25

What are the minimum compute capability requirements for:

  1. CUDA 5.5?
  2. CUDA 6.0?
  3. CUDA 6.5?
talonmies
  • 68,743
  • 34
  • 184
  • 258
Max
  • 391
  • 2
  • 6
  • 14
  • 1
    Related: https://stackoverflow.com/questions/30820513/what-is-the-correct-version-of-cuda-for-my-nvidia-driver/30820690#30820690 – mic May 20 '20 at 04:36

2 Answers2

40
CUDA VERSION   Min CC   Deprecated CC  Default CC  Max CC
5.5 (and prior) 1.0       N/A             1.0
6.0             1.0       1.0             1.0
6.5             1.1       1.x             2.0
7.x             2.0       N/A             2.0
8.0             2.0       2.x             2.0      6.2
9.x             3.0       N/A             3.0      7.0
10.x            3.0       N/A             3.0      7.5 (3.0 deprecated in 10.2)
11.x            3.5       3.x,5.0         5.2      8.6 (11.0:8.0, 11.1:8.6)
(CUDA 11.5 still "supports" cc3.5 devices; the R495 driver in CUDA 11.5 installer does not)
  • Min CC = minimum compute capability that can be specified to nvcc
  • Deprecated CC = If you specify this CC, you will get a deprecation message, but compile should still proceed.
  • Default CC = The architecture that will be targetted if no -arch or -gencode switches are used
  • Max CC = The highest compute capability you can specify on the compile command line via arch switches (compute_XY, sm_XY)
Robert Crovella
  • 131,712
  • 9
  • 184
  • 228
  • Thanks! Actually I had some problems installing CUDA 6 on my GPU with CC 1.2 and thought that it's not supported. But it looks like the problem is somewhere else. – Max Mar 09 '15 at 08:33
  • 4
    Where did you find this information? Is there any official version, that is updated for CUDA 9.0? – Alexander Pacha Jan 03 '18 at 08:40
  • Actually for CUDA 9.0: `GPU card with CUDA Compute Capability 3.0 or higher for building from source and 3.5 or higher for our binaries. ` as stated at https://www.tensorflow.org/versions/r1.7/install/install_linux – Andrzej Piasecki May 25 '18 at 12:06
  • @AndrzejPiasecki that is a requirement specific to Tensorflow (and it may change in the future), not a general CUDA requirement for use of CUDA 9.0. This question is not about Tensorflow. – Robert Crovella May 25 '18 at 13:03
  • @RobertCrovella I have a GTX670M which has a compute capacity of 2.1. So I've installed on my Windows 10 CUDA 8.0 with its patch. Then I installed the right version of TensorFlow and CUDNN (according to this version 8.0 of CUDA) : https://www.tensorflow.org/install/source_windows#gpu (V1.4.0) and https://developer.nvidia.com/rdp/cudnn-archive (V6.0). However, at runtime, my Python interpreter tells that my GTX670M has a CC of 2.1 (right) and that... ... ... my CUDA 8.0 has a minimum CC of .. guess what ... 3.0. :( Why? According to you, Wikipedia, and another SO question, CUDA 8.0 = 2.0 at min – JarsOfJam-Scheduler Aug 02 '19 at 07:58
  • 1
    @JarsOfJam-Scheduler Tensorflow and CUDNN have a minimum cc of 3.0. Not CUDA 8. See [here](https://stackoverflow.com/questions/38542763/how-can-i-make-tensorflow-run-on-a-gpu-with-capability-2-0/38543201#38543201). – Robert Crovella Aug 02 '19 at 11:15
-3

Actually for CUDA 10.0:

  • with Tensorflow: The minimum required Cuda capability is 3.5.
  • with Pytorch: Found GPU0 GeForce GTX 650 which is of cuda capability 3.0. PyTorch no longer supports this GPU because it is too old.
Liz
  • 19
  • 4