AMD Threadripper 2990WX Linux Benchmarks: The 32-Core / 64-Thread Beast

[chithanh]

I had previously played around with switching this setting between madvise and always (taking a page from openSuSe Tumbleweed's book here), but didn't notice an improvement (or degradation) in the observed behavior. I tried disabling it completely but also didn't notice any difference; do I need to reboot my system to properly test the effects of that change, or is it only relevant for new processes?

My understanding was "took effect immediately" but I don't think I have ever confirmed that. If I can find a bit of free time will try to pick through the kernel code and find out for sure.

FWIW, this is what Red Hat documentation has to say:

To disable THP at run time
Run the following commands to disable THP on-the-fly:

  
# echo never > /sys/kernel/mm/redhat_transparent_hugepage/enabled
# echo never > /sys/kernel/mm/redhat_transparent_hugepage/defrag

• NOTE: Running the above commands will stop only creation and usage of the new THP. The THP which were created and used at the moment the above commands were run would not be disassembled into the regular memory pages. To get rid of THP completely the system should be rebooted with THP disabled at boot time.

TR 2990WX destroyed the competition in all benchmarks here. Meanwhile, over at Anandtech they show a completely different picture.

That is because Anandtech benchmarked on Windows. Which royally sucks at scaling. Quoting myself from the 2990WX Linux vs. Windows thread:

I mean, look at this:

and compare to what Phoronix found:

[juanrga]

The Core i9 7980XE continues to be left in the dust compared to the Threadripper 2990WX, but the "Core i9 9900X" is rumored to be coming soon... It will be very interesting to see how Intel responds to the success of AMD's Threadripper 2.

What is coming next month is the i9 9990K and the rest of the 9th series for mainstream desktop. This is Intel response to R7 2700X and rest of Pinnacle Ridge series.

The response to Threadripper 2 is coming latter as Skylake-X refresh (up to 18 core) and Skylake-A (28 core). Current Skylake-X series is 7000. So Skylake refresh will be 8000 series.

[nuetzel]

[-]
As for tar+pxz, it's like a regular tar compression test but using the pxz (parallel xz [ie LZMA2]) compressor instead of the standard xz compressor. I use that compressor any time I'm packing a tar.xz since it's so much faster. Despite this, it's still in the standard Debian repositories. Testing it would involve something like this: tar -c -v --use-compress-program=pxz -f output.tar.xz input_path

Thanks again!

Do you mean:


pixz - Parallel, indexing version of XZ
Pixz supports automatic indexing and parallel compression and decompression using all available CPU cores.

[DavidKL]

TR 2990WX destroyed the competition in all benchmarks here. Meanwhile, over at Anandtech they show a completely different picture.

I can't say I'm surprised. In fact, I can say that is exactly what I expected.

[GreenReaper]

As for tar+pxz, it's like a regular tar compression test but using the pxz (parallel xz [ie LZMA2]) compressor instead of the standard xz compressor. I use that compressor any time I'm packing a tar.xz since it's so much faster. Despite this, it's still in the standard Debian repositories. Testing it would involve something like this: tar -c -v --use-compress-program=pxz -f output.tar.xz input_path

Modern versions of xz already support threaded compression (though not decompression). You can use 'xz -T0' [zero] to use all available threads, or specify an amount. I played with differing levels of compression for our database dumps (which end up being ~4GB compressed) but found using all available threads with the default compression ratio worked well for me - as you'd hope, since it's the default.

[andrei_me]

haasn welcome to the forums

On topic:

Amazing how the Intel Core i9 7980XE, which is supposedly a 165W TDP consumes almost the same energy of the TR 2990WX, which is 250W TDP, even taking the whole system in account

I remember reading at AnandTech that they got the i9 7980XE using ~190W during the tests, seems like Intel is really eager to show the world that TDP != Power consumption

[juanrga]

haasn welcome to the forums

On topic:

Amazing how the Intel Core i9 7980XE, which is supposedly a 165W TDP consumes almost the same energy of the TR 2990WX, which is 250W TDP, even taking the whole system in account

I remember reading at AnandTech that they got the i9 7980XE using ~190W during the tests, seems like Intel is really eager to show the world that TDP != Power consumption

AT used an AVX powervirus that stress Intel uarch but doesn't stress Zen.

Under ordinary loads, Intel chips are within its TDP (unless MCE and other overclocking stuff is enabled via BIOS), whereas many Zen systems aren't within TDP even at stock settings. E.g. the 105W 2700X is in reality a 140W chip.

[chithanh]

AT used an AVX powervirus that stress Intel uarch but doesn't stress Zen.

Under ordinary loads, Intel chips are within its TDP (unless MCE and other overclocking stuff is enabled via BIOS), whereas many Zen systems aren't withing TDP even at stock settings.

You can see from Phoronix graphs that the 2990WX doesn't consume much more power than the 7980XE, and Phoronix doesn't use any "AVX powervirus".

In multi-threaded workloads like Stockfish, the Threadripper 2990WX had an average AC system power draw of around 381 Watts compared to 244 Watts on the 2950X or 315 Watts on the 7980XE.

So this means compared to the 2990WX, the 7980XE has 85 W lower TDP but 66 W lower actual AC system power consumption. The remark about Intel staying within TDP and Zen not seems therefore quite far-fetched when talking about Threadripper.

[LukeP]

It's disappointing there are no machine learning benchmarks. Machine learning is pretty big at the moment. Sometimes useful to use a cpu.

[LukeP]

It would be good to see some machine learning benchmarks. Sometimes cpus are used for this.