13-Way IBM POWER9 Talos II vs. Intel Xeon vs. AMD Linux Benchmarks On Debian

[darkbasic]

Michael is there any reason why you didn't include threadripper and core i9 in the comparison?
Also a Ryzen 2700X and a Core i7 would be good references for those who are not familiar with server CPUs.

[chithanh]

is there any reason why you didn't include threadripper

Yes, the reason is AMD marketing being, well, AMD marketing:

AMD requested I not run any side-by-side tests of Threadripper and EPYC

[darkbasic]

Yes, the reason is AMD marketing being, well, AMD marketing:

LOL that sounds really stupid.

Michael so maybe there is chance to run a second batch of tests in a separate article with just Power 9 plus the workstation cpus (Threadripper and Core i9)?

[Adarion]

Very interesting comparison. It's still astonishing how much results vary during different tests.
One would have to consider the overall price tag for the whole systems (CPUs, board, RAM, PSU, cooling...) before jumping to early conclusions. Moreover, the POWER ones probably might improve with better SW support and they come with Libreboot (and are likely made in smaller numbers than the x86 ones).

[bridgman]

Now that we have announced 32-core TR parts maybe we can get the marketing folks to relax that. I'll see if I can figure out who to ask.

[Vistaus]

Huh, AMD requested that to Michael? But I thought only Intel and nVidia were evil and AMD was the good guy?

[edwaleni]

n;tn

yes right I am interested in the 32core Threadripper 2 for my 2 mining machines to increase Monero/XMR mining output.

i also found the problem i had with one mining pc it was not the MSI mainboard it was a vega-64 who died they run now with 5 instead of 6 vega cards.

s.

What kind of hash rates were you seeing before the Vega died?

One use case for POWER9 and NVidia compute modules is their superior hash rates. But I haven't seen any measurements. The biggest concern is as always, the amount of watts consumed.

[milkylainen]

Are you sure ? My impression was that all the links were in the same speed range, ~25 GB/s raw in each direction.

To my understanding:

CAPI 2.0 built upon PCIe 4.0 (Power9): 16GT/s per lane, PCIe 4.0 16x = 32GB/s each direction.
CAPI 3.0 built upon BlueLink (Power9, same as NvLink): 25GT/s, 8x = 25GB/s in each direction, number of sublinks unknown.
NvLink 2.0 built upon BlueLink (Power9): 25GT/s, 8x = 25GB/s in each direction * 6 sublinks = as one NvLink = 150GB/s.
NvLink 2.0 (NVidia Tesla V100): 25GT/s, 8x = 25GB/s in each direction * 6 sublinks = as one NvLink = 150GB/s.

[nuetzel]

Yes, the reason is AMD marketing being, well, AMD marketing:

Only somewhat OT:

https://www.heise.de/newsticker/meldung/In-eigener-Sache-Nvidia-NDA-als-Maulkorb-fuer-Journalisten-4091751.html

Sorry, German.

[nuetzel]

yes right I am interested in the 32core Threadripper 2 for my 2 mining machines to increase Monero/XMR mining output.

i also found the problem i had with one mining pc it was not the MSI mainboard it was a vega-64 who died they run now with 5 instead of 6 vega cards.

What about this, then:

Ende der Grafikkarten-Ära: 8000 ASIC-Miner für Zcash, Bitcoin Gold & Co.

https://www.heise.de/newsticker/meldung/Ende-der-Grafikkarten-Aera-8000-ASIC-Miner-fuer-Zcash-Bitcoin-Gold-Co-4091821.html