The Open-Source AI Stack
RSS
All hardware

Hardware · Workstation

AMD Radeon AI PRO R9700

AMD

RDNA-generation workstation card aimed at local AI, with 32 GB and a ROCm software path. AMD lists dense and sparse matrix figures side by side; dense values used here.

Compute units 191 TF FP16 640 GB/s memory bus 32 GB GDDR6 (VRAM)
Workstation. Bus width tracks bandwidth (640 GB/s, sets decode speed); the box tracks capacity (32 GB, sets what fits).

Specs

Memory
32 GB GDDR6
Bandwidth
640 GB/s
FP16 dense
191 TFLOPS
FP8 dense
383 TFLOPS
Power
300 W
Form factor
pcie
Interconnect
pcie
Released
2025-07

What it runs (single unit, Q4_K_M, 4K context)

Model Fits? Decode ceiling
Llama 3.1 8B Instruct yes ~127 tok/s
Llama 3.3 70B Instruct no —
Qwen 2.5 72B Instruct no —
DeepSeek-V3 no —

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