AMD EPYC 8635P "Sorano" Benchmarks: Significant Upgrade Opportunity For EPYC 8004 Servers

Written by Michael Larabel in Processors on 3 June 2026 at 11:36 AM EDT. Page 4 of 8. Add A Comment.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Text Generation 128. EPYC 8635P was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Llama-3.1-Tulu-3-8B-Q8_0, Test: Prompt Processing 2048. EPYC 8635P was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Text Generation 128. EPYC 8635P was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 2048. EPYC 8635P was the fastest.

For those evaluating the AMD EPYC 8005 series for edge computing with AI inferencing workloads, Llama.cpp showed the very healthy performance gains from Siena to Sorano at the top-end. If your EPYC 8005 server platform supports up to DDR5-6400 memory rather than a DDR5-5200 limit, you will find even greater performance gains than these results showed.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Text Generation 128. EPYC 8635P was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 2048. EPYC 8635P was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Qwen3.5-9B-Q8_0, Test: Text Generation 128. EPYC 8635P was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Qwen3.5-9B-Q8_0, Test: Prompt Processing 2048. EPYC 8635P was the fastest.

Even with the higher default TDP of the EPYC 8635P, that new processor was often delivering better performance-per-Watt over the EPYC 8534P.

👁 oneDNN benchmark with settings of Harness: Convolution Batch Shapes, Data Type: BF16, Engine: CPU. EPYC 8635P was the fastest.

👁 oneDNN benchmark with settings of Harness: Convolution Batch Shapes, Data Type: INT8, Engine: CPU. EPYC 8635P was the fastest.

👁 oneDNN benchmark with settings of Harness: Matmul Batch Shapes, Data Type: BF16, Engine: CPU. EPYC 8635P was the fastest.

👁 oneDNN benchmark with settings of Harness: Matmul Batch Shapes, Data Type: INT8, Engine: CPU. EPYC 8635P was the fastest.

The EPYC 8635P was also delivering great generational gains in other AI workloads like the oneDNN library.

Add A Comment - Next Page

Page: 1 2 3 4 5 6 7 8 Next Page

URL: https://www.phoronix.com/review/amd-epyc-8635p/4

⇱ AMD EPYC 8635P "Sorano" Benchmarks: Significant Upgrade Opportunity For EPYC 8004 Servers Review - Phoronix

AMD EPYC 8635P "Sorano" Benchmarks: Significant Upgrade Opportunity For EPYC 8004 Servers