Benchmarks Of OpenMandriva's AMD Zen Optimized Linux Distribution Against Ubuntu, openSUSE, Clear Linux

[gregzeng]

Ubuntu tested was with the GNOME interface? MATE is my preferred flavor, but I would guess that the main Ubuntu family should test the same results as each other. Like many readers here, we miss the GIMP test results for Ubuntu.
This series of tests reflect the same results as previous Phoronix results. Clear Linux was designed to show the best results for the Intel CPU. However it also shows the best result also for the AMD CPU.
Almost missed is the summary to these tests:
> " ... tied for first on this AMD Ryzen Threadripper 2950X system was Ubuntu 19.04 and Clear Linux."
This equality with Clear Linux is unexpected. Does the Ubuntu family value benchmark results over other factors? If OpenMandriva values power savings, to prolong the battery life in portable computers, do the Ubuntu family perform poorly in battery life?
It would be so nice if Linux operating systems could have standardised battery-life benchmarks. Android systems have these, so when Linux is ready for the mass market, these benchmarks should be available.

[ms178]

Tumbleweed is now updated from GCC 8 to GCC 9. However LTO is not on yet. It will hopefully do LTO soon - as of today 14 packages needs fixing (mostly because of symbol versioning issues).

Great, so more improvements are on their way. I thought you were already shipping with LTO enabled where the packages work already. I am very much looking forward especially for a PGO+LTO build of Chromium and LTO Mesa.

[hubicka]

Maybe LLVM/Clang does need more tuning for Zen? As a Sony employee does most of the work on this front, expect more Zen tuning with the new PS5. And my takeaway is that Clang still performs a bit worse than GCC at least in the benchmarks shown here.

I was surprised to see that Tumbleweed got better, their recent LTO effort seems to pay off.

Tumbleweed is now updated from GCC 8 to GCC 9. However LTO is not on yet. It will hopefully do LTO soon - as of today 14 packages needs fixing (mostly because of symbol versioning issues).

[wizard69]

Hey Micheal; what about Fedora? Many of us are in the Fedora camp and would find such comparison interesting.

As for this distro it would would be interesting to see what is the root cause of its performance issues. Is it CPUFreq? More likely it is a combination of things but with both versions performing so badly compared to other distros it would be nice to see what is the major problem.

The GIMP results seem to indicate significant build issues beyond any impact governor settings may have. In fact the delta here was actually a big surprise.

[wizard69]

Actually the real truth is that Zen performs better when using Broadwell optimizations.

So? I mean really that doesn’t imply anything special.

ENTIRELY speculation....
I honestly believe Intel donated a Broadwell based architecture to AMD shortly after AMD launched Bulldozer derived architectures.

I can’t see this happening under any circumstance.

AMD then rebased it and redeveloped it into what Zen is. I think Intel realizes they need AMD to continue to exist at some level, either due to contractual obligations or due to legal obligations, but in any case the end results are the same, Zen is very similar to Broadwell and it performs better with Broadwell compiler tuning.

Well it has been awhile since looking at the chip architectures but last I remember there is very little similarity here. Frankly I see a lot of innovation that went into ZEN.

[duby229]

This idea/rumor was floating around the Gentoo camp for a while. I tend to find this skeptical with Intel's knee jerk reactions to Zen. Zen is the work of Jim Keller the lead architect behind the Athlon64 chip. He left AMD in 1999 and returned to create Zen.

Intel has publicly bashed the design but.. if you think of the limits we face with moore's law it may be rather smart to extend CPU performance sideways by adding core count and threading to processes. In a way AMD did the only thing they could and it was a win.

I can't understand how Jim Keller gets so much credit for K8, thye real truth is Fred Weber designed K7 and was the lead architect for K8. Jim keller was there, but Fred Weber was the dude....

EDIT: Please don't forget about Nexgen's K6, which is -exactly- what K7 derived from. That's all -Fred Weber-.... No doubt Jim Kellerdid something to improve K8, but he didn't design it, he already had a complete architecture to work with. No doubt he also had a complete architecture to work with when he started Zen....

[k1e0x]

Actually the real truth is that Zen performs better when using Broadwell optimizations.

ENTIRELY speculation....
I honestly believe Intel donated a Broadwell based architecture to AMD shortly after AMD launched Bulldozer derived architectures. AMD then rebased it and redeveloped it into what Zen is. I think Intel realizes they need AMD to continue to exist at some level, either due to contractual obligations or due to legal obligations, but in any case the end results are the same, Zen is very similar to Broadwell and it performs better with Broadwell compiler tuning.

This idea/rumor was floating around the Gentoo camp for a while. I tend to find this skeptical with Intel's knee jerk reactions to Zen. Zen is the work of Jim Keller the lead architect behind the Athlon64 chip. He left AMD in 1999 and returned to create Zen.

Intel has publicly bashed the design but.. if you think of the limits we face with moore's law it may be rather smart to extend CPU performance sideways by adding core count and threading to processes. In a way AMD did the only thing they could and it was a win.

[duby229]

Actually the real truth is that Zen performs better when using Broadwell optimizations.

ENTIRELY speculation....
I honestly believe Intel donated a Broadwell based architecture to AMD shortly after AMD launched Bulldozer derived architectures. AMD then rebased it and redeveloped it into what Zen is. I think Intel realizes they need AMD to continue to exist at some level, either due to contractual obligations or due to legal obligations, but in any case the end results are the same, Zen is very similar to Broadwell and it performs better with Broadwell compiler tuning.

[betam4x]

I was under the understanding that not very much (depending on the version) of znver1 was different from generic x86_64 and that it was mostly a future stub flag. Clearly a lot of work and understanding of optimizing for ZEN still can be done.

"AMD how you doing over there... would you like to help make your processors go fast?"

Some optimization work has been done, but not a lot. It's pointless anyway. AMD has an entirely new architecture for Zen 2. Zen 1 runs 'well enough' under generic settings. Unlike Intel, AMD has entire architectural changes from here forward, so they can't afford to spend the time/resources on minor optimizations for each CPU release. If EPYC "Rome" sees the adoption rate AMD proclaims, we may see Zen2 optimizations (due to server workloads). Otherwise don't expect much.

[artivision]

Is the Arch Clear Kernel enough to get extra performance in-game or Mesa needs to be also a clear build?