The CudaMiner for OS X 2014-02-28 is now available.
For binaries and instructions see CudaMiner for OS X
Prerequisites
- Mac OS X 10.7 Lion, Mac OS X 10.8 Mountain Lion, or OS X 10.9 Mavericks
- Compute 2.x or Higher NVIDIA GPU (Compute 1.x GPUs are not supported), see https://developer.nvidia.com/cuda-gpus for a listing of supported devices
- CUDA Toolkit and Drivers 5.5
Direct Download Links
- OS X 10.9 Mavericks (2014-02-28): Download
- OS X 10.7 Lion and 10.8 Mountain Lion (2014-02-28): Download
it looks like there is a problem that has been introduced in the versions after the feb-18 version that does not behave like this, and to me it looks like it does not release the GPU properly on termination so that GPU intensive applications are not able to use it after (they just crash) and you have to reboot the system.
You might post this over on the original thread: https://bitcointalk.org/index.php?topic=167229.0. You might request a cleaner way to exit (like cgminer).
Getting out of memory errors.
iMac (Late 2012)
10.8.5
Seen anything like this before?
[2014-03-31 09:44:35] Stratum detected new block
[2014-03-31 09:44:36] GPU #0: cudaError 2 (out of memory) calling ‘cudaFree(0)’ (salsa_kernel.cu line 240)
[2014-03-31 09:44:36] GPU #0: GeForce GT 640M with compute capability 3.0
[2014-03-31 09:44:36] GPU #0: cudaError 2 (out of memory) calling ‘cudaDeviceSetCacheConfig(kernel->cache_config())’ (salsa_kernel.cu line 423)
[2014-03-31 09:44:36] GPU #0: cudaError 2 (out of memory) calling ‘cudaDeviceSetSharedMemConfig(kernel->shared_mem_config())’ (salsa_kernel.cu line 424)
[2014-03-31 09:44:36] GPU #0: interactive: 1, tex-cache: 0 , single-alloc: 0
[2014-03-31 09:44:36] GPU #0: 32 hashes / 0.1 MB per warp.
[2014-03-31 09:44:36] GPU #0: Performing auto-tuning (Patience…)
[2014-03-31 09:44:36] GPU #0: cudaError 2 (out of memory) calling ‘cudaMalloc((void **) &d_idata, mem_size)’ (salsa_kernel.cu line 594)
…
lots more of the same calling different cudaBlahBlah functions
…
[2014-03-31 09:44:36] GPU #0: cudaError 2 (out of memory) calling ‘cudaEventRecord(context_serialize[1][thr_id])’ (salsa_kernel.cu line 303)
[2014-03-31 09:44:36] GPU #0: cudaError 2 (out of memory) calling ‘cudaStreamWaitEvent(context_streams[stream][thr_id], context_serialize[(stream+1)&1][thr_id], 0)’ (salsa_kernel.cu line 946)
[1] 16531 floating point exception
exit
Hi Dan, I’ve seen this on my machine (650M) if I set the configuration values too high (requiring too much GPU RAM). Are you manually setting the configuration or allowing it to auto tune?
I let it auto-tune. However, it looks like maybe it was because a previous mining program did not release the resources properly. After a reboot, it seems to work fine as long as cudaminer is the first miner run.
Ah, yes. Both Photoshop and Pixelmater are also guilty of this (I can’t run the miner and Photoshop at the same time).
Hello john,
I have downloaded your OS X 10.9 Mavericks (2014-02-28) files….
When i do a
./cudaminer -h
dyld: Library not loaded: /Developer/NVIDIA/CUDA-5.5/lib/libcudart.5.5.dylib
Referenced from: /Users/edi3/Downloads/CudaMiner/./cudaminer
Reason: image not found
Trace/BPT trap: 5
Do you know what is the problem?
Thanks
To use the CUDA 5.5 version, you need to have the 5.5 version of the CUDA Toolkit installed: https://developer.nvidia.com/cuda-toolkit-55-archive.
Alternatively, if you have the 6.0 version installed (or 6.5, just replace the version), you can create a symlink so that the program can still find the files it needs. Open Terminal (Applications > Utilities) and do the following:
sudo ln -s /Developer/NVIDIA/CUDA-6.0 /Developer/NVIDIA/CUDA-5.5
You will be prompted for your password.
The above command will create a “fake” CUDA-5.5 directory that is actually a pointer to CUDA-6.0 (or 6.5 if you have that version instead).
hello john,
When i launch cudaminer:
./cudaminer -o stratum+tcp://am01.eu.trademybit.com:3330 -O sshiva1172.2:123 -a scrypt
I have an unknown message:
dyld: Library not loaded: /Developer/NVIDIA/CUDA-5.5/lib/libcudart.5.5.dylib
Referenced from: /Users/edi3/Downloads/CudaMiner/./cudaminer
Reason: image not found
Trace/BPT trap: 5
To use the CUDA 5.5 version, you need to have the 5.5 version of the CUDA Toolkit installed: https://developer.nvidia.com/cuda-toolkit-55-archive.
Alternatively, if you have the 6.0 version installed (or 6.5, just replace the version), you can create a symlink so that the program can still find the files it needs. Open Terminal (Applications > Utilities) and do the following:
sudo ln -s /Developer/NVIDIA/CUDA-6.0 /Developer/NVIDIA/CUDA-5.5
You will be prompted for your password.
The above command will create a “fake” CUDA-5.5 directory that is actually a pointer to CUDA-6.0 (or 6.5 if you have that version instead).