3 ms·
Hopefully at some point deep learning and other applications for GPU technology will become vendor-agnostic, as with CPUs. While I can run Linux on either an In
by snazz 7y ago
Hopefully at some point deep learning and other applications for GPU technology will become vendor-agnostic, as with CPUs. While I can run Linux on either an Intel or an AMD CPU (and many others), it seems awfully silly that I can't run a CUDA application on an AMD GPU.
- gameswithgo 7y agothere are vendor agnostic ways to do it, but performance is just not very good on AMD gpus.
- etaioinshrdlu 7y agoA bigger problem than the performance is the very poor software support. None of the top frameworks officially support AMD or openCL. Only broken patchsets are available for both tensorflow or pytorch. Performance is an issue, but the bigger issue is that your model just won't work. NVIDIA is really the only game in town right now. Google's TPUs are somewhat of an exception but they only sort of count, because you can't just go buy a TPU for training. Makes TPUs a non-starter for a lot of businesses. What's interesting is the slow pace of change. 3 years ago we were in the exact same place. I don't see it changing overnight.
- slavik81 7y agoApple abandoned OpenCL years ago. It's now officially deprecated on macOS. The most recent version they implemented is OpenCL 1.2 (released 2011). For reference, the current version is OpenCL 2.2 (released 2017).
- mdaniel 7y agobecause you can't just go buy a TPU for training but that's what this is, right? https://coral.withgoogle.com/products/ https://coral.withgoogle.com/products/
- hollerith 7y agoThat's for low-power devices.
- bitL 7y agoIt's both poor software support and poor performance, e.g. Vega 20 has 15TFlops in theory, in practice one could get around 7.5TFlops. And only use models supported by ROCm, custom CUDA kernels used by many models nowadays might not work after cross-compilation to AMD instructions.
- colechristensen 7y agoI doubt NVIDIA would invest anywhere near as much on GPU applications if it weren't vendor locked. I bet if NVIDIA significantly investing in GPU applications the competition would too. It makes a whole lot of sense for #1 to vendorlock and #2 onwards to promote open standards. (See Microsoft circa 2001 vs. Microsoft today) If everybody went vendor-agnostic you'd get a lot less support or you would have to find somebody else to pick up the mantle with an economic motive.