Microsoft Azure EPYC Milan-X HBv3 Benchmarks Microsoft Azure HBv3 (Milan) versus HBv3 (Milan-X) benchmarking by Michael Larabel for a future article on Phoronix.com. Looking at performance of AMD EPYC Milan-X in Microsoft Azure cloud for a variety of workloads. HBv3: 64 Cores: Processor: 2 x AMD EPYC 7V13 64-Core (64 Cores), Motherboard: Microsoft Virtual Machine (Hyper-V UEFI v4.1 BIOS), Memory: 442GB, Disk: 2 x 960GB Microsoft NVMe Direct Disk + 32GB Virtual Disk + 515GB Virtual Disk, Graphics: hyperv_fb, Network: Mellanox MT27710 OS: CentOS Linux 8, Kernel: 4.18.0-147.8.1.el8_1.x86_64 (x86_64), Compiler: GCC 8.3.1 20190507, File-System: ext4, Screen Resolution: 1152x864, System Layer: microsoft HBv3 Milan-X: 64 Cores: Processor: 2 x AMD EPYC 7V73X 64-Core (64 Cores), Motherboard: Microsoft Virtual Machine (Hyper-V UEFI v4.1 BIOS), Memory: 442GB, Disk: 2 x 960GB Microsoft NVMe Direct Disk + 32GB Virtual Disk + 515GB Virtual Disk, Graphics: hyperv_fb, Network: Mellanox MT27710 OS: CentOS Linux 8, Kernel: 4.18.0-147.8.1.el8_1.x86_64 (x86_64), Compiler: GCC 8.3.1 20190507, File-System: ext4, Screen Resolution: 1152x864, System Layer: microsoft HBv3: 120 Cores: Processor: 2 x AMD EPYC 7V13 64-Core (120 Cores), Motherboard: Microsoft Virtual Machine (Hyper-V UEFI v4.1 BIOS), Memory: 442GB, Disk: 2 x 960GB Microsoft NVMe Direct Disk + 32GB Virtual Disk + 515GB Virtual Disk, Graphics: hyperv_fb, Network: Mellanox MT27710 OS: CentOS Linux 8, Kernel: 4.18.0-147.8.1.el8_1.x86_64 (x86_64), Compiler: GCC 8.3.1 20190507, File-System: ext4, Screen Resolution: 1152x864, System Layer: microsoft HBv3 Milan-X: 120 Cores: Processor: 2 x AMD EPYC 7V73X 64-Core (120 Cores), Motherboard: Microsoft Virtual Machine (Hyper-V UEFI v4.1 BIOS), Memory: 442GB, Disk: 2 x 960GB Microsoft NVMe Direct Disk + 32GB Virtual Disk + 515GB Virtual Disk, Graphics: hyperv_fb, Network: Mellanox MT27710 OS: CentOS Linux 8, Kernel: 4.18.0-147.8.1.el8_1.x86_64 (x86_64), Compiler: GCC 8.3.1 20190507, File-System: ext4, Screen Resolution: 1152x864, System Layer: microsoft OSPray 1.8.5 Demo: San Miguel - Renderer: SciVis FPS > Higher Is Better HBv3: 64 Cores .......... 52.63 |============================ HBv3 Milan-X: 64 Cores .. 55.56 |============================== HBv3: 120 Cores ......... 83.33 |============================================= HBv3 Milan-X: 120 Cores . 85.86 |============================================== OSPray 1.8.5 Demo: XFrog Forest - Renderer: SciVis FPS > Higher Is Better HBv3: 64 Cores .......... 10.75 |=========================== HBv3 Milan-X: 64 Cores .. 11.36 |============================ HBv3: 120 Cores ......... 16.95 |========================================== HBv3 Milan-X: 120 Cores . 18.41 |============================================== OSPray 1.8.5 Demo: San Miguel - Renderer: Path Tracer FPS > Higher Is Better HBv3: 64 Cores .......... 4.32 |======================== HBv3 Milan-X: 64 Cores .. 4.93 |=========================== HBv3: 120 Cores ......... 7.39 |========================================= HBv3 Milan-X: 120 Cores . 8.50 |=============================================== OSPray 1.8.5 Demo: NASA Streamlines - Renderer: SciVis FPS > Higher Is Better HBv3: 64 Cores .......... 71.43 |========================== HBv3 Milan-X: 64 Cores .. 83.33 |============================== HBv3: 120 Cores ......... 111.11 |======================================== HBv3 Milan-X: 120 Cores . 125.00 |============================================= OSPray 1.8.5 Demo: XFrog Forest - Renderer: Path Tracer FPS > Higher Is Better HBv3: 64 Cores .......... 5.70 |=========================== HBv3 Milan-X: 64 Cores .. 6.20 |============================= HBv3: 120 Cores ......... 9.09 |=========================================== HBv3 Milan-X: 120 Cores . 9.90 |=============================================== OSPray 1.8.5 Demo: Magnetic Reconnection - Renderer: SciVis FPS > Higher Is Better HBv3: 64 Cores .......... 38.46 |=========================== HBv3 Milan-X: 64 Cores .. 40.00 |============================ HBv3: 120 Cores ......... 62.50 |=========================================== HBv3 Milan-X: 120 Cores . 66.67 |============================================== OSPray 1.8.5 Demo: NASA Streamlines - Renderer: Path Tracer FPS > Higher Is Better HBv3: 64 Cores .......... 15.87 |========================== HBv3 Milan-X: 64 Cores .. 17.24 |============================= HBv3: 120 Cores ......... 24.59 |========================================= HBv3 Milan-X: 120 Cores . 27.78 |============================================== Embree 3.13 Binary: Pathtracer - Model: Crown Frames Per Second > Higher Is Better HBv3: 64 Cores .......... 40.78 |========================= HBv3 Milan-X: 64 Cores .. 46.09 |============================ HBv3: 120 Cores ......... 66.50 |======================================== HBv3 Milan-X: 120 Cores . 75.94 |============================================== Embree 3.13 Binary: Pathtracer ISPC - Model: Crown Frames Per Second > Higher Is Better HBv3: 64 Cores .......... 38.91 |========================= HBv3 Milan-X: 64 Cores .. 44.25 |============================ HBv3: 120 Cores ......... 63.29 |======================================== HBv3 Milan-X: 120 Cores . 71.98 |============================================== Embree 3.13 Binary: Pathtracer - Model: Asian Dragon Frames Per Second > Higher Is Better HBv3: 64 Cores .......... 41.86 |======================== HBv3 Milan-X: 64 Cores .. 45.68 |=========================== HBv3: 120 Cores ......... 64.37 |====================================== HBv3 Milan-X: 120 Cores . 78.77 |============================================== Embree 3.13 Binary: Pathtracer ISPC - Model: Asian Dragon Frames Per Second > Higher Is Better HBv3: 64 Cores .......... 42.00 |========================= HBv3 Milan-X: 64 Cores .. 45.66 |=========================== HBv3: 120 Cores ......... 63.44 |====================================== HBv3 Milan-X: 120 Cores . 76.52 |============================================== HPC Challenge 1.5.0 Test / Class: Random Ring Bandwidth GB/s > Higher Is Better HBv3: 64 Cores .......... 1.82992 |============= HBv3: 120 Cores ......... 0.76414 |===== HBv3 Milan-X: 64 Cores .. 6.20242 |============================================ HBv3 Milan-X: 120 Cores . 3.41538 |======================== High Performance Conjugate Gradient 3.1 GFLOP/s > Higher Is Better HBv3: 64 Cores .......... 40.02 |============================================= HBv3 Milan-X: 64 Cores .. 41.13 |============================================== HBv3: 120 Cores ......... 38.72 |=========================================== HBv3 Milan-X: 120 Cores . 39.44 |============================================ HPC Challenge 1.5.0 Test / Class: G-HPL GFLOPS > Higher Is Better HBv3: 64 Cores .......... 99.57 |========================== HBv3: 120 Cores ......... 89.36 |======================= HBv3 Milan-X: 64 Cores .. 175.03 |============================================= HBv3 Milan-X: 120 Cores . 139.04 |==================================== ONNX Runtime 1.9.1 Model: super-resolution-10 - Device: CPU Inferences Per Minute > Higher Is Better HBv3: 64 Cores .......... 6107 |============================================ HBv3 Milan-X: 64 Cores .. 6354 |============================================== HBv3: 120 Cores ......... 5852 |========================================== HBv3 Milan-X: 120 Cores . 6485 |=============================================== OpenVKL 1.0 Benchmark: vklBenchmark ISPC Items / Sec > Higher Is Better HBv3: 64 Cores .......... 120 |================================= HBv3 Milan-X: 64 Cores .. 126 |================================== HBv3: 120 Cores ......... 166 |============================================= HBv3 Milan-X: 120 Cores . 177 |================================================ OpenVKL 1.0 Benchmark: vklBenchmark Scalar Items / Sec > Higher Is Better HBv3: 64 Cores .......... 72 |=============================== HBv3 Milan-X: 64 Cores .. 74 |================================ HBv3: 120 Cores ......... 106 |============================================== HBv3 Milan-X: 120 Cores . 111 |================================================ GraphicsMagick 1.3.33 Operation: Noise-Gaussian Iterations Per Minute > Higher Is Better HBv3: 64 Cores .......... 585 |======================== HBv3 Milan-X: 64 Cores .. 874 |===================================== HBv3: 120 Cores ......... 721 |============================== HBv3 Milan-X: 120 Cores . 1123 |=============================================== HPC Challenge 1.5.0 Test / Class: Max Ping Pong Bandwidth MB/s > Higher Is Better HBv3: 64 Cores .......... 17174.75 |======================================== HBv3: 120 Cores ......... 15815.90 |===================================== HBv3 Milan-X: 64 Cores .. 18347.87 |=========================================== HBv3 Milan-X: 120 Cores . 16082.15 |====================================== Zstd Compression 1.5.0 Compression Level: 19 - Compression Speed MB/s > Higher Is Better HBv3: 64 Cores .......... 85.1 |===================================== HBv3 Milan-X: 64 Cores .. 106.2 |============================================== HBv3: 120 Cores ......... 82.0 |==================================== HBv3 Milan-X: 120 Cores . 93.8 |========================================= Zstd Compression 1.5.0 Compression Level: 19, Long Mode - Compression Speed MB/s > Higher Is Better HBv3: 64 Cores .......... 39.8 |=============================== HBv3 Milan-X: 64 Cores .. 59.8 |=============================================== HBv3: 120 Cores ......... 36.6 |============================= HBv3 Milan-X: 120 Cores . 52.8 |========================================= ASKAP 1.0 Test: tConvolve MPI - Degridding Mpix/sec > Higher Is Better HBv3: 64 Cores .......... 35988.0 |=========================== HBv3 Milan-X: 64 Cores .. 40896.3 |=============================== HBv3: 120 Cores ......... 41724.7 |================================ HBv3 Milan-X: 120 Cores . 57881.4 |============================================ ASKAP 1.0 Test: tConvolve MPI - Gridding Mpix/sec > Higher Is Better HBv3: 64 Cores .......... 38175.0 |============================= HBv3 Milan-X: 64 Cores .. 41160.1 |================================ HBv3: 120 Cores ......... 41287.0 |================================ HBv3 Milan-X: 120 Cores . 57042.5 |============================================ GROMACS 2021.2 Implementation: MPI CPU - Input: water_GMX50_bare Ns Per Day > Higher Is Better HBv3: 64 Cores .......... 7.476 |=================================== HBv3 Milan-X: 64 Cores .. 7.977 |====================================== HBv3: 120 Cores ......... 9.054 |=========================================== HBv3 Milan-X: 120 Cores . 9.705 |============================================== LAMMPS Molecular Dynamics Simulator 29Oct2020 Model: 20k Atoms ns/day > Higher Is Better HBv3: 64 Cores .......... 31.61 |===================================== HBv3: 120 Cores ......... 36.88 |=========================================== HBv3 Milan-X: 64 Cores .. 32.37 |====================================== HBv3 Milan-X: 120 Cores . 39.54 |============================================== LAMMPS Molecular Dynamics Simulator 29Oct2020 Model: Rhodopsin Protein ns/day > Higher Is Better HBv3: 64 Cores .......... 32.96 |======================================= HBv3: 120 Cores ......... 35.41 |========================================== HBv3 Milan-X: 64 Cores .. 33.96 |========================================= HBv3 Milan-X: 120 Cores . 38.47 |============================================== Facebook RocksDB 6.22.1 Test: Random Read Op/s > Higher Is Better HBv3: 64 Cores .......... 324822447 |========================== HBv3 Milan-X: 64 Cores .. 330410911 |=========================== HBv3: 120 Cores ......... 502728808 |======================================== HBv3 Milan-X: 120 Cores . 522387301 |========================================== Facebook RocksDB 6.22.1 Test: Read Random Write Random Op/s > Higher Is Better HBv3: 64 Cores .......... 1357349 |=================================== HBv3 Milan-X: 64 Cores .. 1381157 |==================================== HBv3: 120 Cores ......... 1587743 |========================================= HBv3 Milan-X: 120 Cores . 1684654 |============================================ John The Ripper 1.9.0-jumbo-1 Test: MD5 Real C/S > Higher Is Better HBv3: 64 Cores .......... 5697467 |=============================== HBv3 Milan-X: 64 Cores .. 5913000 |================================ HBv3: 120 Cores ......... 7143267 |======================================= HBv3 Milan-X: 120 Cores . 8141400 |============================================ Kripke 1.2.4 Throughput FoM > Higher Is Better HBv3: 64 Cores .......... 73635521 |================================= HBv3 Milan-X: 64 Cores .. 97373301 |=========================================== HBv3: 120 Cores ......... 88201142 |======================================= HBv3 Milan-X: 120 Cores . 93936541 |========================================= NAS Parallel Benchmarks 3.4 Test / Class: CG.C Total Mop/s > Higher Is Better HBv3: 64 Cores .......... 20940.70 |======================================== HBv3 Milan-X: 64 Cores .. 22323.23 |=========================================== HBv3: 120 Cores ......... 20926.52 |======================================== HBv3 Milan-X: 120 Cores . 21914.51 |========================================== BRL-CAD 7.32.2 VGR Performance Metric VGR Performance Metric > Higher Is Better HBv3: 64 Cores .......... 618492 |========================= HBv3 Milan-X: 64 Cores .. 655183 |========================== HBv3: 120 Cores ......... 1044368 |========================================= HBv3 Milan-X: 120 Cores . 1109486 |============================================ LULESH 2.0.3 z/s > Higher Is Better HBv3: 64 Cores .......... 44262.23 |=================================== HBv3 Milan-X: 64 Cores .. 54759.69 |=========================================== HBv3: 120 Cores ......... 40262.21 |================================ HBv3 Milan-X: 120 Cores . 47341.13 |===================================== NAMD 2.14 ATPase Simulation - 327,506 Atoms days/ns < Lower Is Better HBv3: 64 Cores .......... 0.41157 |============================================ HBv3 Milan-X: 64 Cores .. 0.40802 |============================================ HBv3: 120 Cores ......... 0.27619 |============================== HBv3 Milan-X: 120 Cores . 0.26900 |============================= WRF 4.2.2 Input: conus 2.5km Seconds < Lower Is Better HBv3: 64 Cores .......... 10150.07 |=========================================== HBv3 Milan-X: 64 Cores .. 9294.70 |======================================= HBv3: 120 Cores ......... 8766.54 |===================================== HBv3 Milan-X: 120 Cores . 7804.46 |================================= OpenFOAM 8 Input: Motorbike 60M Seconds < Lower Is Better HBv3: 64 Cores .......... 89.65 |============================================== HBv3 Milan-X: 64 Cores .. 65.50 |================================== HBv3: 120 Cores ......... 80.60 |========================================= HBv3 Milan-X: 120 Cores . 54.03 |============================ NWChem 7.0.2 Input: C240 Buckyball Seconds < Lower Is Better HBv3: 64 Cores .......... 2256.6 |======================================== HBv3 Milan-X: 64 Cores .. 2219.8 |======================================= HBv3: 120 Cores ......... 2557.1 |============================================= HBv3 Milan-X: 120 Cores . 2467.9 |=========================================== Xcompact3d Incompact3d 2021-03-11 Input: X3D-benchmarking input.i3d Seconds < Lower Is Better HBv3: 64 Cores .......... 348.11 |============================================= HBv3 Milan-X: 64 Cores .. 322.88 |========================================== HBv3: 120 Cores ......... 287.76 |===================================== HBv3 Milan-X: 120 Cores . 255.84 |================================= Xcompact3d Incompact3d 2021-03-11 Input: input.i3d 193 Cells Per Direction Seconds < Lower Is Better HBv3: 64 Cores .......... 13.67810920 |======================================== HBv3 Milan-X: 64 Cores .. 10.87805050 |================================ HBv3: 120 Cores ......... 12.28599450 |==================================== HBv3 Milan-X: 120 Cores . 9.93829823 |============================= RELION 3.1.1 Test: Basic - Device: CPU Seconds < Lower Is Better HBv3: 64 Cores .......... 418.48 |============================================= HBv3 Milan-X: 64 Cores .. 414.54 |============================================= HBv3: 120 Cores ......... 312.80 |================================== HBv3 Milan-X: 120 Cores . 274.35 |============================== Parboil 2.5 Test: OpenMP CUTCP Seconds < Lower Is Better HBv3: 64 Cores .......... 1.515548 |=========================================== HBv3 Milan-X: 64 Cores .. 1.127166 |================================ HBv3: 120 Cores ......... 0.976470 |============================ HBv3 Milan-X: 120 Cores . 0.847720 |======================== Timed Linux Kernel Compilation 5.14 Time To Compile Seconds < Lower Is Better HBv3: 64 Cores .......... 24.16 |============================================== HBv3 Milan-X: 64 Cores .. 23.91 |============================================== HBv3: 120 Cores ......... 19.07 |==================================== HBv3 Milan-X: 120 Cores . 18.56 |=================================== Timed Node.js Compilation 15.11 Time To Compile Seconds < Lower Is Better HBv3: 64 Cores .......... 96.35 |============================================== HBv3 Milan-X: 64 Cores .. 93.80 |============================================= HBv3: 120 Cores ......... 75.71 |==================================== HBv3 Milan-X: 120 Cores . 72.45 |===================================