Samples from 9.1 do not run

Accelerated Computing CUDA CUDA Setup and Installation
1

Setup:

Ubuntu 18.04
±----------------------------------------------------------------------------+
| NVIDIA-SMI 390.116 Driver Version: 390.116 |
|-------------------------------±---------------------±---------------------+
| 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 GTX 675M Off | 00000000:01:00.0 N/A | N/A |
| 0% 64C P1 N/A / N/A | 582MiB / 1977MiB | N/A Default |
±------------------------------±---------------------±---------------------+
| 1 GeForce GTX 675M Off | 00000000:07:00.0 N/A | N/A |
| 0% 56C P12 N/A / N/A | 56MiB / 1985MiB | N/A Default |
±------------------------------±---------------------±---------------------+

±----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 Not Supported |
| 1 Not Supported |
±----------------------------------------------------------------------------+

Installed:

sudo sh ./cuda_9.1.85_387.26_linux.run --verbose --silent --toolkit --override

sudo sh ./cuda_9.1.85.1_linux.run --silent --accept-eula

sudo sh ./cuda_9.1.85.2_linux.run --silent --accept-eula

sudo sh ./cuda_9.1.85.3_linux.run --silent --accept-eula

gcc --version

gcc (Ubuntu 6.5.0-2ubuntu1~18.04) 6.5.0 20181026

cd ~/Projects/cuda-testing/samples/

make -j4

All samples build without error but none run actually run with errors like :

cd bin/x86_64/linux/release

./simpleSurfaceWrite

simpleSurfaceWrite starting…
MapSMtoCores for SM 2.1 is undefined. Default to use 64 Cores/SM
MapSMtoCores for SM 2.1 is undefined. Default to use 64 Cores/SM
GPU Device 0: “GeForce GTX 675M” with compute capability 2.1

CUDA device [GeForce GTX 675M] has 8 Multi-Processors, SM 2.1
Loaded ‘lena_bw.pgm’, 512 x 512 pixels
CUDA error at simpleSurfaceWrite.cu:227 code=37(cudaErrorInvalidSurface) “cudaBindSurfaceToArray(outputSurface, cuArray)”

I have seen a number of other posts in this forum regarding these errors which all suggest my setup is still not correct.

Any ideas appreciated.

2

CUDA 9.x does not support devices of compute capability 2.1

Those devices are supported by CUDA 8.0 and prior.

3

Thank you for your assistance and experience.

I am assuming I should remove the 9.1 version prior to installing the 8.0 version or will they happily co-exist?

Also is there a handy chart that indicates the CUDA version to Compute capability compatibility?

Much appreciated!

4

They can coexist. Follow the instructions in the linux install guide carefully, including section 7, to point to the correct version of CUDA.

I believe your existing driver should be OK for usage with a Fermi device, with a few corner-case exceptions (e.g. usage of cuda-memcheck). If you want full capability with CUDA 8, you should install the driver that comes with the CUDA 8 toolkit.

5

I ended up removing 9.1 and installing 8.0 as documented and all the samples now work.

Thanks people!