AMD Adds MiniCPM-V 4.7 Support for Radeon GPUs and Ryzen AI Systems

AMD’s MiniCPM-V 4.7 support covers Radeon GPUs, Ryzen AI Max integrated graphics and Ryzen AI NPUs, giving developers listed GPU and NPU paths for running the multimodal model locally. The support was available when ModelBest released the model on Oct. 14, 2026.

The model handles image-and-text inference and comes in 1B and 35B-A3B variants. Both versions completed inference on three tested configurations: a Radeon AI PRO R9700, a Radeon PRO W7900D and the integrated Radeon 8060S in a Ryzen AI Max 395 system. The outputs matched the test image.

AMD’s GPU validation covered functional results, but it did not include reported measurements for throughput, time to first token or GPU memory use. The listed configurations describe the tested systems; they do not establish compatibility with every Radeon GPU.

For NPU inference, FastFlowLM version 1.0.8 supports MiniCPM-V-4.7-1B on Ryzen AI PCs with XDNA 2 NPUs. The listed systems include Ryzen AI Max, Ryzen AI 400 and Ryzen AI 300 series devices. The NPU package uses Q4_K quantization in NPU2 format, with two files totaling about 2.2 GB.

The NPU example used a Ryzen AI Max+ 395 system running Windows 11, FastFlowLM 1.0.8 and NPU driver version 32.0.20102.3930. In that test, time to first token for a single image was approximately 1.2 seconds. Decoding ran at about 44 tokens per second with a 1,000-token context and about 23 tokens per second with a 32,000-token context. Those performance figures apply to that test setup.

The GPU workflow uses a llama.cpp build with a HIP backend. The NPU workflow uses FastFlowLM and the quantized package. AMD’s reported validation results do not provide a performance comparison between the two paths.