Hi,
I working using Google Cloud Instance with GPU K80 and Ubuntu 16.04.
But I have one question, what is the correct Cuda version, 9.0 , 9.2 or 10 for this hardware ?
in this link you see more information:
Create a VM with attached GPUs | Compute Engine Documentation | Google Cloud
Linux instances:
R384 branch: NVIDIA 384.111 driver or greater
R390 branch: Not yet available
You can use 9.0, 9.2, or 10.0 with K80
You should use an appropriate GPU driver depending on which CUDA toolkit version you are using.
A 410.xx driver should be fine for all CUDA versions.
A 396.xx driver should be OK for CUDA 9.0 or CUDA 9.2
a 384.xx driver will support CUDA 9.0
See table 1 here:
[url]https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#major-components[/url]
Can I use Cuda version 10.1 with GPU Tesla K80?
yes
will the latest CUDA 11 work with Tesla K80?
From the CUDA 11 release notes (1. CUDA 12.3 Update 1 Release Notes — Release Notes 12.3 documentation)
Support for Kepler sm_30 and sm_32 architecture based products is dropped.
Support for the following compute capabilities are deprecated in the CUDA Toolkit:
sm_35 (Kepler)
sm_37 (Kepler)
sm_50 (Maxwell)
The Tesla K80 has compute capability 3.7, i.e. sm_37. It is supported by CUDA 11, but that support is deprecated which means it will likely be removed in the next major CUDA release.
You can look at this table called CUDA Toolkit and Compatible Driver Versions.
What about the Cuda Version 9.1? Is it supported by the K80?
sudo apt-get install -y cuda-compat-12-1
this will give you cuda 12 support for the old k80’s
This will give you the ability to run codes that are already compiled using older versions of CUDA (in some cases). It will not cause the nvcc in CUDA 12 to support e.g. -arch=sm_37.