[Bug] multiple GPUs

[brianleect]

ran it with an instance w 10x GTX 1080 Ti.

Was profanity made with running on instances with multiple gpus in mind?

Gave it a simple command to match for 'dead' but seems that I just get flooded with a stream of total Ghz on terminal without any private key info printed.

[brianleect]

Just tested again with 2x RTX 3090 & seems to work fine. Weird. Will try with a 10x RTX 3090 and see if I can replicate the error seen with 10x GTX 1080 Ti.

[brianleect]

Tested again with a 7x RTX 3090 and it works to generate a addresses at start. However, it later ended up resulting in a seg fault. Seems that the amount of logs printed is a lot more than the usual as well.

[brianleect]

Might be linked to
johguse#69
johguse#45

[k06a]

Thanks for the issue, I will try to investigate. Could you try make your window a bit wider? This could help possibly.

[brianleect]

Np, thanks for the response. Hope the issue regarding seg fault can be identified. The window was already maximized sadly so I don't think that's the issue.

If it helps I was attempting to use vast.ai cloud gpu with nvidia/opencl image to run profanity2.

[galekseev]

Multiple GPUs work, and the environment side of what you ran into is now covered by #62.

Fresh evidence from vast.ai: an 8x RTX 5090 instance ran for about ten minutes without crashing at 12878 MH/s, against 1694 MH/s for a single card of the same model on the same tooling - a 7.6x scale-up, so all eight devices were doing real work. Logs are in bench-logs/ on master, see 8xRTX5090.vastai.log and 1xRTX5090.vastai.log.

You mentioned you were running the nvidia/opencl image on vast.ai. That is exactly what #62 removes the need for: there is now an official image with the OpenCL ICD registered inside it, so a rented machine only needs the image and the scoring options. It also refuses to start with a clear message when no OpenCL device is visible, instead of printing an empty Devices: list, and it writes every result to /workspace/profanity2.log - the hashrate counter goes to stderr and stays out of that file, so results can no longer be lost in the stream of speed updates that filled your first screenshot. See docs/VASTAI.md.

Closing as resolved for the multi-GPU question. The segmentation fault you saw later on 7x RTX 3090 was never reproduced here and nothing in #62 targets it specifically - if it still happens on current master, please open a fresh issue with the GPU model, driver version and the last lines of output, that will be much easier to track down separately.