Return to Level1Techs.com

Odd TR2 2950X Phoronix Results

So, the hpcc is probably explainable by my having done dumb stuff with the libraries. That makes sense.

But I’ve enabled PBO2 with board defaults and am using the enermax TR2 cooler. I have verified that my temps are 55-60c, decently far away from the 68c where TR2 begins to throttle. I even raised the temp throttle limit on the aorus extreme to 80c with no effect (two pbo2 scores).

https://openbenchmarking.org/result/1809071-AR-1808115RA52

Anyone have any ideas as to what the issue might be? It’s sorta odd… it’s been a long week so perhaps I’m also just not thinking clearly and I’ve done something obviously dumb. I’ve tested on fedora 28 (not shown) and got similar results, though.

Performance covernor confirmed, except there is a UEFI bug on the Aorus Extreme right now where the performance governor for hyperthreaded cpu 16 is… missing? From the acpi tables or whatever?

$ cpufreqctl gov
cat: /sys/devices/system/cpu/cpu31/cpufreq/scaling_governor: No such file or directory
performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance performance

kinda doubt that would gimp performance this bad though.

PBO2 or higher where appliccable override and set their own caps, is why; so that’s on a par. Monitoring your temps is the way to go for this, like you did.

Cannot help in regard to the actual question am afraid, but much as you’re gonna hate me saying this, Windows has its uses. Have you run it through its hoops there, compare what you get with what “most” are getting?
Would at least alleviate (or showcase) certain aspects of this equation.

I did some quickie testing on windows and numbers there seemed to line up better than on Linux.

I just switched to Uma more and kernel 4.19-rc2 and each change did not change much. I’d expect the 2950 to exceed 2.0 in indigo. It does not.

Then i’d start from the dumb things :slight_smile:

Disable overrides, run it again. Is it still as bad? Relatively speaking of course, different power factors.

edit: Saw your mentioning UMA, a word of warning. Gigabyte is picky in the BIOS! Forget RM, set it to BIOS as “/die” or “/interleave” accordingly, save and test again.

Same behavior on fedora as ubuntu is weird tho. And the manual oc 1950x is super fast. That link has several tests I did including full auto tests with the 2950 no pbo.

I didn’t upload any fedora tests just ubunto too.

I am running 2666 memory ecc at 15,15,15,34 timings. Might try 3200 tomorrow at tighter timings. Maybe memory is making this much of a difference. That’s a shocker if so.

On the surface, it shouldn’t be as 3200 is the sweetspot for the IF, timings becoming the primary factor once there.
Below the surface… could the ECC chip affect the IF directly?
If you have info on this, i’d love to know please.

  • am now wondering about tracing too… Googleing that now.

No idea about the strange relative performance of your 2950X vs your 1950X. But working mpi libraries made a ~10X difference for me with hpcc.

I think getting the libraries set up correctly might be a necessary first step, at least if the results are to mean anything. Also, notice that the compiler options superimposed on many of the charts differ between your (and my) results vs the benchmarks.

1 Like

So I swapped out to gcc 8.2 and then to just running indigo (which doesn’t recompile itself anyway) now using tight timing samsun bdie memory at 3200 with pbo and all that. Still only getting around 1.8 in indigo and not the expected 2.0+

Performance is worse for a 1950x on windows according to the db at indigo, and yet tech report has even better numbers than phoronix except on indigo.

Hilarious. This is a 2990wx on windows.

Using Process lasso with core 0 turned off.

~1.8 seems like a typical result for 1950X. https://www.indigorenderer.com/benchmark-results?page=1&sort_asc=device_name&filter=cpu That 3.55 for a 2990X with linux is interesting.

I just ran it on my 2700X with Fedora 28 and got 1.25. Same as phoronix results: https://openbenchmarking.org/result/1807207-RA-RYZENOS6018 Side note, indigobench doesn’t run for me in pts docker image.

Indeed. But pts shows about 2 for the 1950x as well.
As does tech report.
Ddr2666 Dr/Ds is not really much slower than 3200 Sr/ss, either. Which is not entirely unexpected. I’m sure 2666 sr/ss is slower… but not Dr/Ds.

FEDORA 2990WX informal result.

I have swapped over to the designare motherboard AND updated my bios, and performance delta is gone. This is a 1950x at stock

So now to see if I can daignose what’s up with this other Mobo.

1 Like

things i woud check is if hpet is on or off.

see if you can change numa/uma settings to the various configurations and also the memory interleave settings