nicehash / nheqminer

Equihash miner for NiceHash
https://www.nicehash.com
MIT License
768 stars 582 forks source link

NiceHash Equihash miner fails to benchmark on GPUs #82

Open nYrXvcpk2l47ciid opened 8 years ago

nYrXvcpk2l47ciid commented 8 years ago

From @nYrXvcpk2l47ciid on October 29, 2016 23:13

The NiceHash Equihash miner fails to benchmark on GPUs when the "t" flag is set to 0. Without the "t" flag set, the benchmark is performed on both the CPU and GPU.

C:\Users\Sydney\Desktop\Files\NiceHashMiner_v1.7.3.1\bin\nheqminer_v0.3a>nheqminer.exe -b -cd 0

        ==================== www.nicehash.com ====================
                Equihash CPU&GPU Miner for NiceHash v0.3a
        Thanks to Zcash developers for providing base of the code.
            Special thanks to tromp and xenoncat for providing
                 optimized CPU and CUDA equihash solvers.
        ==================== www.nicehash.com ====================

Setting log level to 2
[2016-10-29 16:44:46.674864][0x0006c7d4]: Using SSE2: YES
[2016-10-29 16:44:46.676863][0x0006c7d4]: Using AVX: YES
[2016-10-29 16:44:46.678864][0x0006c7d4]: Using AVX2: YES
[2016-10-29 16:44:46.825014][0x0006c7d4]: Benchmarking CUDA worker (CUDA-TROMP) GeForce GTX 750 (#0)
 BLOCKS=28, THREADS=64
[2016-10-29 16:44:46.944156][0x0006c7d4]: Benchmarking CPU worker (CPU-XENONCAT-AVX2)
[2016-10-29 16:44:46.946158][0x0006c7d4]: Benchmarking CPU worker (CPU-XENONCAT-AVX2)
[2016-10-29 16:44:46.948161][0x0006c7d4]: Benchmarking CPU worker (CPU-XENONCAT-AVX2)
[2016-10-29 16:44:46.950162][0x0006c7d4]: Benchmark starting... this may take several minutes, pleas
e wait...
[2016-10-29 16:44:47.431624][0x0006c7d4]: Benchmark done!
[2016-10-29 16:44:47.432624][0x0006c7d4]: Total time : 421 ms
[2016-10-29 16:44:47.433626][0x0006c7d4]: Total iterations: 200
[2016-10-29 16:44:47.435627][0x0006c7d4]: Total solutions found: 4
[2016-10-29 16:44:47.436628][0x0006c7d4]: Speed: 475.059 I/s
[2016-10-29 16:44:47.438630][0x0006c7d4]: Speed: 9.50119 Sols/s

The results reported are likely sums of the units' hash rates.

With the number of CPU threads set to 0, however, the benchmark does not run properly.

C:\Users\Sydney\Desktop\Files\NiceHashMiner_v1.7.3.1\bin\nheqminer_v0.3a>nheqminer.exe -b -cd 0 -t 0

        ==================== www.nicehash.com ====================
                Equihash CPU&GPU Miner for NiceHash v0.3a
        Thanks to Zcash developers for providing base of the code.
            Special thanks to tromp and xenoncat for providing
                 optimized CPU and CUDA equihash solvers.
        ==================== www.nicehash.com ====================

Setting log level to 2
[2016-10-29 16:45:07.748957][0x0006c7d8]: Using SSE2: YES
[2016-10-29 16:45:07.750960][0x0006c7d8]: Using AVX: YES
[2016-10-29 16:45:07.752962][0x0006c7d8]: Using AVX2: YES
[2016-10-29 16:45:07.897099][0x0006c7d8]: Benchmarking CUDA worker (CUDA-TROMP) GeForce GTX 750 (#0)
 BLOCKS=28, THREADS=64
[2016-10-29 16:45:08.019217][0x0006c7d8]: Benchmark starting... this may take several minutes, pleas
e wait...
[2016-10-29 16:45:08.077273][0x0006c7d8]: Benchmark done!
[2016-10-29 16:45:08.079274][0x0006c7d8]: Total time : 37 ms
[2016-10-29 16:45:08.081276][0x0006c7d8]: Total iterations: 200
[2016-10-29 16:45:08.082277][0x0006c7d8]: Total solutions found: 0
[2016-10-29 16:45:08.084279][0x0006c7d8]: Speed: 5405.41 I/s
[2016-10-29 16:45:08.085280][0x0006c7d8]: Speed: 0 Sols/s

Now we don't get correct results for the GPU's hash rate. So of course it is necessary to set the "t" flag to 0, but the miner does not react to this correctly. This could be an issue with nheqminer.cs (https://github.com/nicehash/NiceHashMiner/blob/master/NiceHashMiner/Miners/nheqminer.cs#L66$L71) or the miner itself.

Copied from original issue: nicehash/NiceHashMiner#295

nYrXvcpk2l47ciid commented 8 years ago

From @ansemz on October 30, 2016 2:48

I have the same issue.

nYrXvcpk2l47ciid commented 8 years ago

From @S74nk0 on October 30, 2016 20:41

NiceHash Miner handles this correctly. You can check your logs that generate benchmark input. If you are referring this as an issue with nheqminer open an issue there. But nheqminer is intended to run on CPU by default.

nYrXvcpk2l47ciid commented 8 years ago

From @ansemz on October 31, 2016 3:47

[2016-10-30 09:55:34] [INFO] [BENCHMARK] Benchmark starts [2016-10-30 09:55:34] [INFO] [nheqminer-MINER_ID(6)-DEVICE_IDs(0)] Starting benchmark: -b -t 0 -cd 0 [2016-10-30 09:55:34] [INFO] [nheqminer-MINER_ID(6)-DEVICE_IDs(0)] Using miner: bin\nheqminer_v0.3a\nheqminer.exe [2016-10-30 09:55:35] [INFO] [BENCHMARK] Final Speed: 0.000 H/s [2016-10-30 09:55:35] [INFO] [BENCHMARK] Benchmark ends [2016-10-30 09:55:35] [INFO] [FormBenchmark] EndBenchmark() benchmark routine finished [2016-10-30 09:55:35] [INFO] [CopyBenchmarks] Checking for benchmarks to copy [2016-10-30 09:55:42] [INFO] [NICEHASH] Balance get [2016-10-30 09:55:52] [INFO] [nheqminer-MINER_ID(7)-DEVICE_IDs(NOT_SET)] NEW MINER CREATED [2016-10-30 09:55:53] [INFO] [BENCHMARK] Benchmark starts [2016-10-30 09:55:53] [INFO] [nheqminer-MINER_ID(7)-DEVICE_IDs(0)] Starting benchmark: -b -t 0 -cd 0 [2016-10-30 09:55:53] [INFO] [nheqminer-MINER_ID(7)-DEVICE_IDs(0)] Using miner: bin\nheqminer_v0.3a\nheqminer.exe [2016-10-30 09:55:54] [INFO] [BENCHMARK] Final Speed: 0.000 H/s [2016-10-30 09:55:54] [INFO] [BENCHMARK] Benchmark ends

chronosek commented 8 years ago

0.3a version was compiled only for compute capability 5.2 , your card is older, you need compile cuda_tromp.dll for your card or use older (slower) from version 0.2a .... Or you can try dll compiled by me http://chronx.pl/cuda/cuda_tromp_0.3a.zip (cc2.x-cc6.2)... if not work try decrease number of blocks (-cb option) or threads (-ct option)

ansemz commented 8 years ago

@chronosek Thank you. it works.

sonicdx10 commented 8 years ago

@chronosek Thank you!