AMD Ryzen AI Max+ "Strix Halo" Performance With ROCm 7.0

Written by Michael Larabel in Display Drivers on 22 September 2025 at 10:30 AM EDT. Page 2 of 5. 34 Comments.

👁 vLLM benchmark with settings of Test: Hermes-3-Llama-3.2-8B Latency. Strix Halo was the fastest.

The official rocm/vllm Docker container on Docker Hub still has not been updated for the ROCm 7.0 release. But there are the rocm/vllm-dev container images that continue to see new nightly builds. The vllm-dev container was working fine on this Strix Halo desktop. Such as for Llama 3.2 8B running fine as a starter test case.

👁 vLLM benchmark with settings of Test: deepseek-moe-16b-chat Latency. Strix Halo was the fastest.

DeepSeek MOE 16B chat was also working fine on Strix Halo with the AMDGPU DKMS package on Ubuntu 24.04 LTS.

👁 vLLM benchmark with settings of Test: Hermes-3-Llama-3.1-8B Throughput. Strix Halo was the fastest.

👁 vLLM benchmark with settings of Test: DeepSeek-R1-Distill-Qwen-7B Latency. Strix Halo was the fastest.

👁 vLLM benchmark with settings of Test: DeepSeek-R1-Distill-Llama-8B Latency. Strix Halo was the fastest.

👁 vLLM benchmark with settings of Test: DeepSeek-R1-Distill-Qwen-7B Throughput. Strix Halo was the fastest.

👁 vLLM benchmark with settings of Test: DeepSeek-R1-Distill-Llama-8B Throughput. Strix Halo was the fastest.

DeepSeek R1 and other tested models were also working on the vLLM nightly container with this AMD Ryzen AI Max+ system.

34 Comments - Next Page

Page: 1 2 3 4 5 Next Page

URL: https://www.phoronix.com/review/amd-rocm-7-strix-halo/2

⇱ AMD Ryzen AI Max+ "Strix Halo" Performance With ROCm 7.0 - Phoronix

AMD Ryzen AI Max+ "Strix Halo" Performance With ROCm 7.0