Back in 2015, Amazon had acquired Annapurna labs, an Israel-based firm with the intention to pursue its ambitions of a home-grown chip to energy its AWS servers. The results of that acquisition is the Graviton2 processor, a surprisingly good chip for Amazon’s server setting. The Graviton2 is a 64-core processor with a clock velocity of 2.5GHz and 32MB of cache. The chip helps 8X DDR4 reminiscence at 3200MHz, however maybe essentially the most attention-grabbing factor concerning the chip occurs to be the truth that its an ARM-based processor and never x86. The Graviton2 makes use of ARM’s Neoverse N1 cores, that are a by-product of the A76 cores that we noticed in among the greatest performing smartphones final yr.
Anandtech examined the chip out in its meant setting and in contrast it towards AMD EPYC 7571 and the Intel Xeon Platinum 8259CL. Since server-side efficiency is all about the associated fee, AnandTech examined the three chips when it comes to their performance-per-dollar. Turns out, Amazon’s Graviton2 chip presents virtually 40 p.c higher efficiency per greenback. This was verified by AnandTech by working numerous SPEC exams on every platform, with the time taken for completion transformed to hours. Then the quantity was multiplied by the associated fee per hour of working that individual server, yielding the associated fee per workload quantity. In every occasion, Amazon’s chip is 40 p.c more cost effective than Intel or AMD.
ARM has had ambitions for datacenter chips for a very long time now, and with the Graviton2, these ambitions would possibly lastly be realized. If AnandTech’s findings can actually be translated into real-world efficiency, it will imply severe disruption within the knowledge heart area, one greater than the one introduced by AMD.