AMD Ryzen 9 3900X Linux CPU Frequency Scaling Governor Benchmarks

[shmerl]

According to AMD the maximum temperature for 3900X is 95°C. My 3700X only hit that while forcing manual overclock in prime95 AVX2 on the stock cooler, as I was experimenting. It did not crash but reduced clocks instead.

After days of playing I found out that stock settings resulted in the best ST performance while having marginally worse MT than all-core OC (which reduced ST). With current BIOSes overclocking the CPU looks to be simply not worth it on Zen 2

I'm not overclocking it. Only using XMP profile for RAM (3600MHz / CL16).

[Michael]

no power usage numbers?

"The power consumption between the modes weren't compared due to the WattsUp Pro meters being busy on other systems and this testing was done prior to the recent Windows/Linux power discoveries; in fact, this testing was from back in July albeit slipped under the radar for publishing these figures...The next round of Ryzen CPUFreq benchmarks will also be looking at the performance-per-Watt metrics, but generally from our past findings have shown those to be close as well."

I only have 2~3 working WattsUp Pro meters. Unfortunately the company went out of business like 2 or 3 years back. I've tried buying several used ones off eBay and the like, but they all are borked. There is a known problem where after lots of extended use, they begin reporting like ~0 Watt usage.

[davidbepo]

no power usage numbers?

[numacross]

I'll run a test with lower RAM clocks on that. What do you mean by CPU mounting? Badly placed in the socket somehow?

For the reference, I run stressapptest to stress the RAM - it didn't reboot, but temps were lower (around 54°C). That's why I suspected, the above when building the kernel, it was a temperature related problem.

According to AMD the maximum temperature for 3900X is 95°C. My 3700X only hit that while forcing manual overclock in prime95 AVX2 on the stock cooler, as I was experimenting. It did not crash but reduced clocks instead.

After days of playing I found out that stock settings resulted in the best ST performance while having marginally worse MT than all-core OC (which reduced ST). With current BIOSes overclocking the CPU looks to be simply not worth it on Zen 2

[shmerl]

That's not even a thermally challenging application, as it is leaning on memory access a lot. Check your CPU mounting, and check your RAM stability.

I'll run a test with lower RAM clocks on that. What do you mean by CPU mounting? Badly placed in the socket somehow?

For the reference, I run stressapptest to stress the RAM - it didn't reboot, but temps were lower (around 54°C). That's why I suspected, the above when building the kernel, it was a temperature related problem.

[R41N3R]

I thought schedutil would do a little better, especially after reading the news today that it eventually becomes the default. Anyway, on Arch Linux it seems to be the default for AMD and it works well. For gaming there is gamemode and that will adjust the scheduler anyway to performance.

[xorbe]

Did anyone encounter reboots during heavy load? My Ryzen 9 3900X / Asrock X570 Taichi (firmware 2.0) starts rebooting when CPU temperature reaches somewhere between 70°C and 80°C. For example when building Linux kernel.

That's not even a thermally challenging application, as it is leaning on memory access a lot. Check your CPU mounting, and check your RAM stability.

[blackshard]

Schedutil behaves really poorly, exactly as it was behaving years ago...

[shmerl]

Haven't hit that behavior with stock coolers.

What are your max CPU temperatures during heavy load? Also, this potentially can be Asrock specific issue for me.

[Michael]

Did anyone encounter reboots during heavy load? My Ryzen 9 3900X / Asrock X570 Taichi (firmware 2.0) starts rebooting when CPU temperature reaches somewhere between 70°C and 80°C. For example when building Linux kernel.

Haven't hit that behavior with stock coolers.