intel/ipex-llm issue #12991: Intel Arc B580 llama-bench Vulkan vs SYCL vs IPEX-LLM token-generation tables
forum
Fields
id
sources/gh-intel-ipex-llm-12991-arc-b580-llama-benchPublisher
Intel IPEX-LLM (GitHub, archived 2026-01-28)
Kind
forum
Published
2025-03-21
Accessed
2026-10-11
notes
A GitHub issue thread in which the reporter (TabNahida) posts full llama-bench tables for an Intel Arc B580 12GB on Windows 11 24H2 / Ryzen 9 9950X, comparing the mainline llama.cpp Vulkan and SYCL backends against the IPEX-LLM portable SYCL build, for DeepSeek-R1-Distill-Qwen 1.5B/7B/14B Q4_K_M. The rows are verbatim llama-bench output (pp512/1024/2048 and tg128/256/512 with +/- stdev), which is why the source is citable despite being a forum post. It is a single reporter on a single card, and the repository was archived on 2026-01-28 (read-only), so this captures the March-2025 software state. measured_by is filed as 'third_party' (community measurement); the schema has no 'community' member.
Notes
A GitHub issue thread in which the reporter (TabNahida) posts full llama-bench tables for an Intel Arc B580 12GB on Windows 11 24H2 / Ryzen 9 9950X, comparing the mainline llama.cpp Vulkan and SYCL backends against the IPEX-LLM portable SYCL build, for DeepSeek-R1-Distill-Qwen 1.5B/7B/14B Q4_K_M. The rows are verbatim llama-bench output (pp512/1024/2048 and tg128/256/512 with +/- stdev), which is why the source is citable despite being a forum post. It is a single reporter on a single card, and the repository was archived on 2026-01-28 (read-only), so this captures the March-2025 software state. measured_by is filed as 'third_party' (community measurement); the schema has no 'community' member.
Cross-references
This record references no other records.