Amd Previews Time To Come Gpus As Well As Cpus For The Datacenter

From the supercomputer nerds at Top500, June 7:

This calendar week at the Computex estimator tradeshow inwards Taipei, AMD CEO Lisa Su outlined where her companionship stood amongst its next-generation Radeon Instinct GPU together with its Zen 2 EPYC CPU.

Although much of AMD’s presentation had to produce amongst its client-side processor business, Su position a lot of emphasis on their upcoming server silicon together with specifically took aim at the HPC destination of the business. “We are absolutely most high surgery computing,” she declared.

The server chips inwards enquiry are novel versions of the Radeon Instinct GPU together with the EPYC CPU. Both volition locomote manufactured amongst 7nm procedure technology, which according to AMD, volition furnish twice the transistor density together with position out energy efficiency every bit its 14nm parts, spell delivering 1.35 the performance. We’re non certain how they came upward amongst that surgery figure, given the broad variation inwards application behaviour on a given processor. That aside, the jump from 14nm to 7nm should bump upward surgery significantly, regardless of whatever blueprint upgrades.

In the illustration of the second-generation Radeon Instinct GPU, AMD is sticking amongst its existing Vega architecture, which was used on the master high-end Instinct part, the MI25. That GPU delivered 24.6 teraflops at one-half precision (FP16), 12.29 teraflops at unmarried precision, but simply 0.768 teraflops at total double precision (FP64). The depression amount of 64-bit surgery was a outcome of the production line’s intended market, namely AI/machine learning together with ray-traced icon rendering, where 64-bit math doesn’t brand much sense.

Moving this GPU to 7nm could presumably double those superlative surgery numbers, but we’ll bring to expect together with come across how AMD spreads those smaller transistors around. The novel GPU volition also come upward amongst 32 GB of HBM2, doubling the xvi GB of the MI25, together with matching the HBM2 capacity on NVIDIA’s latest V100 GPU together with Intel’s upcoming “Lake Crest” Neural Network Processor.
The chip is sampling similar a shot to lead customers together with is scheduled to locomote to a greater extent than oftentimes than non available afterwards inwards the year.

The Zen 2 EPYC processor, code-named “Rome,” is nevertheless inwards the lab, although Su did concur upward a sample chip for the crowd....MORE