AMD Instinct™ MI250 Accelerator


World’s Fastest HPC and AI Accelerator1

The era of exascale is here. Immense computational power coupled with the fusion of HPC and AI is enabling researchers and scientists to tackle our most pressing challenges from climate change to vaccine research. With the AMD Instinct™ MI200 accelerators and ROCm™ 5.0 software ecosystem, innovators can tap the power of the world’s most powerful HPC and AI data center GPUs to accelerate their time to science and discovery.1

Based on the 2nd Gen AMD CDNA™ architecture, AMD Instinct™ MI200 accelerators deliver a quantum leap in HPC and AI performance over competitive data center GPUs today. With an up to 4x advantage in HPC performance compared to competitive GPUs, the MI200 accelerator is the first data center GPU to deliver 383 teraflops of theoretical mixed precision FP16 performance for deep learning training, offering users a powerful platform to fuel the convergence of HPC and AI.

1: For more info see the test results here on

Get a Quote
XENON AMD Instinct MI250 MI250X GPU
AMD Instinct™ MI250 Accelerator

GPU Specifications

AMD Instinct™ MI250 Accelerator

  • GPU Architecture: CDNA2
  • Stream Processors: 13,312
  • Peak Half Precision (FP16) Performance: 362.1 TFLOPs
  • Peak Single Precision Matrix (FP32) Performance: 90.5 TFLOPs
  • Peak Single Precision (FP32) Performance: 45.3 TFLOPs
  • Peak INT4 Performance: 362.1 TOPs
  • Peak bfloat16: 362.1 TFLOPs
  • Lithography: TSMC 6nm FinFET
  • Compute Units: 208
  • Peak Engine Clock: 1700 MHz
  • Peak Double Precision Matrix (FP64) Performance: 90.5 TFLOPs
  • Peak Double Precision (FP64) Performance: 45.3 TFLOPs
  • Peak INT8 Performance: 362.1 TOPs
  • OS Support Linux: x86_64

Total Board Power (TBP)

AMD Instinct™ MI250 Accelerator

500W | 560W Peak


GPU Memory

AMD Instinct™ MI250 Accelerator

  • Dedicated Memory Size: 128 GB
  • Memory Interface: 8192-bit
  • Peak Memory Bandwidth: Up to 3276.8 GB/s
  • Dedicated Memory Type: HBM2e
  • Memory Clock: 1.6 GHz
  • Memory ECC Support: Yes (Full-Chip)

Board Specifications

AMD Instinct™ MI250 Accelerator

  • Form Factor: OAM Module
  • Infinity Fabric™ Links: 8
  • Cooling Passive: OAM
  • Bus Type PCIe®: 4.0 x16
  • Peak Infinity Fabric™ Link Bandwidth: 100 GB/s


AMD Instinct™ MI250 Accelerator

  • Board Height: N/A
  • Board Width: N/A

Additional Features

AMD Instinct™ MI250 Accelerator

  • Supported Technologies: AMD CDNA™ 2 Architecture
  • AMD ROCm™ - Ecosystem without Borders
  • AMD Infinity Architecture
  • Page Retirement: Yes
  • RAS Support: Yes
  • Coherency Enabled: No

Software API Support

AMD Instinct™ MI250 Accelerator

  • OpenMP®: Yes
  • HIP: Yes
  • OpenCL™: Yes
  • ROCm™ Open Ecosystem: Yes


AMD Instinct™ MI250 Accelerator

  • TensorFlow: Yes
  • Kokkos: Yes
  • PyTorch: Yes
  • RAJA: Yes

Product Basics

AMD Instinct™ MI250 Accelerator

  • Product Family: AMD Instinct™
  • Platform: Server
  • Product Line: AMD Instinct™ MI Series
  • Launch Date: 11/08/2021

Quick Quote Request

  • This field is for validation purposes and should be left unchanged.

Get a Quote Talk to a Solutions Architect