Tweet
Hi Everyone,
I have just built my new system, installed Arch and doing some initial benchmarks I'm getting very weird results.
Both CPU and GPU show really bad performance in multiple workloads tested with Phoronix Test Suite.
Some data about the system
OS: Arch Linux
Kernel: x86_64 Linux 5.0.4
CPU: Threadripper 2950x currently clocked at 4Ghz
Motherboard: Asus Zenith Extreme Alpha
Bios version: 0207
RAM: 64 GB at 3000Mhz
GPUs: 2 x RTX 2080 ti Gigabyte turbo
nvidia driver version: 418.56
from nvidia-smi:
power usage cap is 250W
Performance level is correctly ramped up by PowerMizer (nvidia-settings). Setting it to max performance manually doesn't change results.
CPU and Gpus are watercooled with a custom loop by EK: 360 + 240 radiators, loop order pupm/res combo, CPU, 360 rad, GPUs (parallel), 240 rad, pump/res combo
Temperatures are ~40 C idle, ~60C highest load. They are higher than I expected, and I might reseat both GPUs / replace thermal paddings, but I don't think that I am reaching thermal throttling.
Everything during BIOS setup and software installation went smoothly, all hardware is recognized properly and Arch seem to be working just fine as always.
BUT.
Performing some PTS unigine benchmarks I am getting ~ 20 fps in every test (valley, heaven, superposition), doesn't matter if I change settings from basic to ultra or whether I toggle occlusion shadows or refraction.
Power draw ranges from ~ 80W to 200W when doing Unigine superposition at Ultra settings.
Gpu utilization ranges from 15% to 60%.
At first I thought about some NVIDIA driver issues, then I started some tests on the CPU and core performance is very bad too.
Example: Scimark2 Computational test: composite yields a result of 488 Mflops, lower than an AMD Athlon II. Just for comparison, a Core i3 2100 on Openbenchmarking.org scores 1065 Mflops.
I get silmilar results from other tests aswell.
I also tried multithreaded benchmarks like himeno and again I get 1087 Mflops. i7 6700K scores 2626 Mflops.
Anomalies during CPU tests I noticed with top:
single core usage doesn't go past ~50% (shouldn't it stress to 100%?)
multithreaded tests are actually using just one core (and it goes to 100% usage in these tests)
from sensors data:
CPU temperature is ~13C idle, reached 25C during tests at worst, then went down quickly.
To complete the overview, ramspeed test with Phoronix Test Suite yields expected results. RAM seem to be working nicely.
Any idea about what's going on? Am I missing something obvious? GPU bad performance could be the result of a bottleneck due to CPU single core severely underperforming - afterall I am pairing lower-than-i3 performance with a double 2080 ti - but then why is my TR behaving that badly? If this was due to some mistake I made during CPU seating I would expect to see way higher temps...
I have just built my new system, installed Arch and doing some initial benchmarks I'm getting very weird results.
Both CPU and GPU show really bad performance in multiple workloads tested with Phoronix Test Suite.
Some data about the system
OS: Arch Linux
Kernel: x86_64 Linux 5.0.4
CPU: Threadripper 2950x currently clocked at 4Ghz
Motherboard: Asus Zenith Extreme Alpha
Bios version: 0207
RAM: 64 GB at 3000Mhz
GPUs: 2 x RTX 2080 ti Gigabyte turbo
nvidia driver version: 418.56
from nvidia-smi:
power usage cap is 250W
Performance level is correctly ramped up by PowerMizer (nvidia-settings). Setting it to max performance manually doesn't change results.
CPU and Gpus are watercooled with a custom loop by EK: 360 + 240 radiators, loop order pupm/res combo, CPU, 360 rad, GPUs (parallel), 240 rad, pump/res combo
Temperatures are ~40 C idle, ~60C highest load. They are higher than I expected, and I might reseat both GPUs / replace thermal paddings, but I don't think that I am reaching thermal throttling.
Everything during BIOS setup and software installation went smoothly, all hardware is recognized properly and Arch seem to be working just fine as always.
BUT.
Performing some PTS unigine benchmarks I am getting ~ 20 fps in every test (valley, heaven, superposition), doesn't matter if I change settings from basic to ultra or whether I toggle occlusion shadows or refraction.
Power draw ranges from ~ 80W to 200W when doing Unigine superposition at Ultra settings.
Gpu utilization ranges from 15% to 60%.
At first I thought about some NVIDIA driver issues, then I started some tests on the CPU and core performance is very bad too.
Example: Scimark2 Computational test: composite yields a result of 488 Mflops, lower than an AMD Athlon II. Just for comparison, a Core i3 2100 on Openbenchmarking.org scores 1065 Mflops.
I get silmilar results from other tests aswell.
I also tried multithreaded benchmarks like himeno and again I get 1087 Mflops. i7 6700K scores 2626 Mflops.
Anomalies during CPU tests I noticed with top:
single core usage doesn't go past ~50% (shouldn't it stress to 100%?)
multithreaded tests are actually using just one core (and it goes to 100% usage in these tests)
from sensors data:
CPU temperature is ~13C idle, reached 25C during tests at worst, then went down quickly.
To complete the overview, ramspeed test with Phoronix Test Suite yields expected results. RAM seem to be working nicely.
Any idea about what's going on? Am I missing something obvious? GPU bad performance could be the result of a bottleneck due to CPU single core severely underperforming - afterall I am pairing lower-than-i3 performance with a double 2080 ti - but then why is my TR behaving that badly? If this was due to some mistake I made during CPU seating I would expect to see way higher temps...
Comment