Advanced Micro Devices continued to play its hot hand at this week’s Computex event in Taipei, Taiwan, highlighting its processor roadmap at the cutting-edge 7-nanometer process node that includes the next generation of its Epyc HPC processor for datacenters and Radeon GPUs for high-volume gaming, and increasingly for enterprise cloud workloads.
“We’ve really been focused on innovation in high-performance computing,” AMD CEO Lisa Su said in a keynote address on Monday (May 27). Sharpening its focus on datacenters and taking aim at server leader Intel Corp., Su stressed that “one of things about being in semiconductors is you have to make big bets.”
Those bets include a shift to Taiwan Semiconductor Manufacturing Co.’s 7-nm process technology ahead of datacenter rival Intel with AMD’s first 7-nm server CPUs scheduled to launch next quarter.
Su claimed more than 60 Epyc-based platforms in production, including Amazon Web Services and Microsoft Azure and Baidu public cloud instances.
Those partnerships build on AMD’s recent win along with Cray to develop a 1.5-exaflops supercomputer for the Energy Department’s Oak Ridge National Laboratory. The exascale machine is based on a future generation of AMD’s Epyc and Radeon “Instinct” processors running inside the Cray Shasta architecture.
“Our goal in 2021 is to have the world’s fastest supercomputer [based] on AMD technology,” Lu said.
Su also announced a deal with Microsoft to bring its HPC processor capability to the Azure cloud via Microsoft’s HB-series virtual machines tuned to HPC applications. The configuration will run on more than 11,500 Epyc cores.
“HB-series VMs feature the cloud’s first deployment of AMD Epyc 7000-series CPUs explicitly for HPC customers,” Microsoft noted in a blog post. “AMD Epyc features 33 percent more memory bandwidth than any x86 alternative, and even more than leading Power and Arm server platforms.”
Separately, AMD demoed its second-generation, 7-nm Epyc processor dubbed “Rome,” billed as delivering twice the horsepower and I/O bandwidth per socket when compared to the previous generation along with a four-fold boost in floating-point workloads per socket.
Forrest Norrod, general manager of AMD’s datacenter and embedded systems group, pitted two 64-core Rome processors running a molecular dynamics simulation (specifically the Apo1 v2.12 benchmark test) against two 28-core Intel Xeon Scalable 8280 “Cascade Lake” processors. In a head-to-head comparison, AMD claimed Rome more than doubled processor performance.
“It’s such a complicated thing [simulating protein that’s part of HDL, 92k atoms], that even the speed of Epyc can only simulate 20 nanoseconds a day — but that’s twice as many as Cascade Lake,” Norrod boasted.
Norrod said AMD would launch its second-generation Epyc Rome processor during the third quarter of 2019.
Meanwhile, AMD continues to build volume for its Radeon GPUs via gaming applications that increasingly run in the cloud. The chipmaker previewed its 7-nm Navi GPUs and said the the 5000 series graphics cards are set for a July launch. “We believe [Navi] can power a tremendous number of applications” via reduced latency and higher bandwidth, Su said. For example, she said Navi is the first gaming GPU using the fourth generation of PCI Express.
Investors liked what they heard from Su: AMD shares jumped about 10 percent today.