BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

AMD's Next-Gen Big Iron EPYC Server CPU Rumored To Pack 64 Cores And Boatloads Of Cache

Following
This article is more than 6 years old.

In the high performance data center, compute density per cubic centimeter is often the name of the game. Though thermal and power constraints always factor in, often times the sheer number of cores per socket, or server that can be packed and racked, rule the day. And it's in this very competitive arena that AMD's rumored next generation EPYC server processors, based on the company's Zen architecture, could really throw their muscle around.

AMD EPYC Processor Held Up By Company CEO Dr. Lisa Su

AMD

The first generation of AMD's EPYC processor architecture actually offers not only a higher number of Zen cores per socket (32 cores currently versus Intel Xeon Scalable at 28), but also significantly more PCI Express IO connectivity at 128 available lanes for connecting co-processors like GPUs (versus 48 with Intel Xeon). EPYC also has more theoretical memory bandwidth as well, with an 8-channel DDR4 controller, versus Intel’s Xeon Scalable 6-channel setup. However, the devil is in the real-world performance details and there are certainly applications and workloads that favor one of these high-end server architectures over the other.

Regardless, if recent rumors over at HotHardware ring true, AMD’s next generation EPYC server platform could not only bring double the CPU cores at 64 per socket but also four times the on-chip cache at 256MB, versus the 64MB L3 cache configuration on the current generation EPYC CPU. That’s a lot of die real estate-consuming memory, though AMD’s core CPU logic is highly efficient, even more so than Intel’s, actually. Memory structures are costly with respect to silicon area. However, with AMD moving to a 12nm FinFET process in 2018, it’s feasible that the company could pull this off.

AMD

Further, with the intrinsic round-trip latency of AMD’s Zen CCX, or CPU Complex architecture, having a lot more on-chip cache resources to keep data fresh and from having to go off-chip, would certainly help mitigate some of that latency. Tie that all up neatly with AMDs’ high speed serial Infinity Fabric between cores and we could be looking at a very compelling high density, big iron server CPU architecture that once again demonstrates the scalable elegance of AMD’s new Zen, literally and figuratively.

AMD

For now this is just a little bit of buzz in the rumor mill but it will be interesting to watch as AMD’s Zen and EPYC architectures evolve over the next 12 to 18 months. Where Intel was once completely dominant in the data center, Chipzilla now has a credible server threat once again, just like it does on the desktop/client side of the house with AMD Ryzen. And that, generally speaking, is nothing but goodness for competition and the market in general.

Follow me on Twitter or LinkedInCheck out my website or some of my other work here