2
$\begingroup$

I installed the CUDA toolkit by directly obeying to the WRI instructions and use MMA 11.1.1.0 as well as the paclet version 10.5.0. While CUDA's deviceQuery sample shows an abbreviated output of 

Device 0: "GeForce GTX 1070"
CUDA Driver Version / Runtime Version          8.0 / 8.0
CUDA Capability Major/Minor version number:    6.1
Total amount of global memory:                 8192 MBytes (8589737984 bytes)
(15) Multiprocessors, (128) CUDA Cores/MP:     1920 CUDA Cores

the MMA SystemInformation[] only show 480 CUDA cores as can be seen here:

enter image description here

Does MMA count different from what Nvidia itself does or did I do something wrong while installing the package. Using both the latest CUDA drivers for macOS and the latest paclet for MMA I am quite sure the problem is not related to deprecated drivers.

Improve this question
$\endgroup$
1
  • $\begingroup$ This is an old post, but I just installed an RTX 3090 and my CUDA Core Count is "Indeterminate", so Mathematica can't use all of the cores since it doesn't know how many I have. I have a feeling this is the case because there are more than 10,000 because if I set batch size to anything over 10k, it sets it back down to 10k in NetTrain. $\endgroup$
    – Gregory Klopper
    Commented Apr 20, 2021 at 3:42

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.