Linux 6.12 To Linux 6.18 LTS Upgrade Offers Worthwhile Benefits For 5th Gen AMD EPYC

Written by Michael Larabel in Software on 19 December 2025 at 10:38 AM EST. Page 3 of 8. 3 Comments.

With the Llama.cpp software the text generation performance was unchanged but there were some significant improvements found in the prompt processing performance when going from Linux 6.12 LTS to Linux 6.18 LTS.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 512. v6.19 18 Dec was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 1024. v6.18 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: granite-3.0-3b-a800m-instruct-Q8_0, Test: Prompt Processing 2048. v6.18 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 512. v6.18 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 1024. v6.18 was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: Mistral-7B-Instruct-v0.3-Q8_0, Test: Prompt Processing 2048. v6.18 was the fastest.

The CPU power consumption was slightly higher with Linux 6.18+ but with far better prompt processing performance to deliver a nice win for driving the performance-per-Watt.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: gpt-oss-20b-Q8_0, Test: Text Generation 128. v6.19 18 Dec was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: gpt-oss-20b-Q8_0, Test: Prompt Processing 512. v6.19 18 Dec was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: gpt-oss-20b-Q8_0, Test: Prompt Processing 1024. v6.19 18 Dec was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: gpt-oss-20b-Q8_0, Test: Prompt Processing 2048. v6.19 18 Dec was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 1024. v6.19 18 Dec was the fastest.

👁 Llama.cpp benchmark with settings of Backend: CPU BLAS, Model: DeepSeek-R1-Distill-Llama-8B-Q8_0, Test: Prompt Processing 2048. v6.19 18 Dec was the fastest.

For anyone doing CPU-based AI inferencing on the AMD EPYC 9005 series servers, moving to the newer Linux kernel was showing a clear win with Llama.cpp performance.

3 Comments - Next Page

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

URL: https://www.phoronix.com/review/linux-618-lts-amd-epyc/3

⇱ Linux 6.12 To Linux 6.18 LTS Upgrade Offers Worthwhile Benefits For 5th Gen AMD EPYC - Phoronix

Linux 6.12 To Linux 6.18 LTS Upgrade Offers Worthwhile Benefits For 5th Gen AMD EPYC