GPUDatacenterReleased Jun 20, 2017

AMD Radeon Instinct MI8

4 GB HBM · 4096-bit · 4,096 cores

The Radeon Instinct MI8 was the compact, low-latency member of AMD's first Instinct line-up, built on the same Fiji GPU as the short Radeon R9 Nano. Its 64 compute units at 1000 MHz and 4 GB of HBM on a 4096-bit bus give 512 GB/s of bandwidth and 8.2 TFLOPS of single-precision throughput within a 175 W board power, a balance AMD aimed at inference servers where memory bandwidth matters more than capacity.

General

Manufacturer: AMD
Model: Radeon Instinct MI8
Segment: Datacenter
GPU chip: Fiji
Architecture: GCN 3
Release date: Jun 20, 2017
Price: Search on Amazon

Memory

Memory size: 4 GB
Memory type: HBM
Bus width: 4096-bit
Bandwidth: 512 GB/s
Memory clock: 500 MHz

Clocks & silicon

GPU clock: 1,000 MHz
Boost clock: —
Shader units: 4,096 Stream Processors (64 CU)
Tensor / AI cores: —
RT cores: —
Transistors: 8.9 billion
Technology (nm): 28 nm
TSMC 28 nm
TDP / TBP: 175 W

Interface & APIs

Bus interface: PCIe 3.0 x16
SLI / multi-GPU: —
DirectX: 12 (12_0)
OpenGL: 4.6
Vulkan: 1.0
Other APIs: OpenCL 2.0, ROCm (HIP)

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: None (no matrix units)
Peak AI TOPS: —
Notes: Peak vector throughput is 8.2 TFLOPS at both FP16 and FP32, as Fiji runs half-precision at the same rate as single-precision, and 512 GFLOPS FP64, a 1:16 rate. There are no matrix units and no INT8 dot-product instructions.

All AMD GPUs →

Datacenter gpus from AMD announced around the same time as the AMD Radeon Instinct MI8. Add any of them to your compare list to see the specs side by side.

gpu