The Linux Benchmarking Continues On The Threadripper 2950X & 2990WX

[phoronix]

Phoronix: The Linux Benchmarking Continues On The Threadripper 2950X & 2990WX

While I haven't posted any new Threadripper 2950X/2990WX benchmarks since the embargo expired on Monday with the Threadripper 2 Linux review and some Windows 10 vs. Linux benchmarks, tests have continued under Linux -- as well as FreeBSD...

http://www.phoronix.com/scan.php?page=news_item&px=Threadripper-2900-Benchmarks-Co

[lem79]

I think at this point, we would really benefit from having CPU usage charts alongside all of the benchmarks somehow. This is very telling: http://openbenchmarking.org/result/1...RA-CPUUSAGED10

So with x265 for example, on the 2990WX, you can basically encode three videos all in parallel, because a single instance only really uses 30% of the available threads. Maybe another data point on existing graphs simply saying average CPU usage during that particular test? Or maybe, if multiple CPUs are being benchmarked, a chart that shows all of them on the one chart, X axis being time, Y axis being % usage. Just a thought anyway, but I think pretty important to get a good idea of how well utilised these extreme core count CPUs are.

[Michael]

I think at this point, we would really benefit from having CPU usage charts alongside all of the benchmarks somehow. This is very telling: http://openbenchmarking.org/result/1...RA-CPUUSAGED10

So with x265 for example, on the 2990WX, you can basically encode three videos all in parallel, because a single instance only really uses 30% of the available threads. Maybe another data point on existing graphs simply saying average CPU usage during that particular test? Or maybe, if multiple CPUs are being benchmarked, a chart that shows all of them on the one chart, X axis being time, Y axis being % usage. Just a thought anyway, but I think pretty important to get a good idea of how well utilised these extreme core count CPUs are.

PTS generates those graphs when MONITOR=cpu.usage environment variable is set and does take care of merging them in the case of mutliple CPU comparisons, etc. But short of adding all of those graphs separate, bit overboard or not a good way for efficiently highlighting all of that data. I would assume if including all the CPU usage graphs people would then complain the number of pages in the article are now twice as long...

[lem79]

Oh right, true.. I forget how good Premium is! They should all just sign up for Premium, being the obvious answer to double the graphs

So to avoid adding any extra graphs, maybe just a single number (average CPU usage in %) on the right hand side of the graph, corresponding to each system. I guess it's not always that simple, as even if there's unused resources, it doesn't mean there will be simple linear scaling by running more tasks in parallel. At least as a rough guide it'd be ok.

[M@yeulC]

Well, if you are interested in cpu usage percentage, keep in mind that it's wrong. It would be *extremely* interesting if the PTS could report both, especially on tests that report "100% CPU usage". We would see a lot more stalling on benchmarks that don't scale extremely well. It could just be a lone IPC figure somewhere on the graph (times the frequency? IPS?)

That said, I suppose that real CPU usage can be directly correlated with the system power draw

[edwaleni]

Phoronix makes Forbes.