AMD Ryzen 9 3900X Linux CPU Frequency Scaling Governor Benchmarks

[shmerl]

According to one report here: https://github.com/groeck/k10temp/is...ment-528016506
reboots could be due to bug in the firwmare:

The recent AGESA 1.0.0.3 ABB dropped the fan temp trip to 75° C. At that point all fans go to 100% speed.

I have custom cooling loop on my 3900X. But if the fans don't run there is no heat exchange taking place and the cpu overtemps. There is a bug in the BIOS that disables the fan headers sporadically. That is what caused by overtemp conditions. I have since removed all fans from the motherboard headers and plugged them into a standalone fan controller. Problem solved. Still waiting on the Beta 0002 BIOS with the fixed fan header issue to make it to an official BIOS.

I have the ASUS Crosshair VII Hero motherboard. BIOS 2703 AGESA 1.0.0.3 Patch ABB.

[zerothruster]

Thanks. That's the one that comes with SystemRescueCD? Is there some version of it that works with UEFI, or they are all legacy BIOS?

I.e. the one I know is memtest86+: https://www.memtest.org

From memory, that fork stopped being developed and the last version locked up on many motherboards if running all cores in parallel. The original version is again being updated and still has free downloads, at least for individual use, from https://www.memtest86.com

[willmore]

Is stressapptest enough for that? I run it a few times, and it was fine.

My personal regimen for breaking in new machines is a day of memtest86+ and then at least one loop through all FFT sizes with prime95. You can optionally do some of the more limited tests which just stress power or L1/L2 cache as they help test power delivery and cooling.

I don't trust a machine that hasn't gone through that.

I really wish memtest86+ hadn't died. I don't really trust the closed source replacement.

[shmerl]

I really wish memtest86+ hadn't died. I don't really trust the closed source replacement.

Stressapptest is open source: https://github.com/stressapptest/stressapptest

[Dedale]

Thank you Michael for these interesting tests.

[AndyChow]

Thanks. That's the one that comes with SystemRescueCD? Is there some version of it that works with UEFI, or they are all legacy BIOS?

I.e. the one I know is memtest86+: https://www.memtest.org

Yeah... there's memtest86 and memtest86+, and both are different. The link you provided is good. However, there is also: https://www.memtest86.com/ which supports UEFI.

Both are fine and work well.

You just boot into it, and it runs a few patterns over all the RAM. If you get a single error, at least one of your stick is bad. The test is never ending, so run it a few passes (which can take a long time), and if you have 0 errors the problem is elsewhere. Most of my experience is with memtest86+, the link you provided. UEFI is mostly not an advantage over legacy BIOS because you're just testing RAM in a single live session, not installing anything nor doing some kernel loading voodoo.

[shmerl]

Most of my experience is with memtest86+, the link you provided. UEFI is mostly not an advantage over legacy BIOS because you're just testing RAM in a single live session, not installing anything nor doing some kernel loading voodoo.

Thanks. I'll try that. There is an option for CSM legacy boot, so it should work.

[AndyChow]

Thanks. I'll try that. There is an option for CSM legacy boot, so it should work.

No problem. If you do get errors, you can start removing sticks to identify which one(s) are bad. I once had two bad sticks from a kit of four. It really sucks to find out you have a bad stick. That said, it's better to find out sooner than later, since a stick that has a few errors will silently corrupt a lot of data over time as you are doing things, in addition to causing random system crashes.

As user willmore has said, "I don't trust a machine that hasn't gone through that." memtest86 is the first thing I run after setting up a new rig, usually for a few days, then I run prime95. I also run badblocks and diskscan on any new hard-drive, though diskscan can give false-fails on shingled drives.

I hope you solve your issue.

[microcode]

Total system energy consumed/dissipated for a given benchmark (with a consistent amount of work, like a compilation task, or a video encoding task) would be a very interesting comparison. Does PTS have support for making charts like that, or would it even be necessary? Michael


IMHO that's more interesting than a chart of smoothed electrical load over time.

[arQon]

numacross - The impression I get re OC on Ryzen (mostly from testing by GN et al) is that there simply isn't much room to begin with, and the chips will take what little there is automatically anyway. So BIOS updates won't meaningfully change that, and short of exotic cooling and a lot of +V not much else will either.