Azure Provides Excellent HPC Cloud Performance With HBv4 Series Powered By AMD EPYC Genoa-X

Written by Michael Larabel in Processors on 4 August 2023 at 03:00 PM EDT. Page 2 of 5. 7 Comments.
High Performance Conjugate Gradient benchmark with settings of Performance Per Core, X Y Z: 160 160 160, RT: 60. HC was the fastest.

Immediately when firing off some HPC benchmarks, the generational leap from Azure's HBv3 to HBv4 top-end VM was very profound.

NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: BT.C. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: BT.C. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: BT.C. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: IS.D. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: IS.D. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: IS.D. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: MG.C. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: MG.C. HBv4 was the fastest.
NAS Parallel Benchmarks benchmark with settings of Performance Per Core, Test / Class: MG.C. HBv4 was the fastest.

With the NASA NPB benchmarks, the HBv4 instances with Genoa-X were radically faster than HBv3 thanks to the Zen 4 CPUs with AVX-512, the larger L3 cache via 3D V-Cache, DDR5 memory, and other significant generational enhancements. Even with the top-end VM costing double the prior generation, the HBv4 176 vCPU VM tended to deliver the best value in most of the benchmarks.

NAMD benchmark with settings of Performance Per Core, ATPase Simulation, 327,506 Atoms. HC was the fastest.
NAMD benchmark with settings of Performance Per Core, ATPase Simulation, 327,506 Atoms. HC was the fastest.
libxsmm benchmark with settings of M N K: 128. HBv4 was the fastest.
libxsmm benchmark with settings of M N K: 32. HBv4 was the fastest.

I wasn't entirely shocked though given all of my prior Genoa-X bare metal testing and being incredibly impressed by AMD's new server offerings.


Related Articles
Ampere Computing 2024 Roadmap Update: 256 Core 3nm CPU In 2025
Benchmarking The First RISC-V Cloud Server: Scaleway EM-RV1 Performance
AMD Announces Ryzen PRO 8840 & PRO 8000G Series CPUs
Intel Announces Gaudi 3 AI Accelerator, Intel Xeon 6 Brand
AMD Zen 4 vs. Zen 4C Performance, Zen 4C Core Scaling With Ryzen 5 8500G
Intel Core i3 14100 / i5 14500 vs. AMD Ryzen 5 8500G / 8600G In 500+ Benchmarks