latest source on my github should compile on Linux without having to tamper with anything. Except I had to make a tiny change to some header file of CUDA 7, because the compiler complained about a "comment in a comment".
Thanks Genoil ! I successfully compile on ubuntu My test with the GTX 970 : - Default ethminer opencl : 17,6MH/s - Genoil ethminer opencl : 18,0MH/s - Genoil ethminer cuda : 18,4MH/s
it isn't a big gap, but it's better ! thanks again ;-)
@Brips you will get a bit more if you add the flags --gpu-workgroup-size 128 --gpu-mining-buffers 4 --gpu-batch-size 19
you add these parameters on the command line ? like this : ./ethminer -U -M --gpu-workgroup-size 128 --gpu-mining-buffers 4 --gpu-batch-size 19 ? i test it but i have a little less 18,3MH/s with cuda
Yes then that's what it is. Had hoped for a bit more on 9xx but the only way to possibly get it faster is to completely turn the implementation of the algorithm upside down, and then still it's a bet...
... [100%] Building CXX object ethminer/CMakeFiles/ethminer.dir/main.cpp.o In file included from /home/o0ragman0o/genoil/cpp-ethereum/ethminer/main.cpp:31:0: /home/o0ragman0o/genoil/cpp-ethereum/ethminer/MinerAux.h: In member function ‘void MinerCLI::execute()’: /home/o0ragman0o/genoil/cpp-ethereum/ethminer/MinerAux.h:304:4: error: ‘setKernelParameters’ is not a member of ‘dev::eth::Ethash::CUDAMiner {aka dev::eth::Ethash::CPUMiner}’ ProofOfWork::CUDAMiner::setKernelParameters(gpuMiningBuffers, gpuBatchSize, gpuWorkgroupSize); ^ make[2]: *** [ethminer/CMakeFiles/ethminer.dir/main.cpp.o] Error 1 make[1]: *** [ethminer/CMakeFiles/ethminer.dir/all] Error 2 make: *** [all] Error 2
After some extensive testing, I came to the conclusion that my CUDA miner is seriously broken. I had been developing with a very specific Proof of Work and difficulty, and ironically, just for that exact problem it produces the same results as the opencl kernel. So I though I was on the right track. For other Proof of Work though, it occasionally prodcues the same results, but most of the times completely different.
[ 36%] Building CXX object libdevcore/CMakeFiles/devcore.dir/Guards.cpp.o /home/darryl/genoil/cpp-ethereum/libethash-cu/ethash_cu_miner_kernel.cu(196): error: identifier "__ldg" is undefined
1 error detected in the compilation of "/tmp/tmpxft_00003fc5_00000000-23_ethash_cu_miner_kernel.compute_20.cpp1.ii". CMake Error at ethash-cu_generated_ethash_cu_miner_kernel.cu.o.cmake:264 (message): Error generating file /home/darryl/genoil/cpp-ethereum/build/libethash-cu/CMakeFiles/ethash-cu.dir//./ethash-cu_generated_ethash_cu_miner_kernel.cu.o
...And after removing the offending "/*" from /usr/local/cuda/include/CL/cl_gl_ext.h:44:4: error: "/*" within comment [-Werror=comment] it compiled and works!
It's not benchmarking as good as it was on Win 7. -U Only getting 6.6 on a benchmark whereas -G still hitting 6.9. But on actual mining -U is giving me a solid 5Mhs whereas -G only gets 2.4MHs.
Do you have any idea why mining hashrate is slower than benchmarks?
@o0ragman0o it should be able to do 7 (GTX750 non-Ti) as before, but I've made a change in how to optimize thread occupancy. It would be great if you could fing the best values for __launch_bounds__ on line 284 of the kernel. You may try (64, 7), (64, 8), (128, 6), (128, 7). When launching, you can never specify a value for --gpu-workgroup-size larger than the first argument. It defaults to 64.
The mining hash rate is lower because of some additional code overhead, but it seems quite high on your setup, esp. on OpenCL.
@Genoil, Seems there's a memory leak or something in your CUDA code. After running overnight the workpackages throw: Error CUDA mining: an illegal memory access was encountered
@o0ragman0othe current version is untested on anything but my own GTX780. I tried to speed up the kernel by completely getting rid of shared memory and replace it with warp shuffles. Took me ages to get it working, only to find out it wasn't any faster than the shared memory approach. And now it doesn't even run on GTX750 . I'll try to post a version today that allows you to choose between shuffle and shared. Was planning on that anyway.
scrap that, you were right, it also failed on my GTX780. most recent version should run. btw it won't accept --gpu-workgroup-size 128 for the moment, just 64.
I haven't been able to mine for 10 ten days so I decided to rebuild, can you provide some info on building your cuda miner?
I've downloaded ethereum sln + your sln, but I am build errors. The right ethash doesn't seem to be included: Error 15 error C2039: 'startCUDA' : is not a member of 'dev::eth::GenericFarm' C:\cpp-ethereum\ethminer\MinerAux.h 404 1 eth
Any tips for getting an nvidia 660ti mining on ubuntu15.04? The gpu's max memory allocation is only 500mb. Can i fit the DAG in there in chunks? I'm using the ubuntu ppa version of cpp-ethereum. Perhaps i need to get on the develop branch & build myself?
Heres the error: 0 H/s
[email protected]:~$ ethminer -G -F http://127.0.0.1:8545 [OPENCL]:Found suitable OpenCL device [GeForce GTX 660 Ti] with 2147155968 bytes of GPU memory miner 23:35:04|ethminer Getting work package...
ℹ 23:35:09|ethminer Loading full DAG of seedhash: #290decd9… miner 23:35:10|ethminer Getting work package... miner 23:35:10|ethminer Grabbing DAG for #5a46dd85… ℹ 23:35:14|ethminer Full DAG loaded miner 23:35:21|ethminer Got work package: miner 23:35:21|ethminer Header-hash: c21c407788fcbfac56d7157526e7a6455ec69e265ec74fe4d2d60f0803c1ac2c miner 23:35:21|ethminer Seedhash: 5a46dd85298b0beff65ccf3006c4b5d2f7fa6ca8a5885a7f2132714fe7a48602 miner 23:35:21|ethminer Target: 000000002bf83e1cdec8746c78083e1637e163db9a5e141cd3a72a4c5c6c9216 miner 23:35:21|ethminer Mining on PoWhash #c21c4077… : 0 H/s = 0 hashes / 0.5 s miner 23:35:21|ethminer Got work package: miner 23:35:21|ethminer Header-hash: 36b08af8fcfc0c408110b59cd1102cfa781c5b2c26a80e1ee382f0f869e0acd3 miner 23:35:21|ethminer Seedhash: 5a46dd85298b0beff65ccf3006c4b5d2f7fa6ca8a5885a7f2132714fe7a48602 miner 23:35:21|ethminer Target: 000000002bf2bfc4ea19045deb590d852d521f098972df68b7edcaf4b110ccb4 miner 23:35:22|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 0.5 s miner 23:35:22|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 1 s miner 23:35:23|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 1.501 s miner 23:35:23|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 2.002 s miner 23:35:24|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 2.503 s miner 23:35:24|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 3.004 s miner 23:35:25|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 3.504 s
Output from geth gpuinfo:
============================================= ============ OpenCL Device Info ============= ============================================= Platform id 0 Platform Name NVIDIA CUDA Platform Vendor NVIDIA Corporation Platform Version OpenCL 1.1 CUDA 7.0.35 Platform Extensions cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_copy_opts Platform Profile FULL_PROFILE
Device OpenCL id 0 Device id for mining 0 Device Name GeForce GTX 660 Ti Vendor NVIDIA Corporation Version OpenCL 1.1 CUDA Driver version 346.59 Address bits 32 Max clock freq 980 Global mem size 2147155968 Max constant buffer size 65536 Max mem alloc size 536788992 Max compute units 7 Max work group size 1024 Max work item sizes [1024 1024 64] ============================================= Found 1 devices. Benchmark first GPU: geth gpubench 0 Mine using all GPUs: geth --minegpu 0 You may also need to enable chunking: --gpuchunks
[OPENCL]:Created one big buffer for the DAG [OPENCL]:Loading single big chunk kernels [OPENCL]:Creating buffer for header. [OPENCL]:Mapping one big chunk. [OPENCL]:Creating mining buffer 0 [OPENCL]:Creating mining buffer 1 Trial 1... 9087658 Trial 2... 9175040 Trial 3... 9087658 Trial 4... 9175040 Trial 5... 9087658 min/mean/max: 9087658/9122610/9175040 H/s inner mean: 3058346 H/s
but actual mining is pretty disappointing:
ℹ 09:32:44|ethminer Mining on PoWhash #690a1a93… : 2441468 H/s = 18350080 hashes / 7.516 s
This may be a stupid question, but what do I do after making cpp-ethereum? There isn't a folder labeled ethminer on github anymore, so I'm confused what command I execute to run my miner. I'm running Arch linux if it matters.
Comments
I successfully compile on ubuntu
My test with the GTX 970 :
- Default ethminer opencl : 17,6MH/s
- Genoil ethminer opencl : 18,0MH/s
- Genoil ethminer cuda : 18,4MH/s
it isn't a big gap, but it's better ! thanks again ;-)
./ethminer -U -M --gpu-workgroup-size 128 --gpu-mining-buffers 4 --gpu-batch-size 19 ?
i test it but i have a little less 18,3MH/s with cuda
I'm afraid it's back to the drawing table...
/usr/local/cuda/include/CL/cl_gl_ext.h:44:4: error: "/*" within comment [-Werror=comment]it compiled and works!It's not benchmarking as good as it was on Win 7. -U Only getting 6.6 on a benchmark whereas -G still hitting 6.9. But on actual mining -U is giving me a solid 5Mhs whereas -G only gets 2.4MHs.
Do you have any idea why mining hashrate is slower than benchmarks?
The mining hash rate is lower because of some additional code overhead, but it seems quite high on your setup, esp. on OpenCL.
Error CUDA mining: an illegal memory access was encounteredthe current version is untested on anything but my own GTX780. I tried to speed up the kernel by completely getting rid of shared memory and replace it with warp shuffles. Took me ages to get it working, only to find out it wasn't any faster than the shared memory approach.And now it doesn't even run on GTX750
. I'll try to post a version today that allows you to choose between shuffle and shared. Was planning on that anyway. scrap that, you were right, it also failed on my GTX780. most recent version should run. btw it won't accept --gpu-workgroup-size 128 for the moment, just 64.
Up from 6.9
Benchmarking on platform: { "platform": "CUDA 7.0", "device": "GeForce GTX 750", "version": "Compute 5.0" }Preparing DAG...
Warming up...
ℹ 11:43:37|cudaminer0 workLoop 0 #00000000… #00000000…
ℹ 11:43:37|cudaminer0 Initialising miner...
Using device: GeForce GTX 750(5.0)
Trial 1... 8388608
Trial 2... 8388608
Trial 3... 8475989
Trial 4... 8388608
Trial 5... 8388608
min/mean/max: 8388608/8406084/8475989 H/s
I've downloaded ethereum sln + your sln, but I am build errors.
The right ethash doesn't seem to be included:
Error 15 error C2039: 'startCUDA' : is not a member of 'dev::eth::GenericFarm' C:\cpp-ethereum\ethminer\MinerAux.h 404 1 eth
Heres the error: 0 H/s
[email protected]:~$ ethminer -G -F http://127.0.0.1:8545
[OPENCL]:Found suitable OpenCL device [GeForce GTX 660 Ti] with 2147155968 bytes of GPU memory
miner 23:35:04|ethminer Getting work package...
ℹ 23:35:09|ethminer Loading full DAG of seedhash: #290decd9…
miner 23:35:10|ethminer Getting work package...
miner 23:35:10|ethminer Grabbing DAG for #5a46dd85…
ℹ 23:35:14|ethminer Full DAG loaded
miner 23:35:21|ethminer Got work package:
miner 23:35:21|ethminer Header-hash: c21c407788fcbfac56d7157526e7a6455ec69e265ec74fe4d2d60f0803c1ac2c
miner 23:35:21|ethminer Seedhash: 5a46dd85298b0beff65ccf3006c4b5d2f7fa6ca8a5885a7f2132714fe7a48602
miner 23:35:21|ethminer Target: 000000002bf83e1cdec8746c78083e1637e163db9a5e141cd3a72a4c5c6c9216
miner 23:35:21|ethminer Mining on PoWhash #c21c4077… : 0 H/s = 0 hashes / 0.5 s
miner 23:35:21|ethminer Got work package:
miner 23:35:21|ethminer Header-hash: 36b08af8fcfc0c408110b59cd1102cfa781c5b2c26a80e1ee382f0f869e0acd3
miner 23:35:21|ethminer Seedhash: 5a46dd85298b0beff65ccf3006c4b5d2f7fa6ca8a5885a7f2132714fe7a48602
miner 23:35:21|ethminer Target: 000000002bf2bfc4ea19045deb590d852d521f098972df68b7edcaf4b110ccb4
miner 23:35:22|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 0.5 s
miner 23:35:22|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 1 s
miner 23:35:23|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 1.501 s
miner 23:35:23|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 2.002 s
miner 23:35:24|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 2.503 s
miner 23:35:24|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 3.004 s
miner 23:35:25|ethminer Mining on PoWhash #36b08af8… : 0 H/s = 0 hashes / 3.504 s
Output from geth gpuinfo:
=============================================
============ OpenCL Device Info =============
=============================================
Platform id 0
Platform Name NVIDIA CUDA
Platform Vendor NVIDIA Corporation
Platform Version OpenCL 1.1 CUDA 7.0.35
Platform Extensions cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_copy_opts
Platform Profile FULL_PROFILE
Device OpenCL id 0
Device id for mining 0
Device Name GeForce GTX 660 Ti
Vendor NVIDIA Corporation
Version OpenCL 1.1 CUDA
Driver version 346.59
Address bits 32
Max clock freq 980
Global mem size 2147155968
Max constant buffer size 65536
Max mem alloc size 536788992
Max compute units 7
Max work group size 1024
Max work item sizes [1024 1024 64]
=============================================
Found 1 devices. Benchmark first GPU: geth gpubench 0
Mine using all GPUs: geth --minegpu 0
You may also need to enable chunking: --gpuchunks
Trial 1... 11097429
Trial 2... 11097429
Trial 3... 11097429
Trial 4... 11010048
Trial 5... 11097429
min/mean/max: 11010048/11079952/11097429 H/s
inner mean: 11097429 H/s
And I picked it up for less than 100 bucks! That means it's as price and power efficient as the Titans. Actual mining is at 6.1Mhs
I tried OpenCL benchtest and mining also but got...
Thanks