AMD Teases More Data Center Processors at 2021 Accelerated Data Center Premiere
At last week's Accelerated Data Center Premiere, AMD revealed a plethora of new processor technology including the 3rd Gen of EPYC processors (the Milan-X), the shift to 5 nm, and a Zen 4 architecture.
Last week, AMD hosted its virtual Accelerated Data Center Premiere, an event focused on hardware innovations with data center-level compute as the primary focus. One of the biggest announcements of this event was the new Instinct MI200 accelerator family, the flagship of which AMD claims to be the world's fastest for HPC and AI.
AMD's Instinct MI200 series of accelerators. Screenshot used courtesy of AMD
Beyond the Instinct MI200, AMD had several other exciting announcements from this event that were equally as noteworthy.
This article will cover some of the new releases from AMD to see the company's direction and vision for the future.
3rd Gen AMD EPYC Processors
From a hardware perspective, the most notable aspect of the Milan-X processors is integrating a new cache technology called 3D V-Cache. The 3D V-Cache is a stacked cache technology that leverages a new hybrid bonding technique to fuse together cache layers.
In the Milan-X processors, 3D V-Cache will enable up to 768 MB of total L3 cache per chip and 804 MB of total cache per chip.
At the event, AMD shared a variety of workloads where this increase in L3 cache improved performance by up to 60%.
Representation of the 3rd Gen EPYC Processor. Screenshot used courtesy of AMD
While not much else is known about Milan-X's hardware specs, what is known is that the offering will include up to 64 Zen-3 cores and be SP3 socket compatible.
Further, AMD announced that Meta (formerly known as Facebook) would be adopting the new CPUs family into its hyperscale cloud environment.
Zen 4 Processors
Teasing to the future, the next big announcement from AMD at their event was a look at their imminent Zen 4 processor family.
The Zen 4 family will consist of a "Genoa" model and a "Bergamo" model. AMD claims that these chips will be on a 5 nm TSMC process to enable the Genoa model to feature 96-cores, while the Bergamo model will reach up to 128-cores.
Known features for the Genoa processor. Screenshot used courtesy of AMD
According to AMD, this transition from 7 nm to 5 nm will enable a 2x improvement in density and power efficiency while increasing performance by 1.25x.
Along with these new chips, the Bergamo model will introduce a "Zen 4c" variant, with the 'c' indicating the model's focus on cloud applications. It is speculated that this Zen 4c, optimized for cloud-specific workloads, will be a more barebones offering, removing any redundant functionality in favor of compute density.
According to AMD, Genoa is anticipated to be released in 2022, while Bergamo will have to wait until 2023.
Heavy Industry Competition
The battle for hyperscale supremacy has never been more competitive than it is today. Battling with the likes of NVIDIA, a newly revamped Intel, and many more, AMD certainly has its work cut out for it.
However, with the slew of releases coming from last week's Accelerated Data Center Premiere, it certainly seems like AMD has a fighting chance to keep the momentum rolling for improving data center processing.