While the Nvidia GeForce RTX 3090 might be the most powerful gaming graphics card that you can buy today, chipmaker AMD has announced a new GPU accelerator that it claims to be the world’s fastest high-performance computing (HPC) GPU.
Clocking in at 11.54 teraflops in internal AMD tests, the Instinct MI100 is the first x86 server GPU to surpass the 10 teraflops performance barrier.
The Santa Clara-based tech giant is pitching the new chip for scientific research, highlighting its ability to handle workloads that matter in scientific computing.
In tests conducted by its performance labs, the AMD Instinct MI100 with 120 compute units delivered a peak FP64 performance of 11.54 teraflops and 23.1 teraflops of peak FP32 performance. This makes it several times faster than AMDs earlier accelerators.
The new GPU accelerator is based on the AMD CDNA architecture and is designed to work with the AMD EPYC server processors. Designed to support the latest PCIe 4.0 specification, it features 32GB of high-bandwidth HBM2 memory that’s clocked at 1.2 GHz and delivers 1.23 TB/s of memory bandwidth.
AMD also highlights that the new chip will enable developers to create high performance applications using its ROCm 4.0 platform that includes compilers, programming APIs and libraries.
It further adds that the company is working with various manufacturers including Dell, Hewlett Packard Enterprise, Gigabyte, and Supermicro, and servers featuring the new chip should hit the shelves by the end of the year.