3 ms·
> CUDA, Tensor Cores, Ansel, etc. I don't really use these things but if I'm paying the same amount I want similar capabilities. CUDA might be a bit tough. It'
by m-p-3 6y ago
> CUDA, Tensor Cores, Ansel, etc. I don't really use these things but if I'm paying the same amount I want similar capabilities.
CUDA might be a bit tough. It's hard for an equivalent to catch on if everyone is doing CUDA. It's the chicken-and-egg problem. There's ROCm* if that matters.
*: https://github.com/RadeonOpenCompute/ROCm https://github.com/RadeonOpenCompute/ROCm
- dathinab 6y agoHonestly at least for now for the desktop marked it doesn't matter that much. Software devs might consider it as a factor because they might want to play around with it, but for everyone else is a irrelevant feature.
- ZekeSulastin 6y agoI thought ROCm didn't support RDNA1 as of yet, let alone RDNA2.
- my123 6y agoYes, it still doesn't... (officially) There is some support that was contributed by the community but it's not complete.
- black_puppydog 6y agoThat's too bad. Gonna start a workstation build and even though it's only gonna do deep learning for prototyping, fiddling with low level stuff is just a non-starter compared to just spinning up anaconda env. Any chance AMD would hire some engineers to upstream rocm support into tensorflow/pytorch?
- RavlaAlvar 6y agoThey are 10yrs behind, I doubt they will ever able to catch up.
- Tom4hawk 6y agoPeople claimed similar things about AMD CPUs and GPUs ;)