Repository navigation
[hardware report] 2026-10-05-13th-gen-intel-r-core-tm-i5-13500-sycl #2116
Description
Activity
Thank you! Every SYCL twin returned the CPU's scores on the UHD 770, which is the first Xe-LP result we have. The one failure,
test_sycl_kernel_scratch, is real: one kernel (ssimulacra2_sycl) needs more registers than Xe-LP has and spills to scratch memory, which we do not allow (ADR-1395). We are reworking that kernel; when a new tester image is out, we will ask you to run it once more.A correction to my closing note: the issue was closed for bookkeeping, because the finding is tracked as
T-SYCL-ROW-KERNELS-SG16-OTHER-DEVICES-2026-10-02under #1721. It is not fixed yet: no build since your report changes the sub-group-16 row kernels on Xe-LP / UHD 770, so re-running the current image would give the same result. We'll mention you here when a tester image carries the fix. Thanks for testing.
report.json
vmaf-tester-report: verdict fail
cpu: 13th Gen Intel(R) Core(TM) i5-13500 (x86_64); dispatch: sse2, ssse3, sse4.1, avx2
dispatch equivalence: identical
reference equivalence: identical
metal equivalence: not_applicable
metal parity gate: not_applicable
metal state rows: not_applicable
unit tests: pass (passed 37, failed 0, skipped 0)
golden gate: not_applicable (passed 0, failed 0, skipped 0)
gpu (sycl): fail, path drm
device 0 Intel(R) UHD Graphics 770 (xe-lp 12.2.0): fail; twins identical, gate pass, tests fail (68 passed, 1 failed, 0 skipped), test_sycl_kernel_scratch fail; state rows 0 passing, 1 failing, 0 not measured
failed: gpu
file name for a pull request: docs/hardware-reports/2026-10-05-13th-gen-intel-r-core-tm-i5-13500-sycl.json
not exercised: tiny-AI / ONNX Runtime; Metal; CUDA and HIP twins; NEON and SVE2 kernels; AVX-512 kernels; golden check; SYCL test test_cuda_parity_gate_default_run; SYCL twins with options in the default-option run
Which package did you run?
Intel GPU image (tester-sycl)
Machine
i5 13500
Credit (optional)
No response
report-gpu.json