You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@mxnet.apache.org by "Marco de Abreu (JIRA)" <ji...@apache.org> on 2018/03/14 15:52:00 UTC
[jira] [Assigned] (MXNET-99) P3 (Nvidia V100) not supported by
CUDA8
[ https://issues.apache.org/jira/browse/MXNET-99?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Marco de Abreu reassigned MXNET-99:
-----------------------------------
Assignee: Marco de Abreu
> P3 (Nvidia V100) not supported by CUDA8
> ---------------------------------------
>
> Key: MXNET-99
> URL: https://issues.apache.org/jira/browse/MXNET-99
> Project: Apache MXNet
> Issue Type: Bug
> Reporter: Marco de Abreu
> Assignee: Marco de Abreu
> Priority: Major
>
> During my tests to upgrade our CI for [Quantization support|[https://github.com/apache/incubator-mxnet/pull/9552],] I have noticed that NVCC supplied with CUDA 8 does not support the GPU architecture 7.0, which represents Volta GPUs used in AWS P3 instances. Our Makefile automatically removes this architecture in [https://github.com/apache/incubator-mxnet/blob/master/Makefile#L280,] which is fine since it would crash otherwise.
>
> This causes pre-compilation to generate incompatible PTX binaries and thus causing issues during runtime - the JIT compiler has to kick in and libraries like CuBLAS report CUBLAS_STATUS_ARCH_MISMATCH.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@mxnet.apache.org
For additional commands, e-mail: issues-help@mxnet.apache.org