Hi all,
I have been using Tensorflow 1.8 in my GTX 1080 with 390.87 Nvidia drivers(which has cuda 9). Now I want to update to RTX 2080. But the challenge is that I have stable version of tensorflow 1.8 running. Can I use the same 390.87 with cuda 9 for my tensorflow 1.8.
The updated version of Cuda(cuda 10) is not compatible with Tensorflow 1.8. Need suggestion ASAP.
Regards,
Prem Rao
The RTX2080 needs a minimum driver version 410, current is 418. While Cuda 9.0 would work with the updated driver, it’s incompatible to the RTX’s cc7.5. So when upgrading to a RTX, you will have to upgrade your Tensorflow/Cuda setup in any case.
By this statement “While Cuda 9.0 would work with the updated driver” do you mean that cuda 9.0 is compatible with 410 to 418?
Yes, the driver itself is downward compatible, so you can use GTX1080+driver418+cuda9.0.
but RTX2080+driver418+cuda9.0 would not reliably work due to cc7.5.
Ok. Thank you. How about 9.1 and 9.2 ? will that be same for RTX 2080?
Good question, 9.0 definitely doesn’t work, IIRC, some users had an RTX+9.2 running. There are already binaries for Tensorflow on cuda 10.0, any reason upgrading to those isn’t possible?