Same GPUs, different mobo, can't keep the clocks

papajo · April 2, 2021

Ok so I had 3 desktops with 6 vega 64 nitro+ GPUs (2 in each dekstop case)

I used them for mining, I downgraded the bios to vega 56 nitro+ and configured custom timings on the HMB memory and used different powertables.

The cards were cramped inside the cases with temps (at full load) around 60C for the GPU and around 70C for the HBM memory

But at 1100Mhz (GPU) and 1045Mhz (HBM) they were mining at around 54.8Mh/s on ethereum (a couple could go 1060Mhz which resulted in 56.7Mh/s)
The cards run just fine 100% stable 24/7
What's the issue then?
Well I decided to buy an asrock h110 pro BTC motherboard and make an open mining rig with extra fans etc.

The temperatures of the cards are much lower now (I also changed the paste they had with Kryonaut Extreme I think it is the best after liquid metal)

Now I have like 30-40ishC on the GPUs and 50-60ishC on the HBM at full load

The issue is that for some reasons the same exactly cards that used to work (same OS same drivers same mining software,same memory timings same powertables) at 1045 and 1060Mhz 100% stable 24/7 at higher temps on the memory now refuse to go a step above 1020Mhz!
It is not the fact that I took them apart and changed the paste because I tested them for a couple of hours (on the desktops) and they run at the same settings with lower temps stable.

I can get a card or two run above 1020Mhz but its always a lottery Its like something happens if I set the frequency of all of them (e.g GPU1 may hang, but on reboot GPU1 may work but GPU4 now hangs) and I get BSOD "thread_stuck_in_device_driver" or something like that.

I think it has something to do with lag/interference with the driver since I use 6 GPUs (even adrenaline gets laggy if I click on GPU boxes to check their settings like a 5 to 10 second delay when clicking e.g GPU2 from GPU1 etc)

I use the GUI of this tool https://github.com/Eliovp/amdmemorytweak

To set the OC profiles.

Any ideas on what to try?

P.S EnableULPS and EnableCrossfireAutolink registry keys are disabled for all the GPUs.