Closed wrt54gl closed 6 years ago
Please report output of command
ethminer -U --list-devices
Listing CUDA devices.
FORMAT: [deviceID] deviceName
[0] GeForce GTX 1070
Compute version: 6.1
cudaDeviceProp::totalGlobalMem: 8513978368
Pci: 0000:07:00
[1] GeForce GTX 1070
Compute version: 6.1
cudaDeviceProp::totalGlobalMem: 8513978368
Pci: 0000:0e:00
[2] P104-100
Compute version: 6.1
cudaDeviceProp::totalGlobalMem: 4238737408
Pci: 0000:10:00
[3] GeForce GTX 1070
Compute version: 6.1
cudaDeviceProp::totalGlobalMem: 8513978368
Pci: 0000:11:00
Have you tried to launch ethminer on each single device ?
First on device 0, then for device 1 and so on ?
$ ethminer --cuda-devices 3 -U stratum+tcp://user.rig:x@us-east.ethash-hub.miningpoolhub.com:17020
m 06:05:55 ethminer ethminer 0.16.0.dev0+commit.c33e009c
m 06:05:55 ethminer Build: linux/release
cu 06:05:55 ethminer Using grid size: 8,192, block size: 128
cu 06:05:55 ethminer Found suitable CUDA device [GeForce GTX 1070] with 8,513,978,368 bytes of GPU memory
Segmentation fault (core dumped)
This happens on all 4 devices. Incidentally it worked on device 3 the first time but not after that.
There must be something with your CUDA setup. Which version of cuda and wich NVIDIA driver you're using ?
I installed cuda_9.2.88_396.26_linux
Tried with:
ethminer --opencl-platform 1 -X stratum+tcp://user.rig:x@us-east.ethash-hub.miningpoolhub.com:17020
It works but does not start the last 2 cards. It seems like the Nvidia (cuda) cards are all running but not all the AMD(rocm opencl)
I guess it is a Heisenbug! It will run fine in GDB but not otherwise! For now I can work around by starting it with gdb. Runs fine and I get the same hash rate!
tried running ethminer - compiled and/or the latest release and it segfaults immediately:
~$ ethminer -U stratum+tcp://uname.rigname:x@us-east.ethash-hub.miningpoolhub.com:17020 m 13:37:07 ethminer ethminer 0.16.0.dev0+commit.c33e009c m 13:37:07 ethminer Build: linux/release cu 13:37:07 ethminer Using grid size: 8,192, block size: 128 Segmentation fault (core dumped)
This only happens with the -U option and works fine with my AMD cards on the same machine and the same pool. Also seems to be somewhat intermittent as it will sometimes (about 1 time in 20) work.
I tried running it in gdb and it won't seg fault. It is on ubuntu 16.04 with kernel 4.13. I have 3 GTX1070 and 1 P104-100 cards.
For now I can run this rig by using opencl but the performance is not as good.