POWER9 & ARM Performance Against Intel Xeon Cascadelake + AMD EPYC Rome

[phoronix]

Phoronix: POWER9 & ARM Performance Against Intel Xeon Cascadelake + AMD EPYC Rome

For those wondering how ARM and IBM POWER hardware stack up against AMD's new EPYC "Rome" processors and that of Intel's existing Xeon "Cascade Lake" processors, here is a round of tests from the POWER9 Talos II, Ampere eMAG, and Cavium ThunderX in looking at the cross-architecture Linux CPU performance currently in the server space.

http://www.phoronix.com/vr.php?view=28180

[dispat0r]

Very impressive result for AMD.

[reavertm]

In order to normalise results somehow, due to lack of cost and power baselines, perhaps performance per core and/or thread could be provided? At least for multithread tests. This would require extending PTS I suppose to retrieve core/thread count from the system to make it generic benchmark statistics feature.

[marty1885]

In order to normalise results somehow, due to lack of cost and power baselines, perhaps performance per core and/or thread could be provided? At least for multithread tests. This would require extending PTS I suppose to retrieve core/thread count from the system to make it generic benchmark statistics feature.

At the first page of the post. Quote:
Power efficiency / performance-per-Watt tests were not conducted due to the remote ARM testing. Likewise, there isn't performance-per-dollar metrics due to many variables at play when it comes to factoring in the ARM and POWER9 costs.

[Sonadow]

With virtually every single open-source application written with the underlying assumption that they will be compiled under x64, I consider POWER to be nothing more than just IBM's personal toy.

I'm not going to give POWER a second look until we reach the day where every application that makes up the Linux desktop can be built as-is and unpatched directly from upstream sources by an end-user with just:

• ./configure, make, make install,
• cmake make make install or
• meson ninja ninja install

as is the current situation with x64.

[Panda_Wrist]

With virtually every single open-source application written with the underlying assumption that they will be compiled under x64, I consider POWER to be nothing more than just IBM's personal toy.

I'm not going to give POWER a second look until we reach the day where every application that makes up the Linux desktop can be built as-is and unpatched directly from upstream sources by an end-user with just:

• ./configure, make, make install,
• cmake make make install or
• meson ninja ninja install

as is the current situation with x64.

this kind of thinking is why x86_64 will always be. other platforms can't compete. power9 has a chance but since most devs use x86_64, it doesn't have a chance. RISC-V has a chance but i doubt it will get past arm levels.

[RussianNeuroMancer]

Benchmarks of 128 core Huawei TaiShan 2280 V2 is badly needed, but it's hard to buy and impossible to rent (60 vCPU is available for rent at most).

[ldesnogu]

this kind of thinking is why x86_64 will always be. other platforms can't compete.

Oh really *always*? So we will endure x86-64 til the end of time?

[concatime]

With virtually every single open-source application written with the underlying assumption that they will be compiled under x64, I consider POWER to be nothing more than just IBM's personal toy.

I'm not going to give POWER a second look until we reach the day where every application that makes up the Linux desktop can be built as-is and unpatched directly from upstream sources by an end-user with just:

• ./configure, make, make install,
• cmake make make install or
• meson ninja ninja install

as is the current situation with x64.

So the issue resides on the developers, which do NOT consider other platforms than x86_64. And having generic code (not optimized assembly) depending on platform, beside glibc and few others, is worrisome.

[reavertm]

At the first page of the post. Quote:
Power efficiency / performance-per-Watt tests were not conducted due to the remote ARM testing. Likewise, there isn't performance-per-dollar metrics due to many variables at play when it comes to factoring in the ARM and POWER9 costs.

Thank you, Captain "I have not read the post before replying". I specifically said, since those baselines you mentioned cannot be provided (due to reasons you mentioned), I proposed for Michael to provide different baseline, that can be provided - number of cores/thread.