14th November 2018 World's first 7nm GPUs Chip giant AMD has unveiled the Radeon Instinct MI60 and MI50 accelerators, the world's first 7 nanometre (nm) GPUs, designed to extend the capabilities of the modern datacentre.
The ability to go down to 7nm allows more transistors to be crammed onto an even smaller package than was possible before. In this case, the MI60 contains 13.2 billion transistors on a package size of 331mm², while the previous generation MI25 had 12.5 billion covering 495mm² – a 58% improvement in terms of transistor density. The MI60 features up to 14.7 TFLOPS of peak floating point performance, running across 64 Compute Units (4,096 stream processors). This is roughly equivalent to a top 10 supercomputer of 2004, compressed into a board length of just 10.5" (267 mm). According to AMD, the new accelerators feature "flexible mixed-precision capabilities", expanding the types of workloads they can address, including a range of high performance computing and deep learning applications. They are specifically designed to train complex neural networks rapidly and efficiently.
The MI60 (32GB) and MI50 (16GB) provide hyper-fast HBM2 (second-generation High-Bandwidth Memory) at up to 1 TB/s. They are also the first GPUs capable of supporting next-generation PCIe 4.02 interconnect – which is 2X faster than other x86 CPU-to-GPU technologies – and feature AMD Infinity Fabric Link GPU that enables GPU-to-GPU speeds that are 6X faster than PCIe Gen 3. "Legacy GPU architectures limit IT managers from effectively addressing the constantly evolving demands of processing and analysing huge datasets for modern cloud datacentre workloads," said David Wang, senior vice president of engineering, Radeon Technologies Group at AMD. "Combining world-class performance and a flexible architecture with a robust software platform and the industry's leading-edge ROCm open software ecosystem, the new AMD Radeon Instinct accelerators provide the critical components needed to solve the most difficult cloud computing challenges today and into the future." The MI60 accelerator is expected to ship to datacentre customers by the end of 2018. The MI50 accelerator is scheduled for Q1 2019.
Comments »
If you enjoyed this article, please consider sharing it:
|
