AMD releases the Instinct MI200 series

AMD announced the launch of the new AMD Instinct MI200 series accelerator, the new AMD CDNA 2 architecture, the first multi-chip GPU that supports 128GB HBM2e memory, and the first Exascale class (exascale). Level) GPU. Instinct MI250X can provide greater performance for double-precision (FP64) high-performance applications, and bring more than 380 teraflops of theoretical half-precision (FP16) peak for AI workloads.

AMD stated that the AMD Instinct MI200 accelerator provides leading HPC and AI performance, helping scientists achieve a leap in research across the ages, while greatly reducing the time from initial hypothesis to discovery. With key innovations in architecture, packaging and system design, AMD Instinct MI200 series accelerators are the company’s more advanced data center GPUs in the history of the company, bringing superior performance to supercomputers and data centers to help supercomputers and data centers Solve complex problems in the world.

The key features and functions of AMD Instinct MI200 series accelerators include:

AMD CDNA 2 architecture: Compared with AMD’s previous generation GPUs, the second-generation matrix core accelerated FP64 and FP32 matrix operations can provide up to 4 times the theoretical peak performance of FP64

Leading packaging technology: As the industry’s first multi-chip GPU design with 2.5D Elevated Fanout Bridge technology (EFB), it can provide 1.8 times the number of cores and 2.7 times the memory bandwidth compared with AMD’s previous generation GPUs, and is industry-leading The theoretical peak memory bandwidth of 3.2 terabytes per second.

Third-generation AMD Infinity Fabric technology: Up to 8 Infinity Fabric links connect AMD Instinct MI200 with the third-generation EPYC processor and other GPUs in the node to achieve unified CPU memory/GPU memory consistency and maximum system throughput Through the powerful performance of the accelerator, the CPU code is simplified.

MI200 series parameters:

Post a Comment