Open manuelelucchi opened 7 years ago
Hi. For the moment there is nothing like the cudnn library for AMD. You have two options:
@manuelelucchi @jeandebleau @KeDengMS We need to raise awareness to Qualcomm which provide Snapdragon SoC for W10: IoTCore, W10A, and W10M
https://developer.qualcomm.com/forum/qdn-forums/software/adreno-gpu-sdk/34628
This question came up also in Gitter, answered by @cha-zhang at https://gitter.im/Microsoft/CNTK?at=59976ff0614889d47562c399.
We started engaging with AMD, still very early and nothing can be said yet.
thank you everyone for the answers!
@veikkoeeva That is the correct answer - vendors need to engage at the management and engineering levels for software to be successful. An open source community can do little without such discussions taking place.
https://www.microsoft.com/en-us/research/blog/microsoft-unveils-project-brainwave/
Third, Project Brainwave incorporates a software stack designed to support the wide range of popular deep learning frameworks. We already support Microsoft Cognitive Toolkit and Google’s Tensorflow, and plan to support many others. We have defined a graph-based intermediate representation, to which we convert models trained in the popular frameworks, and then compile down to our high-performance infrastructure.
via @galvesribeiro from https://gitter.im/dotnet/orleans. With ideas such as https://gitter.im/Microsoft/CNTK?at=59887b6b614889d475275902 could be rather interesting! If only I had some more minutes in a day...
I'll cross-reference https://github.com/Microsoft/CNTK/issues/2243 since relevant.
we need to figure out how to get the discussion OUT from "https://developer.qualcomm.com/forum" into github to widen the scope from just a few Qualcomm software engineers to Qualcomm marketing departments who understand the DEMAND and market projection.
You have two options:
- finding a learning framework supporting OpenCL. I do not know if there are plans from cntk developers to do that. I think there is an Opencl caffe branch sowewhere.
- finding a framwork using ROCm MIOpen. Only limited features up to now.
Or, finding a framework that uses OpenACC, which could compile to both NVIDIA GPUs, AMD GPUs, and other accelerators.
Hi. For the moment there is nothing like the cudnn library for AMD.
Can https://github.com/ROCmSoftwarePlatform/hipDNN replace cuDNN for AMD's GPUs?
I have an AMD GPU on my PC, are you going to support also AMD GPU soon? Thanks for everybody answers me