Amazon’s Custom ARM 7nm 64-Bit Graviton2 Processor Compared With Intel’s Xeon Server-Grade CPUs

Amazon has been making some serious strides in custom processors and purpose-built CPUs for its Amazon Web Services or AWS platform. These custom ARM-based processors are truly powerful and can easily compete against Intel Xeon processors that dominated the server market until recently. With Amazon promising more EC2 Instances on AWS to be powered by the new Amazon Graviton2 processors, the CPU battle has just intensified.

AWS announced its second-generation of ARM-based homegrown processors called Graviton2 at the re:Invent 2019 on Tuesday in Las Vegas. The 2nd Gen Graviton2 processors feature four times higher core count than the A1 instances they succeed. They will power Amazon’s sixth-generation EC2 instances. Given the specifications and features of the new Graviton2 CPUs, Amazon confidently claims the M6g instances have a significant boost in performance and cost-efficiency. Incidentally, the eCommerce giant directly compared the new Graviton2 processor with the Intel-based M5 instance.

Amazon Custom ARM-based 7nm Graviton2 Processor Specifications And Features:

The Amazon Graviton2 processor is a custom AWS-designed and purpose-built processor that has been fabricated on the 7nm process. The processor is based on 64-bit ARM Neoverse cores. With 30 billion transistors, the Graviton2 CPUs are claimed to have a 7x performance uplift compared to the Graviton-based A1 instances.

Interestingly, the first generation of the Graviton processors was announced just last year at re:Invent 2018. The 1st Gen Graviton was based on the 64-bit Armv8 Cortex-A72 microarchitecture first announced in 2015. The first-generation 16nm Neoverse platform supported the processor. The CPU sported four quad-core clusters. Each cluster had 2MB of L2 cache. In all, the 1st Gen Amazon Graviton CPU had a total of 16 cores running comfortably at 2.6 GHz.

The Graviton2 CPU takes a significant leap in die size reduction, and is now built on the 7nm Neoverse platform, codenamed Ares. ARM claims the 4-wide Neoverse N1 microarchitecture delivered a 30 percent increase in power efficiency (at the same frequency) and a 60 percent uplift in IPC. The processor also provides twice the floating-point SIMD performance per core.

Although the architecture is quite similar to the Cortex- A76, which is commonly deployed in smartphones, the N1 microarchitecture can be scaled up to 128 Cores. However, the 2nd Gen Graviton2 processors have 64 Cores. These are connected by a 2TB/s mesh architecture. The Graviton2 has twice the amount of L2 cache per core. It has 8 DDR4-3200 channels, which is essentially 5X faster than the previous generation. The processor has support for 64 PCIe 4.0 lanes and also for FP16 and INT8 numerics. Moreover, the processors are always-encrypted.

Amazon has three types of Graviton2-powered EC2 instances: General Purpose, Compute-Optimized, and Memory-Optimized. All the instances have up to 25 Gbps of network bandwidth, and 18 Gbps of EBS-Optimized bandwidth. AWS is so confident about the processors, they will reportedly power Amazon EMR, Elastic Load Balancing, Amazon ElastiCache, and other AWS services.

Amazon’s Custom ARM 7nm 64-Bit Graviton2 Processor Compared With Intel Xeon Server-Grade Processors:

Amazon offered a comparison between its own indigenously developed ARM 7nm 64-Bit Graviton2 processor with Intel’s Xeon server-grade processors. Incidentally, AWS compared their EC2 Instances with Intel-based M5 instances. Amazon claims the instances deliver 20 percent cost savings and up to 40 percent higher performance.

Compared to Intel Xeon processors, Amazon’s Graviton2 has both higher core-count and higher performance per virtual core. However, Amazon did not take into consideration one important aspect that Intel Xeon processors possess. Intel’s CPUs have HyperThreading, resulting in two vCPUs per core, which grants it a significant performance boost.

Moreover, ARM’s use in HPCs and Data Centers has been quite small compared to x86 systems. Still, ARM’s CPUs are now increasingly powering multiple and intensive cloud-computing instances for several remote users of AWS. Additionally, AMD’s highly credible 7nm EPYC platform has always offered power and efficiency for servers. AMD’s EPYC platform runs x86 code just like the Intel servers. This directly means Intel has some serious and tough competition from not just one but two companies.

Alap Naik Desai
A B.Tech Plastics (UDCT) and a Windows enthusiast. Optimizing the OS, exploring software, searching and deploying solutions to strange and weird issues is Alap's main interest.