HPC: Google Cloud launches 3rd generation of AMD Epyc instances

Google Cloud recently complemented its high performance computing IaaS services with 3rd generation AMD Epyc processor-based instances. These VMs are available in many cloud regions in the US, Asia and Europe (in the Netherlands).

This fall, Google Cloud began offering virtual machines based on 3rd generation AMD Epyc (former codenamed Milan) processors for IaaS services. The first of its kind launched with the Tau family, T2D, is optimized for workloads that require horizontal telescoping. The cloud provider then introduced its third-generation AMD Epyc chips on general-purpose N2D instances, stating that they improved the offered price/performance ratio by 30%. Now announces the C2D family. This provides virtual machines optimized for high-performance computing, with instances based on 3rd generation AMD Epyc processors, taking advantage of advances in the architecture (Zen 3 cores). Use cases are those for HPC and intensive workloads (simulations in the pharmaceutical sector for vaccines, computational fluid dynamics (CFD), exploration of the universe, electronic design automation for the future industry, etc.) that require large memory, bandwidth, and I/O capabilities.

The C2D VM family supports hard disks and provides a bandwidth of 32 Gbit/s by default. Instances can be configured for up to 112 vCPUs (56 cores), 896GB of memory, and 3TB of local SSD storage. The C2D family is available in standard, high-CPU, and high-memory versions, each offering 7 hardware types for an optimal memory-to-core ratio to suit different workloads, Google Cloud says in a post. High-memory C2D configurations are designed for HPC and EDA (Electronics Design Automation) workloads that require high memory capacities, while high-CPU and standard instances will suit the demands of high-performance web servers, media conversion, and high-budget video game development. .

Covert computing support coming soon

In this post, Google Cloud compares the performance of its C2D instances based on 3rd generation Epyc (former codename Milan) with N2D VMs based on 2nd generation Epyc (former codenamed Rome). The cloud provider states that it has worked with AMD engineers to create a performance metric for some key high-performance computing applications, such as weather forecasting, molecular dynamics, or computational fluid dynamics and to compare, in particular, performance on floating point computing and memory. Among its users, Google Cloud cites clients such as Airshaper, a CFD platform for performing aerial simulations in the automotive industry. He says he was able to reduce costs by nearly 50% and simulation times by 30% compared to previous generation HPC instances in use.

Google Cloud worked with AMD to create benchmarks to compare performance between C2D instances based on the third generation of Epyc chips and N2D instances based on the second generation of Epycs. (credit: GCP)

For containerized workloads, the Google Kubernetes Engine supports C2D instances. Secret computing support will soon be available to protect sensitive data in process, including in-memory encryption. GCP says it is committed to providing confidential computing VM, GKE, and Dataproc services that provide AMD SEV (Secure Encrypted Virtualization) security functionality. “We will support SEV with the latest generation of Epyc processors in the medium term and plan for other security capabilities in the future,” the vendor says. C2D instances are available in multiple cloud regions, USA, Asia, Europe and the Netherlands.

Leave a Comment