Posts

[AMD EPYC] A new generation of CPU for cloud/server

Updated: Apr 29



Hello there!

This is Jon from Beyond GTA, and this blog will briefly describe features of new CPU generation "AMD EPYC" in cloud usage such as AWS and GCP


AMD EPYC Overview

● World's highest performance x86 CPU

● Performance improvement with the latest “Zen3” core

● AMD EPYC ™ 7003 PROCESSORS (development code name “MILAN”)

● Compatibility to use Rome platform with drop-in (BIOS update required)

● Expansion of security functions

● Support for 4ch / 6ch / 8ch channel memory

● For memory performance (up to 32MB per Core x86 maximum capacity L3 cache)


Features of each cloud


​Cloud name

​Features

Overview

​Main target instances

AWS

· Priced up to 10% to 45% lower than comparable EC2 instances.

· Seamless workload migration and flexibility.

· EC2 G4ad instance with AMD Radeon Pro V520 GPU and 2nd generation AMD EPYC processor

· Up to 45% cheaper than Nvidia G4dn instance for graphics-intensive applications such as virtual workstations / game streaming / graphics rendering.

・ M5a / M5ad (general purpose)

・ R5a / R5ad (memory optimization)

・ T3a (burst possible general purpose)

・ C5a (computing optimization)

・ G4ad (graphics heavy use)

・ 3rd generation EPYC core Amazon EC2 instance * Coming soon

-Supports a wide range of applications / workloads.

-Up to 80K core equipped with HBv2 that realizes HPC in the cloud.

· Azure D32as v4 virtual machines with 2nd generation AMD EPYC processors cost up to 53% less than Azure D32ds v4.

・ Dav4 (general purpose)

・ Eav4 (memory optimization)

・ HBv2 (HPC optimization)

・ HBv3 (HPC optimization) * Coming soon

・ Lsv2 (storage optimization)

・ NVv4 (GPU + CPU optimization)

GCP

· AMD-based N2D instances save up to 13% compared to comparable Intel N-series instances.

· Up to 39% performance improvement on Coremark benchmarks compared to comparable Intel N1 instances.

· AMD-based N2D VMs improve the performance of data management workloads that take advantage of AMD's higher memory bandwidth and higher throughput per system (available in larger VM choices), up to 224 vCPUs Equipped with Google's largest general-purpose VM.

・Benchmarks up to 2.5x improvement in benchmarks that benefit from AMD N2D's improved memory bandwidth, such as STREAM, resulting in 20-30% performance improvement for many workload types.

・ N2D Standard (general purpose)

・ N2D HighMEM (memory optimization)

・ N2D HighCPU (HPC optimization)

・ Confidential VMs

・ N2D Milan    * Coming soon  ・ C2D for HPC * Coming soon

・ Tau T2D * Coming soon

・Oracle Cloud

---

---

・ E2 Standard (general purpose)

・ E3 Standard (high performance)

・ Elastic Block Storage V2

・ Standard E4 Flex Sizing (general purpose) * Coming soon

・Alibaba Cloud

---

---

・ G6a (general purpose)

・ c6a (computing optimization)

・ r6a (memory optimization)


Reference material

● AWS: AMD EPYC ™ & AWS EC2 instances

● Microsoft Azure: AMD EPYC ™ & Microsoft Azure Virtual Machine

● Google Cloud: AMD EPYC ™ & Google Cloud Platform instance


Summary

Recently, AMD CPU instances have been released in the cloud as well, so when developing services / content in the cloud, why not consider it as an option according to the specifications of the application running on the cloud?


The information, shared above was originally written mid-September by Ohara Yuya who works in our Technology Sales Department of HQ in Japan.