mxnet-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sheng Zha <szha....@gmail.com>
Subject Re: Ampere GPUs and slack access
Date Fri, 11 Sep 2020 20:34:08 GMT
Hi Michael,

Invite sent. Welcome!

The master branch should already support CUDA 11 & CuDNN 8, though we
are still in the progress of adding a CI check for it. Dick can
probably help elaborate more on it.

For the performance issue on quadro 4000, we will need information
such as the workload, mxnet version, environment, etc. I'd recommend
filing a bug report in github issues [1].

Best,
Sheng

[1] https://github.com/apache/incubator-mxnet/issues/new/choose

On Fri, Sep 11, 2020 at 10:13 AM Shannon, Michael J.
<mjs2399@cumc.columbia.edu> wrote:
>
> Dear all
>
> Hope you are having a good Friday.
>
> I would like to join the slack for support reasons - would this be possible?
>
> In addition, I was hoping that you might be able to comment on the likelihood of current
MXnet builds being able to work with the Ampere cards from NVIDIA. Namely we are going to
invest in an RTX 3090 for our machine learning needs (for now), and I am wondering whether
a specific software we use will be able to run MXnet to contact the GPU.
>
> Finally, currently using MXnet with our quadro 4000 card, I can only get it to use about
1 percent of the GPU. I wonder - is this a problem with MXnet?
>
> Thanks,
>
> Michael
>
>
> Michael Shannon, PhD
> Postdoctoral Research Scientist (Mace Lab),
> Department of Pediatrics,
> Columbia University Irving Medical Center
> P&S 10-451 630 W. 168th st.
> New York, NY, USA 10032
> 929-319-6335

Mime
View raw message