inference/research
verifiedhigh · 0.85updated 2026-10-17

llama.cpp SYCL backend documentation

repo

Fields

id
sources/llama-cpp-sycl-backend-doc
Publisher
ggml-org
Kind
repo
Accessed
2026-10-17
notes
The SYCL backend's own docs page on master, read 2026-10-17. It is a living document with no pinned revision, so numbers taken from it are the state of master on that date, not a release. The page carries the backend's changelog (the News section), the runtime environment-variable contract, the verified-device table, and the design rules. The one measured number this repo records from it — about 50% faster pp512 on Qwen3-30B-A3B UD-IQ3_XXS on an Arc Pro B60 from the XMX dequant-GEMM paths — is stated in the GGML_SYCL_XMX_GATHER_TYPES entry and is the backend authors' own example, not an independent benchmark.

Notes

The SYCL backend's own docs page on master, read 2026-10-17. It is a living document with no pinned revision, so numbers taken from it are the state of master on that date, not a release. The page carries the backend's changelog (the News section), the runtime environment-variable contract, the verified-device table, and the design rules. The one measured number this repo records from it — about 50% faster pp512 on Qwen3-30B-A3B UD-IQ3_XXS on an Arc Pro B60 from the XMX dequant-GEMM paths — is stated in the GGML_SYCL_XMX_GATHER_TYPES entry and is the backend authors' own example, not an independent benchmark.

Cross-references

This record references no other records.