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

[dkasak]

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.

Um ... you've copied / pasted your comment from another article. Seems more like lazy trolling than actual serious discussion. I don't think anyone at IBM will lose sleep over a Phoronix troll who "refuses" anything. Storm in a teacup ...

[elatllat]

"performance-per-Watt tests were not conducted" I'll just wait till they are.

[Michael]

"performance-per-Watt tests were not conducted" I'll just wait till they are.

You'll likely be waiting a while.

I have no ThunderX hardware locally and Cavium has refused through now to provide any hardware. I do have an Ampere eMAG pre-production server locally but they told me previously any power numbers would be inaccurate compared to their production servers.

Beyond that, the Arm and POWER9 servers don't expose any mechanisms I am aware of at this point to query just the CPU SoC/package power consumption under Linux so I would be monitoring the AC power draw just as I do when normally doing CPU testing, but here would then also be incorporating the various motherboard differences, etc, between platforms -- that is assuming I ever get those Arm platforms locally for being able to conduct such tests.

[elatllat]

...I would be monitoring the AC power draw...

Just as diffent app versions are used in OS comparison,
whole box price and power consumption of purchasable products is a fair real world comparison.

[DavidKL]

With virtually every single open-source application written with the underlying assumption

... that they will run atop black box hardware that may be actively working against the "owner"/"user" of the equipment...

I do fully understand, though, that it's much easier to accept a broken toy than the fact that more effort is required to obtain and operate the fixed one.

[mitcoes]

Thanks a lot.

But I also would like a lot price/performance benchmarks, even if they are not very accurate with some disclaimer, to not have one. And in this case scalability.

I cannot learn if 2 or 4 Power9 would scale to be better and cheaper - or not - than AMD newer server CPUs from these benchmarks.

Or you can make an "almost same price - including 2 (or x) years power supply costs if it matters - " multi-computer and or multi-CPU server configuration, to benchmark them.

[c117152]

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

It's possible. Intel implements the x86-64 ISA mostly as a decoder targeting a different "microcode" ISAs. It's highly likely they'll be able to shoe-horn the ISA in similar ways on anything Turing complete and nothing short of massive performance loses and security flaws could make the market ditch it.

[mattmatteh]

What is the phoronix-test-suite benchmark?

[wizard69]

Very impressive result for AMD.

In some ways shocking. I’m hoping this leads to significant sales and thus profits to pour into drivers and R&D.

[wizard69]

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.

There is no such assumption across the open source world. Why you would believe that baffles me.

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:

That will never happen on any single day. Honestly X86 distros frequently have to patch software to get it to run on their platforms.

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

as is the current situation with x64.

I don’t think you grasp the current situation.