I was not aware of such command before, thank you very much!
1 Like
Yep, thermal throttling. Reduce the memory to 0 and slowly raise until it starts throttling.
To get the most out of your card you’ll need to upgrade the thermal pads on the memory chips.
1 Like
Is this to query thermal throttling on the core or memory junction temps?
1 Like
any thermal throttle, but since you can see the core temp and unless the fans arent spinning at all, it will be because of the memory temps. pretty common on gddr6x cards.
1 Like
I have been running stable for 10 plus days on these settings for 3090 FEs. The only thing different that I did today is change the fans to auto to test its stability. I changed out all the thermal pads for the 3090 FEs.
Change to locked clocks and it should bring your hashrate up and power down. No need for power limits
Is there a stable locked clock you’d recommend?
It will vary depending on how high youre pushing the memory. For non lhr the goal is raise the memory as high as you can until it’s unstable, then find the lowest locked core clock that maintains full hashrate
switch to locked core clocks instead
looks good. 50+ watts less and same hashrate, gpus will run slightly cooler too.
Hello,
I also have a RTX 3090 FE and the same low hashrate.
After doing the thermal pad mod, my temps improved dramatically (around ~20ºC less on the memory), but they still hover over 90ºC (summer in spain can be VERY hot, more than 35ºC in the room sometimes), so it still throttles
Running:
nvidia-smi -q | egrep ‘Thermal Slowdown’
returns: SW Slowdown: Active as expected
So… my question is… Can I increase the thermal throttling temperature for memory a tiny bit? My guess is that they would last longer at 95ºC than at 110, and that will be the tiny bit I need to mantain 120MH/s. At least while the summer lasts…
thermal throttling on the memory side only happens at 110c. thats a built in to the gpu setting, not something you can change yourself
I don’t know what happens then, because GPU is around 70ºC and memory at 90ºC, I should have no throttling at all (but according to nvidia-smi, its throttling indeed, it only runs at 120MH/s briefly after a restart). Memory temp shows in red starting at 90ºC. I will check thermal pads tomorrow, but since I have much better temps I suspect that this is not the problem
I’m on 510 driver to unlock other LHR GPUs and lastest NBMiner, I don’t know if older drivers can help with this.Any clue on what might be causing the throttling?
using a locked core clock?
70c is too hot for a 3090 core imo.
Maybe I exaggerated a bit with the 70c, but definitely more than 60 with 100% fan. Tried both a locked and offset core clock. Keep in mind ambient temps in the room are way above 30ºC. Lately I am using -500 core offset. But using a fixed core like 1200 has the same effect: 120-ish on boot quickcly dropping to 95-100 after a couple of minutes. Using a really low fixed core like 800Mhz removes the throttling (both core and memory cools down) but drops the hashrate to hell as expected
I cannot check the rig at this moment to confirm the temps but i have a screenshot prior to the mod, I had 52ºC on the core, 110ºC on the memory, core -300 RAM 2400 Power limit 320. Had to drop to -600 on core, 1600 RAM 280 Power limit to at least mantain 100MH, else it dropped even lower.
dont use offsets at all, ~1140 mhz is all you need for 125mh.