A Quick Guide to Benchmarking AI models on Azure: Mask R-CNN with MLPerf Training v2.1

By Sonal Doomra, Hugo Affaticati, Reza Soroushmehr, and Jon Shelley

November 9, 2022

Introduction

Azure is pleased to showcase results from our MLPerf Training v2.1 submission. For this submission, we benchmarked our NC A100 v4-series and NDm A100 v4-series offerings, which are our flagship virtual machine (VM) types for mid-end and high-end AI training workloads, respectively. NC A100 v4-series features up to four NVIDIA A100 PCIe Tensor Core GPUs with 80GB memory each, 96 3rd generation AMD EPYC Milan processor cores and 880 GiB of system memory.  NDm A100 v4 VMs are powered by eight NVIDIA A100 SXM 80GB Tensor Core GPUs with NVIDIA (NVLink 3.0 technology) and 96 physical 2nd generation AMD EPYC™ 7V12 CPU cores. In addition, with support for NVIDIA Quantum InfiniBand networking with in-network computing, the NDm A100 v4 VM series enable customers to scale out and meet the computational demands of multi-node distributed training workloads at scale.

In this document we share the highlights of MLPerf™ Training v2.1 results along with how to run these benchmarks on Azure. These MLPerf™ benchmark results demonstrate Azure’s commitment to providing our customers with the latest GPU offerings, being in line with on-premises performance and its commitment to enabling our customers to run AI at scale in the cloud.

MLPerfTM from MLCommons®

MLCommons® is an open engineering consortium of AI leaders from academia, research labs, and industry where the mission is to “build fair and useful benchmarks” that provide unbiased evaluations of training and inference performance for hardware, software, and services—all conducted under prescribed conditions. MLPerf™ Training benchmarks consist of real-world compute-intensive AI workloads to best simulate customer’s needs. MLPerf™ tests are transparent and objective, so technology decision makers can rely on the results to make informed buying decisions.

 Highlights from MLPerfTM training v2.1 benchmark results

  1. Azure is the only cloud provider that demonstrated distributed training at scale with up to 128 NVIDIA A100 Tensor Core GPUs.
  2. Azure achieved a 5.3% reduction in training time from the previous round (MLPerf Training v2.0) on an eight NVIDIA GPU system for the Mask R-CNN benchmark.
  3. Azure training times are at par with the on-premises submitters across all the benchmarks.

Full results on MLCommons website.

How to replicate the results in Azure

Pre-requisites:

NC A100 v4-series (single node): Deploy and set up a virtual machine on Azure by following Getting started with the NC A100 v4-series.

NDm A100 v4-series (multi node): Deploy and set up a cluster of virtual machines with CycleCloud by following the script Setup CycleCloud to run NGC containers

Set up the environment

First, we need to export the path to the directory where we will perform the benchmarks.

The path for NC A100 v4-series (single node) is:

export PATH_DIR=/mnt/resource_nvme

The path for NDm A100 v4-series (multi node) is:

export PATH_DIR=/shared/mlcommons

Then, we can clone the repository in the directory we created previously.

cd $PATH_DIR
git clone https://github.com/mlcommons/training_results_v2.1.git

Get the dataset for Mask R-CNN

The Mask R-CNN script operates on COCO, a large-scale object detection, segmentation, and captioning dataset. To download and verify the dataset, use the following commands.

cd $PATH_DIR/training_results_v2.1/Azure/benchmarks/maskrcnn/implementations/ND96amsr_A100_v4/maskrcnn/dataset_scripts
./download_dataset.sh
./verify_dataset.sh

This should return PASSED.

Then, to extract the dataset, use:

DATASET_DIR=$PATH_DIR/mlperf/data/maskrcnn_data/
mkdir -p $DATASET_DIR
DATASET_DIR=$DATASET_DIR ./extract_dataset.sh

Mask R-CNN uses pre-trained ResNet50 as a backbone. To download and verify the RN50 weights use:

DATASET_DIR=$DATASET_DIR ./download_weights.sh

Make sure DATASET_DIR is writable.

To speed up loading of coco annotations during training, the annotations can be pickled since unpickling is faster than loading a json. Repeat the command below for all the annotations files in the json format to convert it to the pickle format.

python pickle_coco_annotations.py --root $DATASET_DIR --ann_file $DATASET_DIR/coco2017/annotations/<FILENAME>.json --pickle_output_file $DATASET_DIR/pkl_coco/<FILENAME>.json.pickled

Run the Mask R-CNN benchmark

The steps to run the benchmark consist of building the docker container, sourcing the configuration file, and starting the benchmark. The path and run commands differ depending on the virtual machines you are testing.

For NC A100 v4-series:

cd $PATH_DIR/training_results_v2.1/Azure/benchmarks/maskrcnn/implementations/NC96ads_A100_v4
docker build -t mlperftrainingv21/maskrcnn:latest .
source ./config_DGXA100_NC.sh
CONT= mlperftrainingv21/maskrcnn:latest DATADIR=$PATH_DIR/mlperf/data/maskrcnn_data/ ./run_with_docker.sh

For NDm A100 v4-series:

cd $PATH_DIR/training_results_v2.1/Azure/benchmarks/maskrcnn/implementations/ND96amsr_A100_v4
docker build -t mlperftrainingv21/maskrcnn:latest .
enroot import -o maskrcnn.sqsh "dockerd://mlperftrainingv21/maskrcnn:latest"
source ./config_DGXA100.sh
CONT=./maskrcnn.sqsh DATADIR=$PATH_DIR/mlperf/data/maskrcnn_data/ sbatch -N $DGXNNODES -p ndmv4 --exclusive --gpus-per-node=${DGXNGPU} ./run.sub

The above steps can be replicated for the other MLPerf Training v2.1 benchmarks or cluster sizes. One would have to adapt to the right configuration files and steps to preprocess the data.

#AzureHPCAI #MakeAIYourReality

Return to Solution Channel Homepage
Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

PFAS Regulations, 3M Exit to Impact Two-Phase Cooling in HPC

January 27, 2023

Per- and polyfluoroalkyl substances (PFAS), known as “forever chemicals,” pose a number of health risks to humans, with more suspected but not yet confirmed – and, as a result, PFAS are coming under increasing regu Read more…

Sweden Plans Expansion for Nvidia-Powered Berzelius Supercomputer

January 26, 2023

The Atos-built, Nvidia SuperPod-based Berzelius supercomputer – housed in and operated by Sweden’s Linköping-based National Supercomputer Centre (NSC) – is already no slouch. But now, Nvidia and NSC have announced Read more…

Multiverse, Pasqal, and Crédit Agricole Tout Progress Using Quantum Computing in FS

January 26, 2023

Europe-based quantum computing pioneers Multiverse Computing and Pasqal, and global bank Crédit Agricole CIB today announced successful conclusion of a 1.5-year POC study “to evaluate the contribution of an algorithmi Read more…

Critics Don’t Want Politicians Deciding the Future of Semiconductors

January 26, 2023

The future of the semiconductor industry was partially being decided last week by a mix of politicians, policy hawks and chip industry executives jockeying for influence at the World Economic Forum. Intel CEO Pat Gels Read more…

Riken Plans ‘Virtual Fugaku’ on AWS

January 26, 2023

The development of a national flagship supercomputer aimed at exascale computing continues to be a heated competition, especially in the United States, the European Union, China, and Japan. What is the value to be gained Read more…

AWS Solution Channel

Shutterstock_1687123447

Numerix Scales HPC Workloads for Price and Risk Modeling Using AWS Batch

  • 180x improvement in analytics performance
  • Enhanced risk management
  • Decreased bottlenecks in analytics
  • Unlocked near-real-time analytics
  • Scaled financial analytics

Overview

Numerix, a financial technology company, needed to find a way to scale its high performance computing (HPC) solution as client portfolios ballooned in size. Read more…

Microsoft/NVIDIA Solution Channel

Shutterstock 1453953692

Microsoft and NVIDIA Experts Talk AI Infrastructure

As AI emerges as a crucial tool in so many sectors, it’s clear that the need for optimized AI infrastructure is growing. Going beyond just GPU-based clusters, cloud infrastructure that provides low-latency, high-bandwidth interconnects and high-performance storage can help organizations handle AI workloads more efficiently and produce faster results. Read more…

Supercomputer Research Predicts Extinction Cascade

January 25, 2023

The immediate impacts of climate change and land-use change are severe enough, but increasingly, researchers are warning that large enough changes can then snowball into catastrophic changes. New, supercomputer-powered r Read more…

PFAS Regulations, 3M Exit to Impact Two-Phase Cooling in HPC

January 27, 2023

Per- and polyfluoroalkyl substances (PFAS), known as “forever chemicals,” pose a number of health risks to humans, with more suspected but not yet confirmed Read more…

Critics Don’t Want Politicians Deciding the Future of Semiconductors

January 26, 2023

The future of the semiconductor industry was partially being decided last week by a mix of politicians, policy hawks and chip industry executives jockeying for Read more…

Riken Plans ‘Virtual Fugaku’ on AWS

January 26, 2023

The development of a national flagship supercomputer aimed at exascale computing continues to be a heated competition, especially in the United States, the Euro Read more…

Shutterstock 1134313550

Semiconductor Companies Create Building Block for Chiplet Design

January 24, 2023

Intel's CEO Pat Gelsinger last week made a grand proclamation that chips will be for the next few decades what oil and gas was to the world over the last 50 years. While that remains to be seen, two technology associations are joining hands to develop building blocks to stabilize the development of future chip designs. The goal of the standard is to set the stage for a thriving marketplace that fuels... Read more…

Royalty-free stock photo ID: 1572060865

Fujitsu Study Says Quantum Decryption Threat Still Distant

January 23, 2023

Global computer and chip manufacturer Fujitsu today reported that a new study performed on its 39-qubit quantum simulator suggests it will remain difficult for Read more…

At ORNL, Jeff Smith Becomes Interim Director, as Search for Permanent Lab Chief Continues

January 20, 2023

UT-Battelle, which manages Oak Ridge National Laboratory (ORNL) for the U.S. Department of Energy, has appointed Jeff Smith as interim director for the lab as t Read more…

Top HPC Players Creating New Security Architecture Amid Neglect

January 20, 2023

Security of high-performance computers is being neglected in the pursuit of horsepower, and there are concerns that the ignorance may be costly if safeguards ar Read more…

Ohio Supercomputer Center Debuts ‘Ascend’ GPU Cluster

January 19, 2023

Less than 10 months after it was announced, the Columbus-based Ohio Supercomputer Center (OSC) has debuted its Dell-built GPU cluster, “Ascend.” Designed to Read more…

Leading Solution Providers

Contributors

SC22 Booth Videos

AMD @ SC22
Altair @ SC22
AWS @ SC22
Ayar Labs @ SC22
CoolIT @ SC22
Cornelis Networks @ SC22
DDN @ SC22
Dell Technologies @ SC22
HPE @ SC22
Intel @ SC22
Intelligent Light @ SC22
Lancium @ SC22
Lenovo @ SC22
Microsoft and NVIDIA @ SC22
One Stop Systems @ SC22
Penguin Solutions @ SC22
QCT @ SC22
Supermicro @ SC22
Tuxera @ SC22
Tyan Computer @ SC22
  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire