HPC performance on IaaS with GPUs & HPC-IA
The use of accelerators can provide large gains in performance while also decreasing the costs associated with running HPC apps in the public cloud. Most public cloud service providers offer a selection of instances or VMs incorporating some of the most advanced server GPUs. Furthermore, several providers have already announced the upcoming availability of Nvidia Ampere architecture.
HPC users can choose between running their workloads in a single instance or tailor GPU clusters to their timely needs. This flexibility is especially critical because of the large upfront cost of GPU servers and fluctuating workloads. The performance and scalability of HPC apps on GPU clusters is a complex function that depends on many factors and where a large variety of variables can affect the final performance. Particularly critical is scalability at increasing cluster size. Contact us directly at if you are interested in using GPU-enabled IaaS.
Benchmarks with different apps have shown the ability of GPU-enabled IaaS to perform at levels similar to state-of-the-art supercomputers. An example is LAGHOS (LAGrangian High-Order Solver), which solves the time-dependent Euler equations of compressible gas dynamics in a moving Lagrangian frame and requires vast amounts of computational power. These early results show numerical performance for clusters built using AWS IaaS (p2 and p3 instance series). The comparisons use as a reference results from Lassen, which is currently ranked as the 14th most powerful supercomputer in the world.