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 4 of 5. 34 Comments.

Llama 3.1 was also showing much better performance with Vulkan on Strix Halo for Llama.cpp 6401 compared to HIP defaults.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Text Generation 128). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 512). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 1024). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 2048). Vulkan was the fastest.

Mistral was also running fine with ROCm 7.0 on Strix Halo but the best performance was found simply using Vulkan with the stock RADV Vulkan API driver.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Text Generation 128). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 512). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 1024). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 2048). Vulkan was the fastest.

Some of these Vulkan leads is much more significant than what we have seen out of ROCm HIP vs. Vulkan for Llama.cpp on discrete Radeon GPUs.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 512). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 1024). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 2048). Vulkan was the fastest.

Llama.cpp with ROCm 7.0 installed on Ubuntu 24.04 LTS was working with the AMD Ryzen AI Max+ 395 (Strix Halo) in the Framework Desktop. But the best performance continued to be with using the Vulkan back-end instead when testing in the default configuration.

34 Comments - Next Page

Page: 1 2 3 4 5 Next Page

URL: https://www.phoronix.com/review/amd-rocm-7-strix-halo/4

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

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