[
https://issues.apache.org/jira/browse/SINGA-423?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16786547#comment-16786547
]
wangwei commented on SINGA-423:
-------------------------------
Hi,
The task is to run Singa code on AMD GPUs by calling AMD's deep learning
library.
You need to update the files here
[https://github.com/apache/incubator-singa/tree/master/src/model/operation
|https://github.com/apache/incubator-singa/tree/master/src/model/operation.]
[
https://github.com/apache/incubator-singa/tree/master/src/core/tensor|https://github.com/apache/incubator-singa/tree/master/src/model/operation.]
https://github.com/apache/incubator-singa/tree/master/src/core/device
to add the operations implemented using AMD's APIs.
> Deep learning over AMD GPUs
> ---------------------------
>
> Key: SINGA-423
> URL: https://issues.apache.org/jira/browse/SINGA-423
> Project: Singa
> Issue Type: New Feature
> Reporter: wangwei
> Priority: Major
> Labels: gsoc2019
>
> GPU is a fueling factor for deep learning. Most deep learning libraries
> (including Singa) are using Nvidia GPUs because
> [cuDNN|https://developer.nvidia.com/cudnn] from Nvida provides almost all
> deep learning operations, which are also highly optimized.
>
> This ticket is to provide support for AMD GPUs in Singa. AMD has provided a
> similar [library |https://rocm.github.io/dl.html] as cuDNN. The task is then
> to integrate this library with Singa's programming model, write the
> documentation and conduct the corresponding test.
>
> Programming language: C++, CMake, OpenCL or
> [HIP|https://gpuopen.com/compute-product/hip-convert-cuda-to-portable-c-code/]
> OS: Linux (Ubuntu or CentOS)
> Tools: Github, JIRA,
> Machine learning (ML) background: basics of ML and DL (deep learning)
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
