I assembled my 7401P with Tyan S8026 and 8 DDR4-2666 DIMM. It's supposed to have an "all-core boost" of 2.8GHz, but I am only getting 2.35GHz when running BOINC projects. Searching online, information has been thin but I am definitely not alone: EPYC 7401 - Slow Rendering in 3Dsmax and VRay
With 48 threads of work, it starts with 2.6GHz, quickly drops below 2.5GHz, and takes quite a few minutes before stabilizing at 2.35GHz. The frequency fluctuates around that frequency. If I only run stress -n 24, it can sustain higher frequencies a bit longer. I didn't wait long enough for it to drop since I am not going to run only 24 threads. cpufreq has scaling_max_freq at 2GHz (the base frequency), so I don't think governor is playing a role here.
CPU temperature hovers around 32C and I have NH-U14S. It's surprisingly low so I lowered fan speed to see temperature climb in order to verify sensor is correct. (Phoronix just published an article today matching my data: Cooling AMD EPYC With Noctua Coolers: NH-U9 TR4-SP3, NH-U12S TR4-SP3, NH-U14S TR4-SP3 - Phoronix. )
This MB has a setting called "Determinism Slider" and "cTDP control", but setting to performance and maximal allowed TDP doesn't seem to make any difference.
Does anyone here also have tested EPYC 7401P, or any one of the three single socket version? Is what I am seeing normal? I didn't find any thing related to frequency in reviews I saw so far, including the one on our site: https://www.servethehome.com/amd-epyc-7401p-linux-benchmarks-and-review-something-special/
With 48 threads of work, it starts with 2.6GHz, quickly drops below 2.5GHz, and takes quite a few minutes before stabilizing at 2.35GHz. The frequency fluctuates around that frequency. If I only run stress -n 24, it can sustain higher frequencies a bit longer. I didn't wait long enough for it to drop since I am not going to run only 24 threads. cpufreq has scaling_max_freq at 2GHz (the base frequency), so I don't think governor is playing a role here.
CPU temperature hovers around 32C and I have NH-U14S. It's surprisingly low so I lowered fan speed to see temperature climb in order to verify sensor is correct. (Phoronix just published an article today matching my data: Cooling AMD EPYC With Noctua Coolers: NH-U9 TR4-SP3, NH-U12S TR4-SP3, NH-U14S TR4-SP3 - Phoronix. )
This MB has a setting called "Determinism Slider" and "cTDP control", but setting to performance and maximal allowed TDP doesn't seem to make any difference.
Does anyone here also have tested EPYC 7401P, or any one of the three single socket version? Is what I am seeing normal? I didn't find any thing related to frequency in reviews I saw so far, including the one on our site: https://www.servethehome.com/amd-epyc-7401p-linux-benchmarks-and-review-something-special/
