» Current | 2018 | 2017 | 2016 | Subscribe

Linley Newsletter

AMD Ships Industry’s First 7nm GPU

December 4, 2018

Author: Linley Gwennap

Jumping ahead of archrival Nvidia, AMD announced the industry’s first 7nm GPU, with shipments expected to begin this quarter. But the new chip will bypass PC gamers and head straight for the data center, serving mainly AI and high-performance-computing (HPC) workloads. To get to market sooner, the company reused its Vega architecture, shrinking it to TSMC’s leading-edge 7nm technology. AMD will offer the 7nm Vega in two Radeon Instinct PCIe cards: the flagship MI60 board is scheduled to ship in December, followed by the MI50 board in early 2019.

Whereas the older Vega10 GPU features two High Bandwidth Memory (HBM2) channels, the 7nm Vega20 doubles that number. This change, along with a slightly faster (1.0GHz) interface, gives it more than 1.0TB/s of memory bandwidth. The 16-lane PCI Express interface also doubles in bandwidth by implementing the new Gen4 specification. Vega20 adds two Infinity Fabric links that can connect up to eight GPUs in a cache-coherent cluster. AMD’s new accelerator cards offer the same 300W TDP rating as the earlier MI25 card.

The new manufacturing technology increases the peak clock speed by 20%, lifting performance accordingly. Whereas Vega10 had poor FP64 performance, the new design handles FP64 at half the rate of FP32, making it better suited to HPC. It also adds support for 8-bit-integer (INT8) data, doubling performance relative to FP16.

For floating-point applications that require single precision (FP32) or double precision (FP64), the MI60 delivers performance similar to that of Nvidia’s high-end Tesla V100 GPU, particularly when compared with the slower PCIe version. But applications that use Nvidia’s tensor engines will lag considerably on the Vega cards, which lack tensor acceleration.

Subscribers can view the full article in the Microprocessor Report.

Subscribe to the Microprocessor Report and always get the full story!

Purchase the full article

Events

Linley Fall Processor Conference 2018
Covers processors and IP cores used in embedded, communications, automotive, IoT, and server designs.
October 31 - November 1, 2018
Hyatt Regency, Santa Clara, CA
More Events »

Newsletter

Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »