IBM Power11 CPUs Launching In 2025 - Linux 6.13 Preps KVM Nested Guests For Power11

[coder]

POWER is used by tons of large enterprises with needs too big for x86 shitboxes from amd/intel but to small to justify the cost of a Z-arch mainframe. This is firmly in the territory of "if you need to ask, you can't afford it, but if you need it, nothing else will do the job."

POWER chips tend to come out only every 5 years or so, but typically leapfrog whatever amd/intel have by at least 3-4. It almost always is a full generation ahead on PCIe and either is a full generation ahead on DDR or has a faster bespoke memory system like OMI. Many POWER features and capabilities never find their way into everyday x86 server chips. eg. POWER has supported 4-way SMT (or even 8-way SMT) for several generations now, while AMD has only started to flirt with 4-way.

Wow, you're high as hell on copium with that shit. No, POWER stopped being competitive 5 years ago and all it's doing now is following the mainframe playbook of milking market niches who are mostly smoking the same dope as you.

OMI is dead. OpenCAPI folded and gave its IP to the CXL consortium. POWER had got the jump on PCIe 4.0 (beating AMD by about 9 months), but that's the last time it'll ever be ahead in anything. POWER10 only scaled up to 30 cores, at a time when AMD was already doing 64. Even though POWER can scale to 16 sockets, you'd much rather have fewer CPUs, each with more cores, and Intel can do 8-socket scalability now with 128 Hyperthreaded cores or 144 E-cores.

OpenPOWER was a good thing. But, as usual, when industry finally does the right thing about opening up, it's usually too late. It will not save POWER.

[NeoMorpheus]

LOL, the mobile problem was indeed solved, but by P.A. Semi. However, it came too late. When Jobs bought them, he had already charted his course with Arm.

https://vita.militaryembedded.com/2885-p-processor-ever-designed/

I'm pretty sure there were emulators, though. I saw Windows NT running on a DEC Alpha and I thought I also saw x86 software running on it under emulation.

It was a translation layer, called FX!32 and only worked on DEC Alpha, not on PowerPC.

https://en.m.wikipedia.org/wiki/FX!32

[ilgazcl]

Oh, c'mon! That wasn't Apple's fault. Intel had been steadily increasing TDPs since Skylake! The year Apple launched M1 was also the year of Intel's Rocket Lake. Their mobile CPU of the time was Tiger Lake, which was basically a higher-clocking version of Ice Lake and not terribly efficient, either.

It was always the case, even when the day Apple switched to Intel. I remember Mac Intel early adopters rushing to forums asking about the absurd temperature levels and whether it will "burn". The raw power of them were damn impressive for a G4 switcher, remember G5 never shipped for the mobile.

Things look like night&day today, fanless laptops, 20 hours battery life, everything X86 can't touch.

[ilgazcl]

As if Amiga the company wasn't dead.

If all were well in Amiga, financially but with the same idiotic management, they would ship a 68040 box rather than something like PA-RISC+3D as proposed by their engineers and die again. You probably know the Amiga 1000 (and 500 etc) aren't the original envisioned computers anyway.

[edwaleni]

No, those don't use POWER ISA.

IBM Power Systems use the POWER CPU. Since RHEL 8, Linux can run in little endian mode on POWER. AIX and i-Series run big endian.

IBM Z does not use Power. However, RHEL on Z/VM requires the big endian version.

[hiryu]

They're not ANYTHING based. Their CPUs are a from-scratch implementation of the OpenPOWER sepc.

They're also likely vaporware, but that's an entirely separate issue.

If Raptor says that's what they're going with... I suspect they're not vaporware.

I should have been more clear on what I meant by "POWER11 based". I didn't mean based on IBM's internal P11 design... But likely at least partially based on the P11 ISA and feature set updates. I'm sure there will be differences, but perhaps you could say they'll like be the same "generation" rather than Solid Silicon being a newer release that's stuck on P9 or P10 feature sets and capabilities.

[Vistaus]

I'm not sure what you're talking about, but the Amiga machines from the 1990's were all based on Motorola 68k.

Like I said, NEW Amiga's. I didn't talk about old Amiga's. Both the current X1000 and X5000 use PowerPC:

https://en.wikipedia.org/wiki/AmigaOne_X1000

https://wiki.amiga.org/index.php?title=X5000

And even the older AmigaOne 500 that I have uses PowerPC: https://acube-systems.biz/index.php?page=hardware&pid=7

PowerPC is also the only hardware supported by AmigaOS 4 (AmigaOS 3 for 68k is also supported still): https://www.amigaos.net/content/72/supported-hardware

[Vistaus]

Amigas have been using PowerPC for decades. 68k systems are the older Amigas.

Thank you! Finally someone that's knowledgable.

[Vistaus]

Atari and Commodore died before jumping to PowerPC.

AmigaOS is very much alive and has been using PowerPC on new hardware for years.

https://www.amigaos.net/content/72/supported-hardware

[Developer12]

Wow, you're high as hell on copium with that shit. No, POWER stopped being competitive 5 years ago and all it's doing now is following the mainframe playbook of milking market niches who are mostly smoking the same dope as you.

OMI is dead. OpenCAPI folded and gave its IP to the CXL consortium. POWER had got the jump on PCIe 4.0 (beating AMD by about 9 months), but that's the last time it'll ever be ahead in anything. POWER10 only scaled up to 30 cores, at a time when AMD was already doing 64. Even though POWER can scale to 16 sockets, you'd much rather have fewer CPUs, each with more cores, and Intel can do 8-socket scalability now with 128 Hyperthreaded cores or 144 E-cores.

OpenPOWER was a good thing. But, as usual, when industry finally does the right thing about opening up, it's usually too late. It will not save POWER.

You're making comparisons using POWER9 (2017) and POWER10 (2020). No shit they're behind. POWER11 comes out next year, meaning those generations of POWER are nearly EoL. Would you like me to compare them against an intel Kabylake server and see how they do? No?

OMI had higher bandwidth than any DDR technology that existed when POWER10 came out. Frankly, it's an embarrassment that DDR still uses a parallel interface when everything else (PCIe, SATA, USB, Ethernet, etc.) have switched to independent lanes of differential pairs. OMI was a massive step in the right direction. No shit it's dead, it's being replaced by CXL which wasn't a thing at the time and is typically built on PCIe with diff pairs.

You're damn right POWER9 shipped PCIe 4.0 before AMD, and it's very likely POWER11 will ship PCIe 6.0 next year. Any ideas when AMD well get around to that? No? Those POWER10 cores each do 8-way SMT btw, so you're talking about 240 POWER threads vs 256 from AMD. That narrows the gap significantly, and those POWER cores are being fed by memory bandwidth that AMD could only dream of.

Even though POWER can scale to 16 sockets, you'd much rather have fewer CPUs, each with more cores, and Intel can do 8-socket scalability now with 128 Hyperthreaded cores or 144 E-cores.

No, you don't. Not at the scale POWER machines operate at. That's joe blow consumer logic. That sheer number of sockets should have tipped you off.

The constraining factor on CPUs at this scale is power dissipation. One of the remarkable things about AMD's new Turin processors (announced last month) that they can hit an *all core* 5.0 GHz. Intel has never been able to do that. They have so many cores crammed onto one die and their cores are so inefficient that essentially every core needs to fall back the instant one or two P-cores start to turbo.

By spreading out the system across more sockets and making use of high-bandwidth OMI-style links between them (technically, it's a different fabric between sockets) you get more cores, more cache per core, more system memory, higher sustained clocks per core, AND can keep the heat per socket low enough to actually be physically possible with current process nodes and air cooling. The result is that current (soon to be replaced) POWER chips have been able to hit top speed across all cores on all sockets at the same time, something AMD only just caught up to.

(while also having a larger system with more cores and ram, really important for the workloads people buy POWER machines for which are too large to ever fit on an AMD box. When was the last time you saw an AMD server with 16 TB of ram in 2020?)