AMD ROCm 7.1 vs. RADV Vulkan For Llama.cpp With The Radeon AI PRO R9700

Written by Michael Larabel in Display Drivers on 18 November 2025 at 10:45 AM EST. Page 3 of 4. 24 Comments.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Text Generation 128). Vulkan was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 512). AMD ROCm HIP was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 1024). AMD ROCm HIP was the fastest.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 2048). AMD ROCm HIP was the fastest.

For Llama 3.1 8B, the AMD ROCm back-end returned to delivering the best prompt processing performance at least for this AMD RDNA4 graphics card used for testing. Even so it remained impressive how well the Vulkan API is working out for AI workloads these days.

👁 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). AMD ROCm HIP was the fastest.

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

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

Mistral came out faster on ROCm over Vulkan for prompt processing.

👁 Llama.cpp benchmark with settings of Backend Comparison (Model: DeepSeek-R1-Distill-Llama-8B-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). AMD ROCm HIP was the fastest.

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

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

For the DeepSeek-R1 testing, the ROCm back-end was a fair amount faster than Vulkan.

24 Comments - Next Page

Page: 1 2 3 4 Next Page

URL: https://www.phoronix.com/review/rocm-71-llama-cpp-vulkan/3

⇱ AMD ROCm 7.1 vs. RADV Vulkan For Llama.cpp With The Radeon AI PRO R9700 - Phoronix

AMD ROCm 7.1 vs. RADV Vulkan For Llama.cpp With The Radeon AI PRO R9700