-
Notifications
You must be signed in to change notification settings - Fork 229
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
GCN optimized code for RandomX #270
Conversation
Tested and working on Polaris (RX470/570/480/580) in Win10. Used 1 thread per gpu. Tried various intensities to get the best result. More testing to follow. |
@Bathmat there is a bug with 2 threads per GPU in this pull request - it allocates dataset 2 times, that's why it doesn't work on 4 GB GPUs and works poorly on 8 GB GPUs. I'll fix it soon. |
That would explain why it was crashing when I tried 2 threads. Thanks. I just thought it was RandomX design to not be able to use 2 threads. |
Dual thread working now. Approx 7-8% boost in hashrate for rx/loki, 15-20% boost for rx/wow. I just took the intensities I was using for single thread and halved them. So for rx/loki, I was using Anything over 832 total intensity would crash/very poor hashrate for 470/570 on rx/loki. |
rx/loki
|
rx/wow
|
Optimized code works only on Vega and Polaris GPUs with Adrenaline drivers (Windows) or amdgpu-pro drivers (Linux). Vega 64 (1700/1100 MHz) does 1283 h/s on rx/test, 1296 h/s on rx/wow, 1124 h/s on rx/loki.