← GPU latency results

gpu/isolated_ops

Legacy report: ingestion provenance is unrecorded; no raw artifact is inferred.

Original report

gpu/isolated_ops: CUDA public API benchmark

Measured cells: 1078. Unsupported cells: 62, including 50 numerically verified but timing-excluded cells. Unsupported cells do not count as measured coverage.

Raw run: data/results/nvidia-gpu/gpu/isolated_ops/20261013_212123/.

Values are median (IQR) µs per operation. Samples time batches of declared operations plus device synchronization; retained output memory is limited to 256 MiB. Batch counts and raw durations are preserved in JSONL.

Setup, input transfers, library/plan initialization, JIT compilation, validation downloads and output destruction are outside timing. Intrinsic output allocation is inside timing; output-reuse rows instead use preallocated destinations across samples. Accumulating destinations are restored from prepared GPU seeds and synchronized before every batch clock, including initialization and calibration. Direct and eager each use one entered session across samples; direct FFT uses cached FftExecutor. JAX eager and compiled JIT are separate arms.

All measured output elements are checked against independently generated NumPy/SciPy reference results before and after sampling. Unsupported rows include unavailable APIs and APIs whose internal setup cannot be excluded; these are not performance measurements. Short batches limited by output memory remain identifiable in the sampling table.

I32/I64 outputs require exact integer equality without conversion to floating point. Division and remainder truncate toward zero; axis-1 sum preserves the input dtype. Arithmetic fixtures avoid overflow, and I64 unary/order fixtures include values above 2**53 to detect loss of integer precision. A zero verification value in these rows means exact equality.

Sampling