title | description | services | ms.service | ms.subservice | ms.topic | ms.date | ms.reviewer | ms.author | author |
---|---|---|---|---|---|---|---|---|---|
HX-series VM sizes performance and scalability |
Learn about performance and scalability of HX-series VM sizes in Azure. |
virtual-machines |
virtual-machines |
hpc |
article |
05/23/2023 |
cynthn |
padmalathas |
padmalathas |
Applies to: ✔️ Linux VMs ✔️ Windows VMs ✔️ Flexible scale sets ✔️ Uniform scale sets
Performance expectations using common HPC microbenchmarks are as follows:
Workload | HX |
---|---|
STREAM Triad | 750-780GB/s of DDR5, up to 5.7 TB/s of 3D-V Cache bandwidth |
High-Performance Linpack (HPL) | Up to 7.6 TF (Rpeak, FP64) for 144-core VM size |
RDMA latency & bandwidth | < 2 microseconds (1 byte), 400 Gb/s (one-way) |
FIO on local NVMe SSDs (RAID0) | 12 GB/s reads, 7 GB/s writes; 186k IOPS reads, 201k IOPS writes |
The STREAM memory test can be run using the scripts in this GitHub repository.
git clone https://github.com/Azure/woc-benchmarking
cd woc-benchmarking/apps/hpc/stream/
sh build_stream.sh
sh stream_run_script.sh $PWD “hbrs_v4”
The HPL benchmark can be run using the script in this GitHub repository.
git clone https://github.com/Azure/woc-benchmarking
cd woc-benchmarking/apps/hpc/hpl
sh hpl_build_script.sh
sh hpl_run_scr_hbv4.sh $PWD
The MPI latency test from the OSU microbenchmark suite can be executed as shown. Sample scripts are on GitHub.
module load mpi/hpcx
mpirun -np 2 --host $src,$dst --map-by node -x LD_LIBRARY_PATH $HPCX_OSU_DIR/osu_latency
The MPI bandwidth test from the OSU microbenchmark suite can be executed as shown. Sample scripts are on GitHub.
module load mpi/hpcx
mpirun -np 2 --host $src,$dst --map-by node -x LD_LIBRARY_PATH $HPCX_OSU_DIR/osu_bw
[!NOTE] Define source(src) and destination(dst).
The Mellanox Perftest package has many InfiniBand tests such as latency (ib_send_lat) and bandwidth (ib_send_bw). An example command is shown.
numactl --physcpubind=[INSERT CORE #] ib_send_lat -a
[!NOTE] NUMA node affinity for InfiniBand NIC is NUMA0.
- Learn about scaling MPI applications.
- Review the performance and scalability results of HPC applications on the HX VMs at the TechCommunity article.
- Read about the latest announcements, HPC workload examples, and performance results at the Azure HPC Microsoft Community Hub.
- For a higher-level architectural view of running HPC workloads, see High Performance Computing (HPC) on Azure.