A moment ago I received a notification that cuDNN 5 is now out. On the webpage[1] it mentions “Supported on Windows, Linux and MacOS systems with Pascal, Kepler, Maxwell, Tegra K1 or Tegra X1 GPUs.” However, last time I checked with cuDNN 3, it requires at least CUDA 7.0. Tegra K1 (being a 32-bit platform) never received support for this version of CUDA. As a result, I was unable to use CUDA-accelerated CAFFE on our Jetson board.
Is there a good reason to assume cuDNN 5 will work on the Tegra K1? Or is the linked webpage incorrectly claiming compatibility?
Hi RSpliet,
Thanks for reporting this issue.
Actually, TK1 support ended with cuDNN v2, the information on https://developer.nvidia.com/cudnn will be corrected soon.
Thanks
Hello kayccc,
Thank you for your reply. You might want to chase your colleagues a bit more, given how the web page has not yet been updated to reflect this finding.
Cheers,
Roy