Closed nohboogy closed 5 years ago
same question,need some help
Did you achieve to solve the problem? I have the same problem with the RTX2080 and ubuntu 18.04 and I really need to solve it. Thanks
@zhangqijun @pablovicentem I solved this problem with reinstalling CUDA 10.0. CUDA 9.0 dosen't support rtx 2080 even though it looks like we could install many applications with it.
I am with CUDA 10.0, still suffers this issue:
nvcc fatal : Unsupported gpu architecture 'compute_75'
when install Torch on Ubuntu 16.04.6 LTS
Fri Mar 29 10:12:20 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.78 Driver Version: 410.78 CUDA Version: 10.0 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce RTX 208... Off | 00000000:19:00.0 Off | N/A |
| 27% 29C P8 11W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 GeForce RTX 208... Off | 00000000:1A:00.0 Off | N/A |
| 27% 31C P8 10W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 2 GeForce RTX 208... Off | 00000000:67:00.0 Off | N/A |
| 27% 35C P8 21W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 3 GeForce RTX 208... Off | 00000000:68:00.0 On | N/A |
| 27% 34C P8 21W / 250W | 71MiB / 10986MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 3 1055 G /usr/lib/xorg/Xorg 59MiB |
+-----------------------------------------------------------------------------+
I am with CUDA 10.0, still suffers this issue:
nvcc fatal : Unsupported gpu architecture 'compute_75'
when install Torch onUbuntu 16.04.6 LTS
Fri Mar 29 10:12:20 2019 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 410.78 Driver Version: 410.78 CUDA Version: 10.0 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce RTX 208... Off | 00000000:19:00.0 Off | N/A | | 27% 29C P8 11W / 250W | 11MiB / 10989MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce RTX 208... Off | 00000000:1A:00.0 Off | N/A | | 27% 31C P8 10W / 250W | 11MiB / 10989MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 2 GeForce RTX 208... Off | 00000000:67:00.0 Off | N/A | | 27% 35C P8 21W / 250W | 11MiB / 10989MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 3 GeForce RTX 208... Off | 00000000:68:00.0 On | N/A | | 27% 34C P8 21W / 250W | 71MiB / 10986MiB | 0% Default | +-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+ | Processes: GPU Memory | | GPU PID Type Process name Usage | |=============================================================================| | 3 1055 G /usr/lib/xorg/Xorg 59MiB | +-----------------------------------------------------------------------------+
Hi, have you solved the torch installition problem ? I have the same problem and cannot solve it. Thanks!
I am with CUDA 10.0, still suffers this issue:
nvcc fatal : Unsupported gpu architecture 'compute_75'
when install Torch onUbuntu 16.04.6 LTS
Fri Mar 29 10:12:20 2019 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 410.78 Driver Version: 410.78 CUDA Version: 10.0 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 GeForce RTX 208... Off | 00000000:19:00.0 Off | N/A | | 27% 29C P8 11W / 250W | 11MiB / 10989MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 1 GeForce RTX 208... Off | 00000000:1A:00.0 Off | N/A | | 27% 31C P8 10W / 250W | 11MiB / 10989MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 2 GeForce RTX 208... Off | 00000000:67:00.0 Off | N/A | | 27% 35C P8 21W / 250W | 11MiB / 10989MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 3 GeForce RTX 208... Off | 00000000:68:00.0 On | N/A | | 27% 34C P8 21W / 250W | 71MiB / 10986MiB | 0% Default | +-------------------------------+----------------------+----------------------++-----------------------------------------------------------------------------+ | Processes: GPU Memory | | GPU PID Type Process name Usage | |=============================================================================| | 3 1055 G /usr/lib/xorg/Xorg 59MiB | +-----------------------------------------------------------------------------+
Hi, have you solved the torch installition problem ? I have the same problem and cannot solve it. Thanks!
I am sorry, I didnt get a solution for this.
@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun
Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10.
The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit.
So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the samenvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:
export TORCH_CUDA_ARCH_LIST="7.0"
Just before running ./install.sh from the torch directory. Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.
@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10. The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit. So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same
nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:export TORCH_CUDA_ARCH_LIST="7.0"
Just before running ./install.sh from the torch directory. Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.
Thanks! It works.
@RishengZhou By the way, another approach that worked for me without any tweaks needed, on a machine with Ubuntu 18.04 and CUDA 10.1 is simply using the following git repo instead of the official one: https://github.com/nagadomi/distro
@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10. The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit. So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same
nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:export TORCH_CUDA_ARCH_LIST="7.0"
Just before running ./install.sh from the torch directory. Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.
Thanks ! My configuration is a RTX 2080 CUDA 7.5, Ubuntu 16.04,
export TORCH_CUDA_ARCH_LIST="5.0"
Succeed
@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10. The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit. So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same
nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:export TORCH_CUDA_ARCH_LIST="7.0"
Just before running ./install.sh from the torch directory. Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.
Thanks! I met the issue when i install detectron2. It works! You are amazing!
I had 2 different versions of CUDA, but solved it by removing everything and re-installing 10.1
I'm on Ubuntu 18.04 with 2080
Highly recommend following the instructions in this blog, worked perfectly for me: https://medium.com/@exesse/cuda-10-1-installation-on-ubuntu-18-04-lts-d04f89287130
@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10. The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit. So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same
nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:export TORCH_CUDA_ARCH_LIST="7.0"
Just before running ./install.sh from the torch directory. Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.
I follow this solution but seem it does not work for me. I have tried other lower versions, but still not working.
While trying to install torch, same error comes out.
I think it is because I'm using RTX graphic driver, but I cannot find any way to install torch now. Anyone can solve this problem?