AMD
|

AMD Unveils Record-Breaking Datacenter GPUs

AMD has announced their new range of datacenter graphics cards that offer up to 4.9x faster performance than their competitors for HPC.

https://cdn.mos.cms.futurecdn.net/3tSRriHsiFKjstasQTtDWF-970-80.jpg.webp

 

The new Instinct MI200 series claims to deliver up to 4.9x more performance than the Nvidia A100 GPUs in supercomputing, reaching speeds of up to 47.9 teraFLOPS. As for AI training workloads, the MI200 is 1.2x faster than its competitive accelerators according to AMD’s internal testing. The architecture for the accelerators is the new AMD CDNA 2 which features the world’s first ever multi-die GPU design which results in a record-breaking3.2TB/second peak theoretical memory bandwidth, which is 2.7x higher than AMD’s last generation of Instinct GPUs. The MI200 series has two different versions: the MI200 open accelerator module (OAM) which is built for very demanding HPC and AI workloads, and the MI210 PCIe for servers. The MI200 is currently available, with the MI210 to follow shortly.

https://cdn.mos.cms.futurecdn.net/sqqMSARXX2YmrhZaDKyBKF-970-80.jpg.webp

AMD Instinct MI200

Specifications

Below is a graph of the specs of the two new MI200 GPUs and last year’s MI100 as a comparison:
Model Compute Units Stream Processors Peak Half Precision (FP16) Performance Peak Single Precision (FP32) Performance Peak Double Precision (FP64) Performance Dedicated Memory Size Dedicated Memory Type Peak Memory Bandwidth
AMD Instinct™ MI250X 220 14,080 383 TFLOPs 47.9 TFLOPs 47.9 TFLOPs 128 GB HBM2e 3276.8 GB/s
AMD Instinct™ MI250 208 13,312 362.1 TFLOPs 45.3 TFLOPs 45.3 TFLOPs 128 GB HBM2e 3276.8 GB/s
AMD Instinct™ MI100 120 7,680 184.6 TFLOPs 23.1 TFLOPs 11.5 TFLOPs 32 GB HBM2 1228.8 GB/s

The Fastest HPC GPU in the World

AMD made the decision last year to split their GPUs into two channels, their gaming range and HPC range. This allowed them to focus on what was most important for each of them. This resulted in the birth of their Instinct MI100 range of HPC GPUs, which took the role of the fastest HPC GPU in the world. Now AMD aims to continue with this progress with their MI200 cards, which will deliver a “multi-generational leap in performance.”

In a statement from AMD’s SVP and GM of their datqa center branch, Forrest Norrod, he said that “AMD Instinct MI200 accelerators deliver leadership HPC and AI performance, helping scientists make generational leaps in research that can dramatically shorten the time between initial hypothesis and discovery.”

“With key innovations in architecture, packaging and system design, the AMD Instinct MI200 series accelerators are the most advanced datacenter GPUs ever, providing exceptional performance for supercomputers and datacenters to solve the world’s most complex problems.”

For more details on their MI200 GPUs, visit AMD’s website where they go into more details with regards to the workings and specifications of the graphics cards.

Similar Posts