Stability issues on 'high efficiency' build

nwraae · October 4, 2020

Hi all, I got really inspired by the LTT Video from December 'We built a PC more efficient than a console!' and tried to tackle it for myself. I'm not new to building PCs, but this was something a little different for me.

I live in a 12VDC powered tiny home and really want to prioritize performance per watt. Unfortunately, it feels like there are SO MANY variables to mess with in chasing a stable system.

System Specs:

Ubuntu 20.04 (kernel 5.8.13-lowlatency, OIBAF PPA amdgpu drivers)

250W DC-DC M4-ATX Power supply

AMD Ryzen 7 3700x CPU

Gigabyte B550M DS3H Mobo (most recent BIOS)

G.Skill TridentZ Series 2x 16GB DDR4

Crucial P1 500GB NVMe

Gigabyte Radeon RX 5600 XT

Based on the video referenced above, I input a bunch of power-saving BIOS tweaks:

CPU Clock 2.3Ghz

Vcore 0.875V

25W Package power limit

PCI-E Link speed Gen 1

Memory clock 3000Mhz

Mem voltage 1.25v

Timings 14-14-14-30 1T

VTT_SOC 0.788

VDDG 0.9v

VDDP 0.7v

And then, using CoreCtl to control GPU settings, I have:

GPU Freq range 800-1650

GPU Mem max @ 750Mhz

Voltage 850mv @ All frequencies

75W Power limit

Unfortunately, I seem to get GPU Black Screen crashes under medium-heavy load. (amdgpu ring gfx timeout error). Screen locks, a few seconds of audio continue, screen goes black, hard reset required.

Things I've tried: PCIE Gen move to 3.0, Messing with memory subtimings (nothing passes memtest86+), slight increases in BIOs voltage or GPU voltage... nothing seems to work.

So, it feels like I have too many variables to keep track of. Can anyone give me some guidance on what has worked for you? Or, maybe an 'order of operations' for trying stuff out?

Thanks

porina · October 4, 2020

Establish what works and what doesn't.

Don't run any mods, run stock. Is it stable?

Only apply CPU mods, is it stable?

Only apply GPU mods, with CPU stock, is it stable?

You can then try focusing on what area was unstable.

Edit: personally, I'd really not want to tinker with clock/voltages on either as a first step. Just try power limit by itself as a stepping point.

nwraae · October 4, 2020

Thanks, I'll start there. Although with this small PSU, I'm not sure the stock values aren't going to trip up against the power limits. For example, I'm not sure the 3700x can be PPT limited to 25w without the other values in play. But, I'll try it all.