At a dedicated event in San Francisco today under the 'Next Horizon' tagline, AMD announced that it is collaborating with TSMC to make the jump to the 7nm process node with its next-generation Zen 2 architecture.
As you may remember, two years ago, AMD debuted its 14nm Zen core architecture at an event dubbed New Horizon, architecture which underpins its Ryzen, EPYC, and Threadripper offerings. Earlier this year, the half-generational refresh dubbed Zen+ debuted, complete with its 12nm process found in the second-gen Threadripper CPUs unveiled in August.
According to the company, the 7nm node is more dense, consumes half the power - while delivering the same performance -, and sees a 1.25x performance increase at the same power draw.
Zen 2 has a number of improvements overall, among which a better branch predictor and enhanced instruction pre-fetching, larger micro Op Cache, double the floating point width (256-Bit), as well as double the load / store bandwidth.
As opposed to the first generation's multi-chip design, Zen 2 uses a so-called 'chiplet' design, which is very modular and delivers improved latency and thus performance. AMD claims that it has double the throughput with the same power draw, a higher number of instructions per clock, and of course, enhanced security.
Although the company is sampling Zen 2 today, it stated that CPUs using this architecture are on track for availability in 2019.
EPYC on AWS
While AMD has also hinted at the upcoming Zen 2-based EPYC family of CPUs (codenamed Rome), it was keen to make yet another related announcement. In keeping with its goal to bring its processors to as many users as possible, it stated that there was an obvious need to "partner with the number one cloud provider in the world". As such, it has jointly announced with Amazon that EPYC-based Elastic Compute Cloud (EC2) instances are available immediately.
Specifically the ones in question are R5a, M5a, and T3a as variants of EC2's already existing memory-optimized and general purpose instance lineups. R5 and M5 instances can be launched via the AWS Management Console or AWS CLI with current availability across the following regions: US East (Ohio, North Virginia), US West (Oregon), Europe (Ireland), and Asia Pacific, soon to be followed by others. As far as T3 instances are concerned, those should show up in the following weeks.
R5 and M5 instances will be available in six sizes, capping off at 96 vCPUs and 768 GB of memory, while T3 instances will be divvied up into seven sizes with up to 8 vCPUs and 32GB of memory. In terms of purchasing, these will be available as On-Demand, Reserved, or Spot instances.
Be sure to stick around for more news out of AMD's ongoing Next Horizon event.