GPUDatacenterReleased Nov 16, 2020

AMD Instinct MI100

32 GB HBM2 · 4096-bit · 7,680 cores

AMD's first accelerator built on the CDNA architecture, launched in November 2020 as a GPU designed only for HPC and AI rather than graphics. AMD introduced it as the first x86 server GPU to exceed 10 TFLOPS of double-precision performance, and it debuted the Matrix Cores used for mixed-precision AI workloads. The passively cooled, dual-slot PCIe card uses Infinity Fabric links to connect up to four GPUs in a single server.

General

Manufacturer: AMD
Model: Instinct MI100
Segment: Datacenter
GPU chip: Arcturus
Architecture: CDNA
Release date: Nov 16, 2020
Price: Search on Amazon

Memory

Memory size: 32 GB
Memory type: HBM2
Bus width: 4096-bit
Bandwidth: 1,228.8 GB/s
Memory clock: 1,200 MHz

Clocks & silicon

GPU clock: —
Boost clock: 1,502 MHz
Shader units: 7,680 Stream Processors (120 CU)
Tensor / AI cores: —
RT cores: —
Transistors: 25.6 billion
Technology (nm): 7 nm
TSMC N7
TDP / TBP: 300 W

Interface & APIs

Bus interface: PCIe 4.0 x16
SLI / multi-GPU: AMD Infinity Fabric Link (3 links, 92 GB/s peak)
DirectX: N/A (compute only)
OpenGL: N/A (compute only)
Vulkan: N/A
Other APIs: ROCm, OpenCL

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: Matrix Cores (CDNA)
Peak AI TOPS: 92.3 TOPS
Notes: Matrix Cores accelerate FP16 (184.6 TFLOPS), bfloat16 (92.3 TFLOPS), INT8 and INT4 (92.3 TOPS each). FP64 runs at half the FP32 rate (11.5 vs 23.1 TFLOPS), with full-chip ECC. No FP8 or TF32 support, which arrived with later CDNA generations.

All AMD GPUs →

Datacenter gpus from AMD announced around the same time as the AMD Instinct MI100. Add any of them to your compare list to see the specs side by side.

gpu