debian style multiarch libdir and implement qemu-user based (foreign) multilib support. This could be fun...

[madscientist159]

Isn't POWER is in a pickle now that Globalfoundries has abandoned its 7nm node?

No. POWER is perfectly fine. I cannot comment further beyond noting that per public news reports, there are at least two other 7nm foundries available. AMD already publicly moved from Global Foundries to TSMC. Shutting down a process node is not something that happens in a vacuum or broadsides anyone ( other than possibly Intel )these days.

madscientist159
i have some questions on this new product i hope you could reply
a) will this product be an atx/microatx/miniitx?
b) do we have the option to buy just the motherboard + cpu + ram because for people outside USA your product are more expensive because of taxes
c) the cpu do have an integrated video card or do we need an external one?
d) will be capable to handle virtual machine with virtualization technology and or iommu?
e) what about spectre meltdown and foreshadow do they are mitigated at hardware level or software mitigation are needed
f) is possible to use a custom fan or heatsink for the cpu?
g) ethernet and sata chipset do still need firmware blobs to run or do you freed them?

a.) No comment. Think small.
b.) Of course -- we've done this from day one and will keep doing so. This is something that made x86 easy to get and modify in the past, and is now a major selling point for our POWER hardware as well.
c.) No POWER9 CPUs have a integrated video card. Most offerings have a basic 2D VGA output via the BMC, but if you are looking to run Blender, FreeCAD, games, you will want an external 3D card.
d.) Of course! Just like Talos II / Talos II Lite can today.
e.) These are largely mitigated by disabling vulnerable hardware before the CPU cores even come out of reset. Meltdown has been patched in kernel for nearly a year now with a simple L1 flush on return from kernel space -- this is actually good security practice anyway and carries very little performance impact.
f.) We provide mechanical specifications, so all that is remaining is someone to manufacture a compatible heatsink / fan. We support such efforts (unofficially) where we can.
g.) SATA needs no blobs controller side (leaving aside disk firmware, which is not technically part of Talos itself). Ethernet is still being freed, see https://raptorcs.com/TALOSII/nic_fw_contest.php

QEMU isn't difficult at all, and there is libvirt to make it even easier. Mind you, if you want to run x86 binaries (ie proprietary games) you really need an x86 CPU. QEMU emulation is excellent, but you only get near metal performance through hardware virtualisation - which means your virtualised hardware is the same architecture as your actual hardware. Emulating x86 on POWER would probably be acceptable for compatibility with the odd legacy application, but totally inadequate for AAA gaming. On the other hand, if POWER actually starts to become popular with Linux users, maybe Valve will begin to compile their binaries for POWER and really make things interesting...

Have you seen Unreal Tournament 4 Alpha running on POWER natively? We can usually port a game of that complexity in a matter of a week or two (multiple separate vector engines including the various PhysX engines), so something like a simple OpenGL game or even the Steam Client should be a matter of days. Just a matter of getting Valve's attention for the most part. If you want to play UT4 on POWER, check out the instructions at https://wiki.raptorcs.com/wiki/Softw...ealTournament4

As far as QEMU goes, we tried to run UT2004 (x86-only) using the QEMU user support. The main sticking point is that QEMU's TCG does a terrible job of translating vector guest instructions to vector host instructions, so the vector units (think SSE on x86) are basically idle on POWER -- QEMU literally translates SSE vector instructions to host scalar instructions. Even so, UT2004 is almost sort of playable emulated, and older games (especially those with the option to turn off SSE) would likely work fine:

https://wiki.raptorcs.com/wiki/File:...64_demo_nr.mp4

[freespirit]

madscientist159
thank you very much for the replies this project seems what i was waiting for, i have some other question if you don't mind
a) for the storage will be avalaible a pcie m.2 socket for faster ssd or just the sata 3?
b) about the graphics, for fullhd video the integrated video is enough? integrated video output are hdmi and/or DP or vga only like talos? i really hope will have hdmi and dp allowing to use latest monitor and also multimonitor setup
c) a termal paste is required or suggested for the cpu? or do you provide a custom solution? i'm asking this because i saw a video about talos 2 and i saw no termal paste
d) normal PSU and their cable are ok to use for the system, or some other custom solution is needed?
e) what's the correct procedure to check and update system firmware and bios?
f) about the eth contest do you know if someone are working on it, do you have some info you can share about progress?
g) what about ram? will be possible to use the non-ecc cheapest ram?

[vegabook]

Why does anybody care about POWER architecture when IBM has sold all its foundries and its only partnership, Globalfoundries, has abandoned the leading edge?

Somebody please explain to me how this architecture is not a dead end. I'm all for competition, but that is going to come from AMD.

All this intricate comment detail.... it's all irrelevant if the backing superpower of the architecture has been neutered for the future.

[freespirit]

Why does anybody care about POWER architecture when IBM has sold all its foundries and its only partnership, Globalfoundries, has abandoned the leading edge?

Somebody please explain to me how this architecture is not a dead end. I'm all for competition, but that is going to come from AMD.

All this intricate comment detail.... it's all irrelevant if the backing superpower of the architecture has been neutered for the future.

Globalfoundries has abbandoned the 7nm, i see no disaster on it, if in the future IBM will need that technology could make an agreement with another partner
If you are asking why people care about POWER, i can give you my 2cent, it's because of x86 is full of backdoor and unknow instruction, it's time to drop technology who care only about their creator and start to support company and technology who care about be open and user friendly

the only cons i see on power arch, is the ibm fee, they should start give it for free, like risc-v or they will loose the competition, linux and bsd now are everywhere (android, embedded component) because they are free of charge, windows and osx are just on pc because of marketing, otherwise they will not exist anymore, time are changing, if 15 years ago the high % was on windows because computer=desktop/laptop and cpu=x86=intel/amd now computer=mobile>>laptop/desktop so android that mean linux and arm. I don't know what will happen in the next 15 years, but i'm pretty sure something like power or risc-v will gain value