The Open-Source AI Stack
RSS
All hardware

Hardware · Datacenter

AMD Instinct MI300X

AMD

Highest memory capacity in its generation at launch. AMD publishes dense and sparse figures side by side; the dense values are used here.

×8 over Infinity Fabric Compute units 1307 TF FP16 5.30 TB/s memory bus 192 GB HBM3 (VRAM)
Datacenter. Bus width tracks bandwidth (5.30 TB/s, sets decode speed); the box tracks capacity (192 GB, sets what fits). Drawn per unit; a typical node is 8×.

Specs

Memory
192 GB HBM3
Bandwidth
5.30 TB/s
FP16 dense
1307 TFLOPS
FP8 dense
2615 TFLOPS
Power
750 W
Form factor
oam
Interconnect
infinity-fabric
Released
2023-12

What it runs (8× unit, Q4_K_M, 4K context)

Model Fits? Decode ceiling
Llama 3.1 8B Instruct yes ~7160 tok/s
Llama 3.3 70B Instruct yes ~887 tok/s
Qwen 2.5 72B Instruct yes ~862 tok/s
DeepSeek-V3 yes ~1716 tok/s

Ceiling is the theoretical rooflineruntimeA performance model that bounds throughput by either compute or memory bandwidth, whichever is the limiting resource for an operation's arithmetic intensity. Open full entry ; open the explorer to set quant, context, and runtime and see the realistic range.

Sources