Claymore-Dual-Miner Does not work on Volta (V100) class hardware

Does not work on Volta (V100) class hardware

Open zebrax0r opened this issue 6 years ago • 15 comments

Hi.

I have some Volta class hardware - the V100's to be exact. The current release (10) does not seem to detect or understand the V100 + CUDA9 under CentOS 7.4...

Setting DAG epoch #147 for GPU1 Setting DAG epoch #147 for GPU0 Create GPU buffer for GPU1 Create GPU buffer for GPU0 GPU 0, GpuMiner cu_kd failed 8 (0), invalid device function GPU 0, Calc DAG failed! GPU 1, GpuMiner cu_kd failed 8 (0), invalid device function GPU 1, Calc DAG failed! ETH: 10/23/17-16:31:16 - New job from eth-eu.dwarfpool.com:8008 ETH - Total Speed: 0.000 Mh/s, Total Shares: 0, Rejected: 0, Time: 00:00 ETH: GPU0 0.000 Mh/s, GPU1 0.000 Mh/s SC - Total Speed: 0.000 Mh/s, Total Shares: 0, Rejected: 0 SC: GPU0 0.000 Mh/s, GPU1 0.000 Mh/s ETH: 10/23/17-16:31:17 - New job from eth-eu.dwarfpool.com:8008 ETH - Total Speed: 0.000 Mh/s, Total Shares: 0, Rejected: 0, Time: 00:00 ETH: GPU0 0.000 Mh/s, GPU1 0.000 Mh/s SC - Total Speed: 0.000 Mh/s, Total Shares: 0, Rejected: 0 SC: GPU0 0.000 Mh/s, GPU1 0.000 Mh/s GPU 1 failed Setting DAG epoch #147 for GPU1 GPU 0 failed Setting DAG epoch #147 for GPU0 GPU 1, GpuMiner cu_kd failed 8 (0), invalid device function GPU 1, Calc DAG failed! GPU 0, GpuMiner cu_kd failed 8 (0), invalid device function GPU 0, Calc DAG failed! Quit signal received...

Hopefully I can help you add support for it, if you have not already considered it.

Oct 23 '17 06:10 zebrax0r

Wanted to do some benchmarking here at Paperspace and also ran in to this issue.

V100+CUDA 9 +CuDNN 7 + Ubuntu 16.04

Oct 26 '17 16:10 dte

Unfortunately, still broken in 10.1...

Setting DAG epoch #148 for GPU0
Setting DAG epoch #148 for GPU1
Create GPU buffer for GPU0
Create GPU buffer for GPU1
GPU 0, GpuMiner cu_kd failed 8 (0), invalid device function
GPU 0, Calc DAG failed!
GPU 1, GpuMiner cu_kd failed 8 (0), invalid device function
GPU 1, Calc DAG failed!
GPU #0: Tesla V100-PCIE-16GB, 16152 MB available, 80 compute units, capability: 7.0
GPU #1: Tesla V100-PCIE-16GB, 16152 MB available, 80 compute units, capability: 7.0

Oct 28 '17 18:10 zebrax0r

Any update on this, finding same thing... Is there any mining software out there that works with Tesla V100s at this time?

Nov 06 '17 20:11 cyclecole

Nov 17 '17 06:11 tejdeeep

any updates on this?

Dec 03 '17 16:12 Prefinity

Nuh.

On 10.2, slightly newer CUDA9 packaging, still a fail...

GPU 1, GpuMiner cu_kd failed 8 (0), invalid device function GPU 1, Calc DAG failed! GPU 0, GpuMiner cu_kd failed 8 (0), invalid device function GPU 0, Calc DAG failed! GPU 2, GpuMiner cu_kd failed 8 (0), invalid device function GPU 2, Calc DAG failed! GPU 1 failed Setting DAG epoch #157 for GPU1 GPU 0 failed Setting DAG epoch #157 for GPU0 GPU 2 failed Setting DAG epoch #157 for GPU2 GPU 1, GpuMiner cu_kd failed 8 (0), invalid device function GPU 1, Calc DAG failed! GPU 0, GpuMiner cu_kd failed 8 (0), invalid device function GPU 0, Calc DAG failed! GPU 2, GpuMiner cu_kd failed 8 (0), invalid device function GPU 2, Calc DAG failed! GPU 1 failed GPU 0 failed GPU 2 failed Quit signal received...

Odd. Etherminer seems to work just fine?

Dec 11 '17 20:12 zebrax0r

Dec 28 '17 08:12 cockroachzl

Hi,

Any news on this issue? Miner just crash with errors provided before and with ethminer hash rate start on 90+ mh/s then it goes down to 50mh/s. And it will be cool if somebody feel free to share ./ethminer command using to run on his miner that assure stable hashrate. Sys info; Ubuntu 16.04 with last CUDA drivers.

Thanks

Jan 11 '18 17:01 ZEROF

Sigh. Was really hoping to have some action on this by now. It is the fastest GPU in the world. I'd have expected to see support for it by now. Perhaps there is much more to it than I understand, in terms of implementation difficulties.

I can get a stable hashrate at 93.5MH/s using the following string (using ethminer).

./ethminer --cl-parallel-hash 4 --farm-recheck 200 -G -F http://127.0.0.1/blah

Jan 15 '18 11:01 zebrax0r

Fastest is not necessarily best. I had a 1080 Ti which used about 250W. I replaces it with TWO 1070 cards that did nearly double the work for less power. IMHO hashes/s/W is more important that speed.

Jan 15 '18 11:01 Roaders

Granted, and agree entirely - but when power consumption is irrelevant - it'd be nice to have the support for the GPU, nonetheless.

Perhaps there is more to this - and your perspective is one that Claymore shares. I.e - they want to help the miners with the most common devices achieve gains - rather than support specific devices to just "go fast".

Who knows. This is just conjecture, I suppose.

Jan 15 '18 11:01 zebrax0r

Any updates here?