Mixing graphics cards
34 minutes ago, Mike171 said:AMD graphics card does not support tensorflow-gpu because of opencl
And it never will because AMD have made a tensorflow port that uses HCC instead of OpenCL (which is mostly dead and dying as an API at this point). The AMD Tensorflow can be found here. The installation is not at all easy if you're not experienced in Linux. AMD don't support Windows at all for this stuff. You also need R9 Fury or newer, ideally Vega or RX series (doesn't say in guides but I tried R9 290 and it didn't work).
34 minutes ago, Mike171 said:mix two different graphics card architectures for GPGPU
There is no hard and fast rule. It depends on the workload, program, etc. For Tensorflow you can do it.
The architecture of the GPU doesn't matter as long as it is supported. Tensorflow currently needs CUDA® Compute Capability 3.5 or higher. This is supported by GTX 700 series and newer (except for 760/770, which are rebranded 600 series).
24 minutes ago, Mike171 said:cannot take any advantages from sli
SLI is for graphics rendering (such as games) only. It is not used in compute and is not needed. Only NVLink can be used for compute and only on Quadro class cards, not consumer ones. From what I have read NVLink doesn't do much for Compute workloads which can be split easily across GPUs and so it isn't needed for Machine Learning uses.
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now