Posted: Sep 30, 2020 4:17 am
by SAK917
Note if you have an NVIDIA card and the CUDA is not being detected or not working:
1) Use GPU-z to check CUDA ticked.
2) If you have a CUDA dev kit installed try removing that and run default.

Thanks Kiore, I noted in my logs that it was detecting CUDA but unable to initialize and it was due to having the CUDA Dev Kit installed. I am not using it right now so I uninstalled and everything is working great and my Estimated PPD has gone from about 3.4M to 4.4M! We will see if that holds, but that is a huge increase just from optimizing drivers. My GPU temp has maybe gone up a degree or so, but I am still below 80C (it is sitting at 79C now) and the GPU utilization is parked at 100%.

Running CPUID's HWMonitor I notice something interesting as well. Before the new CUDA update I would often see the flags for Reliable Voltage Limit or Operational Voltage Limit being raised indicating the card was doing some throttling despite the GPU utilization not being at 100%. Now after watching it with the new core, I haven't had either of those flags raised at all, a first that I am aware of. It does, however, have my GPU clock pretty consistently at about 1845MHz for a slight overclock (the card is doing this on its own) and the Power Limit flag is on continuously, so I am assuming that the new core is now almost optimally using my card. That would help explain the 1M PPD increase in my estimated PPD... nice!