The fourth-generation AWS Graviton4 processor, now usually out there, is a high-performance CPU based mostly on Arm Neoverse-V2 cores, boasting important enhancements over its predecessor.
It options 96 cores, offering on common 30% larger compute efficiency and 40% sooner efficiency for databases and Java software program.
For those who’re questioning the way it compares with Intel and AMD rivals (and its predecessors), Phoronix has the reply. The positioning benchmarked 5 AWS cases within the “16xlarge” configuration, every with 64 vCPUs and 512GB of reminiscence. Graviton4 (r8g.16xlarge) was pitted in opposition to Graviton3 (r7g.16xlarge), Graviton2 (r6g.16xlarge), AMD EPYC 9R14 (r7a.16xlarge), and Intel Xeon 8488C (r7i.16xlarge). The cases had been examined utilizing Ubuntu 24.04 with the Linux Kernel 6.8 and inventory GCC 13.2 compiler.
Spectacular efficiency
The Graviton4 carried out impressively throughout the complete vary of exams and within the Geometric imply of all of the take a look at outcomes it got here in second behind AMD’s EPYC processor – and you’ll view the entire take a look at outcomes right here.
The outcomes matched with Amazon’s personal claims for its processor, with Graviton4 barely forward of the Intel Xeon occasion by about 5%, though AMD’s EPYC processor continues to steer by roughly 25% in general efficiency.
Michael Larabel, Phoronix founder, famous, “We are at the stage where the Neoverse-V2 cores with Graviton4 are effectively matching Intel Sapphire Rapids core-for-core and depending upon the workload also matching or outperforming AMD 4th Gen EPYC. Of course, Graviton4 just went GA and will be going up against AMD EPYC Turin and Intel Xeon 6 Granite Rapids in the coming months.”
Whereas Intel and AMD will proceed to steer when it comes to sheer efficiency, Graviton4 isn’t too far behind. It presently leads in ARM64 server efficiency, and given the generational leap AWS made with its newest processor, it is going to be attention-grabbing to see what enhancements it can deliver with Graviton5.
