We are already using NVIDIA RTX A5000 GPU in one of our workstation with CUDA 11.6 version.
We are planning to have one more workstation with GeForce RTX 4060 Ti GPU.
I was trying to understand if 4060 Ti GPU’s CUDA version is compatible with A5000’s. Then I came to know about Compute Capability of GPUs from here. And I got compute capability vs CUDA support from here. Which shows this.
As per my understanding,
NVIDIA RTX A5000 has compute capability of 8.6 (Ampere architecture) and hence supports CUDA 11.0 to 12.2
Nvidia GeForce RTX 4060 Ti has compute capability of 8.9 (Ada Lovelace architecture) and hence supports CUDA versions 11.8 to 12.2. (No support of 11.0 to 11.7.1).
Is my understanding correct?
Thanks & Regards,
If you look at the the line in the table for Cuda 11.0, it ends in 8.0. Your A5000 is 8.6 and so needs a minimum of 11.1.
The 4060 Ti understanding is correct.
@rs277 Thank you for correcting.
Just to be sure about my understanding, I will rewrite,
RTX A5000 supports = 11.1 to 12.2 (as of today)
RTX 4060 TI supports = 11.9 to 12.2 (as of today)
Also, is there any better sheet of CUDA vs Compute capability? The Table which I have attached is slightly confusing for many actually.
Perhaps what is confusing, is that the Toolkit SDK numbers are not a continuous sequence - you can find the list here. Compute Capability is separate from the Toolkit SDK number and indicates a hardware version.
There is no 11.9 Toolkit, so the first one to support it is 11.8 and another page you may find useful, with the same chart, is here.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.