Announced as the card for professional AI at the edge, the Radeon AI PRO R9700 uses 4nm Navi 48, bringing 64 compute units for a total of 4480 Stream processors, and 128 AI accelerators. AMD also paired it up with 32GB of 20Gbps GDDR6 memory on a 256-bit memory interface, which adds up to 640GB/s of memory bandwidth. It needs two 8-pin PCIe power connectors, and its TDP is set at 300W. According to AMD, the performance peaks at 96 TFLOPs of FP16 throughput, and up to 1,531 AI TOPS (INT4 sparse).

The 32GB VRAM should definitely help with larger AI models, and AMD was keen to put it against GeForce RTX 5080, which is not actually the best comparison, but it does end up significantly faster. The Radeon AI PRO R9700 also supports driver-level scalability for up to four GPUs, increasing the memory to 128GB.

The Radeon AI PRO R9700 is coming in July 2025, and AMD has yet to give any details regarding the price.