Core i9 7980XE vs. Threadripper 2990WX - The Pre/Post 2018 Linux Kernel Performance

[phoronix]

Phoronix: Core i9 7980XE vs. Threadripper 2990WX - The Pre/Post 2018 Linux Kernel Performance

Earlier this week I posted some benchmarks looking at the Linux kernel performance from the start to end of 2018 using an Intel Core i9 7980XE system. Here is the second part of that testing in looking at the same Linux 4.14 vs. 4.20 kernel benchmarking while putting the i9-7980XE performance side-by-side against the AMD Ryzen Threadripper 2990WX for seeing how its performance was impacted under the same kernel tests.

http://www.phoronix.com/vr.php?view=27274

[schmidtbag]

Interesting how inconsistent the results were. Though, it does seem that Intel overall took a bigger performance hit (proportionately). Strange how often AMD seems to perform worse with the mitigations turned off.

[microcode]

Where's the hardware config chart? Did both systems use the same memory? Same cooler performance?

[caligula]

So the AMD CPU is around 10% cheaper? What about the whole platform (mobo, certified memory) ?

[ms178]

I also wonder why the AMD CPU is performing worse with mitigations disabled than enabled in some benchmarks. I would expect consistently better numbers with the mitigations disabled. Hackbench behaves very odd as well, there is quite a big difference between 4,14 - 4.20 with both CPUs. Could this be a Kernel performance regression?

[schmidtbag]

Where's the hardware config chart? Did both systems use the same memory? Same cooler performance?

Michael isn't an idiot. I'm sure both platforms were cooled plenty adequately without thermal throttling.
Knowing memory speeds would be nice, but most of the oddities in these results are clearly due to software.

[L_A_G]

That's a lot of server benchmarks (databases, PHP, Node.js, etc.) to be running on two components that are pretty explicitly sold as workstation parts. I know servers have to do a lot of context switching and are thus the parts hit the hardest by Spectre and Meltdown, but you'd think that it would be worth the effort to use actual server hardware when doing server benchmarks.

[pete910]

I also wonder why the AMD CPU is performing worse with mitigations disabled than enabled in some benchmarks. I would expect consistently better numbers with the mitigations disabled. Hackbench behaves very odd as well, there is quite a big difference between 4,14 - 4.20 with both CPUs. Could this be a Kernel performance regression?

Might be not updated BIOS/AGESA , Can't see it mentioned whether they where both updated or not

[tildearrow]

Typo:

In some of the Java tests, the Core i9 7980XE performance slowed down e nough over

[milkylainen]

This looks like other slowdowns in the kernel beside those controlled by flags.
If you look at all patches made against spectre/meltdown I suspect you get a better answer than just the generic disables.