This means that theoretically both manufacturers (Nvidia/AMD) should be equally capable at hashing. In practice however one will have slightly better support for OpenCL than the other.
Roger that. This must have happened while I was out of town. Any ideas on the command set to assign GPUs? The wiki is pretty lacking on how to operate - https://github.com/ethereum/ethash/wiki
@ConradJohnson - To be 100% specific, it's benchmarking eth mining. For various reasons I haven't gotten to actually GPU mine eth on the blockchain locally or on the testnet.
The latest develop branch has support for GPU mining with -G command line option. For some reason that only worked in combination with the -M (benchmark) option. To force it into real blockchain GPU mining, I had to make a change in the source code. And then that for some obscure reason fails after having mined 2 blocks.
--edit--
the very very latest build now has a benchmark ranking built in:
Benchmarking on platform: { "platform": "NVIDIA CUDA", "device": "GeForce GTX 78
0", "version": "OpenCL 1.1 CUDA" }
Preparing DAG...
*** [ 16:39:48 | main ] Spawning anon
WaUsing platform: NVIDIA CUDA
rming upUSING 0: GeForce GTX 780 (OpenCL 1.1 CUDA)
...
Using device: GeForce GTX 780 (OpenCL 1.1 CUDA)
Trial 1... 16951978
Trial 2... 16864597
Trial 3... 16864597
Trial 4... 16858977
Trial 5... 16864597
*** [ 16:40:09 | main ] Stopping anon
*** [ 16:40:09 | anon ] Finishing up worker thread
*** [ 16:40:09 | main ] Stopped anon
min/mean/max: 16858977/16880949/16951978 H/s
inner mean: 16862723 H/s
Phoning home to find world ranking...
Ranked: 10 of all benchmarks.
I guess somebody has to do a CUDA port for me to rank any higher
Also tested on NVidia GTX750Ti 2GB. Somehow runs with CUDA OpenCL 1.2 instead of 1.1 (CUDA 7), but performs really bad. Longer warm up time needed, highly variable results between 350-600 KH/s. What's also kind of odd is that the hash rate often is exactly 2 ^19 = 524288.
Hehe. I sold my R9 for other reasons than speed . It did 15MH/s in the earlier ethash benchmarks.
The basic workflow to get a GPU benchmark on Win64 is as follows:
0. Have a 64-bit Windows and a GPU (NVidia / AMD) 1. Get Visual Studio Express 2013 Desktop 2. Get OpenCL SDK for your card (Intel won't work) 3. Get tar (MingW) and curl for windows and add binary paths to PATH 4. Get git and cmake 5. Checkout latest develop branch 6. run getstuff.bat in cpp-ethereum/extdep. If it doesn't work, fix 3. 7. create cpp-ethereum/build and go there 8. cmake -DETHASHCL=1 -G "Visual Studio 12 2013 Win64" .. 9. open ethereum.sln 10. build eth project 11. launch with eth.exe -G -M
Or simply download the files attached. If for some reasons it tries to benchmark against Intel OpenCL, add --opencl-platform 1 . It's a feature that I just added on my own fork.
@ConradJohnson afaik you can currently only use a single GPU per instance of the client. But there is (work being done on) code that allows you to run a farm of multiple mining clients. That said, I don't see a reason why multiple GPUs per client wouldn't be possible.
I've Nvidia 845M. I tried your test... but it see only my Intel graphic card.. Is it a OpenCL problem ?... It seems that my 845m is not supported by Cuda Nvidia Software... @Genoil Could you help me ?
@pedrosoft the latest develop (attached) has an option to select a different OpenCL platform. Launch it like this:
eth.exe -G -M --opencl-platform 1
Default platform is 0, but in some machines, the GPU OpenCL platform is 1.
@michaelaudoux the temperature heavily depends on OC and fan settings, as well as outside temp, case air flow etc. etc. so it doesn't really make sense to post these. GPU-Z does record 99% load though.
I also noticed that a 4-core CPU is nice to have running alongside your GPU. My home setup is a G1840 dualcore with the gtx750ti, and while mining, the system is completely unusable. so perhaps the bad scores actualy come from a cpu bottleneck. I'll see if I have time to swap out the work/home cards this week.
And lastly a note on NVidia drivers. The latest 350 "GTA V" release has OpenCL 1.2 support. Now on my GTX 780, I also need to warm up longer, and the results are about 10% lower than with the OpenCL 1.1 drivers.
@pedrosoft if you look here it says OpenCL 1.1 is supported on the 840m (couldn't find the 845m page). So as long as you have reasonably up to date graphics drivers, you should be fine.
I wouldn't recommend you to overclock a notebook GPU. It's not not going to give you a lot of extra Hashes before it overheats. But if you understand the risks, go get MSI Afterburner, and follow one these guides
Comments
This means that theoretically both manufacturers (Nvidia/AMD) should be equally capable at hashing. In practice however one will have slightly better support for OpenCL than the other.
Should be able to find it here: https://github.com/ethereum/ethash
Edit: Apologies, I had this confused with the CPP client and cannot confirm whether the miner will be on Windows at this point.
More information on the miner to follow.
Nice, but any AMD R9 based card will beat this. Their OpenCL implementation is simply better.
The latest develop branch has support for GPU mining with -G command line option. For some reason that only worked in combination with the -M (benchmark) option. To force it into real blockchain GPU mining, I had to make a change in the source code. And then that for some obscure reason fails after having mined 2 blocks.
--edit--
the very very latest build now has a benchmark ranking built in:
Benchmarking on platform: { "platform": "NVIDIA CUDA", "device": "GeForce GTX 78 0", "version": "OpenCL 1.1 CUDA" } Preparing DAG... *** [ 16:39:48 | main ] Spawning anon WaUsing platform: NVIDIA CUDA rming upUSING 0: GeForce GTX 780 (OpenCL 1.1 CUDA) ... Using device: GeForce GTX 780 (OpenCL 1.1 CUDA) Trial 1... 16951978 Trial 2... 16864597 Trial 3... 16864597 Trial 4... 16858977 Trial 5... 16864597 *** [ 16:40:09 | main ] Stopping anon *** [ 16:40:09 | anon ] Finishing up worker thread *** [ 16:40:09 | main ] Stopped anon min/mean/max: 16858977/16880949/16951978 H/s inner mean: 16862723 H/s Phoning home to find world ranking... Ranked: 10 of all benchmarks.
I guess somebody has to do a CUDA port for me to rank any higher
--edit--
Found this: http://gav.ethdev.com:3000/
63MH/s..I wonder if that's from a multi-GPU setup or not...
The basic workflow to get a GPU benchmark on Win64 is as follows:
0. Have a 64-bit Windows and a GPU (NVidia / AMD)
1. Get Visual Studio Express 2013 Desktop
2. Get OpenCL SDK for your card (Intel won't work)
3. Get tar (MingW) and curl for windows and add binary paths to PATH
4. Get git and cmake
5. Checkout latest develop branch
6. run getstuff.bat in cpp-ethereum/extdep. If it doesn't work, fix 3.
7. create cpp-ethereum/build and go there
8. cmake -DETHASHCL=1 -G "Visual Studio 12 2013 Win64" ..
9. open ethereum.sln
10. build eth project
11. launch with eth.exe -G -M
Or simply download the files attached. If for some reasons it tries to benchmark against Intel OpenCL, add --opencl-platform 1 . It's a feature that I just added on my own fork.
Might be worth looking into 7970 as well.
eth.exe -G -M --opencl-platform 1
Default platform is 0, but in some machines, the GPU OpenCL platform is 1.
@michaelaudoux the temperature heavily depends on OC and fan settings, as well as outside temp, case air flow etc. etc. so it doesn't really make sense to post these. GPU-Z does record 99% load though.
I also noticed that a 4-core CPU is nice to have running alongside your GPU. My home setup is a G1840 dualcore with the gtx750ti, and while mining, the system is completely unusable. so perhaps the bad scores actualy come from a cpu bottleneck. I'll see if I have time to swap out the work/home cards this week.
And lastly a note on NVidia drivers. The latest 350 "GTA V" release has OpenCL 1.2 support. Now on my GTX 780, I also need to warm up longer, and the results are about 10% lower than with the OpenCL 1.1 drivers.
btw i just found out this client version does not mine in non-benchmark mode. if it does for you (leave out -M parameter), please let me know
this is my benchmark with a NON overclocked geforce 845m