search

apache / mxnet

Lightweight, Portable, Flexible Distributed/Mobile Deep Learning with Dynamic, Mutation-aware Dataflow Dep Scheduler; for Python, R, Julia, Scala, Go, Javascript and more
https://mxnet.apache.org
Apache License 2.0
20.77k stars 6.79k forks source link

Installation Guide for MXNet RTX 30 Series GPU #21027

Closed TNTran92 closed 2 years ago

TNTran92 commented 2 years ago

Description

Hello all, I am new to MXNet, so I am trying to have it installed and run on GPU. I am trying to install mxnet-cu101 to run on an RTX 3060. However, as soon as run the lines below, the code stalls. Nothing happens after that.

from mxnet import nd, gpu, gluon, autograd from mxnet.gluon import nn from mxnet.gluon.data.vision import datasets, transforms import time

x = nd.ones((3,4), ctx=gpu())

Note that if I take out ctx=gpu(), it works just fine.

x = nd.ones((3,4)) x

[[1. 1. 1. 1.] [1. 1. 1. 1.] [1. 1. 1. 1.]] <NDArray 3x4 @cpu(0)>

Also, it seems MXNet recognize there is a GPU

import mxnet as mx mx.context.num_gpus() 1 from mxnet.runtime import feature_list feature_list() [✔ CUDA, ✔ CUDNN, ✖ NCCL, ✔ CUDA_RTC, ✖ TENSORRT, ✖ CPU_SSE, ✖ CPU_SSE2, ✖ CPU_SSE3, ✖ CPU_SSE4_1, ✖ CPU_SSE4_2, ✖ CPU_SSE4A, ✖ CPU_AVX, ✖ CPU_AVX2, ✔ OPENMP, ✖ SSE, ✖ F16C, ✖ JEMALLOC, ✔ BLAS_OPEN, ✖ BLAS_ATLAS, ✖ BLAS_MKL, ✖ BLAS_APPLE, ✔ LAPACK, ✖ MKLDNN, ✔ OPENCV, ✖ CAFFE, ✖ PROFILER, ✖ DIST_KVSTORE, ✖ CXX14, ✖ INT64_TENSOR_SIZE, ✔ SIGNAL_HANDLER, ✖ DEBUG]

Question: 1/ What am I doing wrong? 2/ Is there a comprehensive guide on how to install mxnet for 30 series gpu on Windows in conda?

Below is my specs

CUDA 10.1 and CuDNN 8.0.5.39 mxnet-cu101 Geforce Game Ready Driver: Version 512.77

Device name MyPC Processor AMD Ryzen 9 5900X 12-Core Processor 3.70 GHz Installed RAM 32.0 GB System type 64-bit operating system, x64-based processor Pen and touch Pen support

Edition Windows 11 Pro Version 21H2 Installed on ‎1/‎2/‎2022 OS build 22000.675 Experience Windows Feature Experience Pack 1000.22000.675.0

github-actions[bot] commented 2 years ago

Welcome to Apache MXNet (incubating)! We are on a mission to democratize AI, and we are glad that you are contributing to it by opening this issue. Please make sure to include all the relevant context, and one of the @apache/mxnet-committers will be here shortly. If you are interested in contributing to our project, let us know! Also, be sure to check out our guide on contributing to MXNet and our development guides wiki.

dai-ichiro commented 2 years ago

RTX 3060(Ampere Architecture) don't support CUDA 10.1.

https://docs.nvidia.com/deeplearning/cudnn/support-matrix/index.html

TNTran92 commented 2 years ago

I see. In that case, is there mxnet > 1.8.0 for windows? I saw there is 1.9.0 with CUDA 11.2 support, but that is for Linux https://pypi.org/project/mxnet-cu112/

TristonC commented 2 years ago

@josephevans Does MXNet have 1.9.0 wheels (support CUDA 11) for windows?

EpicKJ commented 2 years ago

I have the same problem. Nothing happens after I run the test: x = nd.ones((3,4), ctx=gpu()), even though my RTX 3070 is recognized.

TNTran92 commented 2 years ago

Confirmed that on Windows, MXNet only support up to mxnet-cu102. The issue is that Cuda 10.2 is only supported up to Turing (20-series and older). Also confirmed that mxnet-cu102 run seemlessly on GTX 1650.

After spending a few days trying to build it from source, I have given up on getting version 1.9 working on Windows. I installed Ubuntu and is trying to get mxnet-cu112 to work. For those interested, CUDA 11.2 (and compatible Cudnn) only supports up to Ubuntu 20.04

TNTran92 commented 2 years ago

Confirmed mxnet-cu112 works on RTX 3060 in Ubuntu 20.04. As of now, looks like using Ubuntu is the only way to get 30 series card to work.

EpicKJ commented 2 years ago

Confirmed mxnet-cu112 works on RTX 3060 in Ubuntu 20.04. As of now, looks like using Ubuntu is the only way to get 30 series card to work.

All right. Thanks for your information