Dell Pro Max GB10 vs. AMD Ryzen AI Max+ Framework Desktop For Llama.cpp, OpenCL & Vulkan Compute

Written by Michael Larabel in Computers on 7 January 2026 at 11:42 AM EST. Page 3 of 5. 90 Comments.

👁 Llama.cpp benchmark with settings of Backend: ROCm Or CUDA, Model: gpt-oss-20b-Q8_0, Test: Text Generation 128. Dell Pro Max GB10 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: ROCm Or CUDA, Model: gpt-oss-20b-Q8_0, Test: Prompt Processing 512. Dell Pro Max GB10 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: ROCm Or CUDA, Model: gpt-oss-20b-Q8_0, Test: Prompt Processing 2048. Dell Pro Max GB10 was the fastest.

For GPT-OSS 20B, the Dell Pro Max with GB10 continued with dominating performance over the Framework Desktop with Ryzen AI Max+ 395 Strix Halo. The Dell Pro Max with GB10 was delivering the best performance-per-Watt. At best for the Framework Desktop it could offer similar performance-per-dollar to the Dell Pro Max with GB10 when running GPT-OSS 20B.

👁 Llama.cpp benchmark with settings of Backend: ROCm Or CUDA, Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 2048. Dell Pro Max GB10 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: ROCm Or CUDA, Model: Qwen3-8B-Q8_0, Test: Prompt Processing 2048. Dell Pro Max GB10 was the fastest.

Across the variety of models tested with Llama.cpp, the Dell Pro Max with GB10 was offering far better performance than AMD's Strix Halo in the Framework Desktop and also delivering the best performance-per-Watt even with higher wall power. Typically the Dell Pro Max with GB10 even with its much higher price tag was delivering competitive value to the Framework Desktop with the flagship Strix Halo APU.

90 Comments - Next Page

Page: 1 2 3 4 5 Next Page

URL: https://www.phoronix.com/review/dell-pro-max-gb10-llama-cpp/3

⇱ Dell Pro Max GB10 vs. AMD Ryzen AI Max+ Framework Desktop For Llama.cpp, OpenCL & Vulkan Compute - Phoronix

Dell Pro Max GB10 vs. AMD Ryzen AI Max+ Framework Desktop For Llama.cpp, OpenCL & Vulkan Compute