← GPU latency results

gpu/routine_command_wall

profile: nvidia-gpuno machine-readable summary

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

Original report

GPU command-wall observations

Wall time includes preparation, validation, sampling and reporting. These are end-to-end workflow costs; operation-performance ratios are not inferred. Missing phase instrumentation is displayed as unavailable, not zero. Filtered or unsuccessful runs do not establish full suite coverage.

gpu/dense / 20261013_234113

Raw run: data/results/nvidia-gpu/gpu/dense/20261013_234113.

tenferro-rs 9655b4e0daced17e0aa35a30d35710b069e10644; harness 9e89674e200526e294a01595152425befafd53f5. GPU: NVIDIA GeForce RTX 3060 (GPU-a78d5217-eba3-72c2-3d5b-8ae496ebbc2e).

Command through this suite: 69.863 s; collection including build/release delays: 61.754 s; report generation/copy: 0.139 s.

Scope: script clock-read through this suite report copy; devcontainer CLI startup and final wall-record write excluded.

Backend Build s Process s (startup/setup/checks/samples)
tenferro-cuda-eager 5.016 16.183
pytorch-cuda 0.000 14.185
jax-cuda 0.000 20.066

Case phases below exclude process startup/JSONL writing. Preparation includes input generation, transfers and reference preparation; lazy initialization can occur in that phase. Sampler walls include untimed batch preparation/cleanup. Command totals are cumulative across suites in one invocation, so do not sum those totals.

Problem Backend Status Backend init s Prepare s Compile s Probe/init s Warmup s Calibration s Sampling s Validation s
dense_matmul_f64_3072 tenferro-cuda-eager ok 0.000 2.376 — 0.343 0.941 0.314 2.196 2.587
dense_batched_matmul_f64_b16_32 tenferro-cuda-eager ok 0.000 0.009 — 0.000 0.000 0.006 0.018 0.000
dense_qr_f64_1536 tenferro-cuda-eager ok 0.000 0.041 — 0.344 0.191 0.095 0.477 0.655
dense_solve_f64_512_rhs16 tenferro-cuda-eager ok 0.000 0.252 — 0.401 0.008 0.004 0.020 0.002
dense_solve_f64_2048_rhs16 tenferro-cuda-eager ok 0.000 0.102 — 0.050 0.097 0.048 0.242 0.057
dense_svd_f64_256 tenferro-cuda-eager ok 0.000 0.007 — 0.133 0.164 0.082 0.409 0.006
dense_eigh_f64_1024 tenferro-cuda-eager ok 0.000 2.018 — 0.085 0.158 0.075 0.372 0.189
dense_matmul_f64_3072 pytorch-cuda ok 1.686 1.761 0.000 0.357 0.941 0.314 2.196 1.059
dense_batched_matmul_f64_b16_32 pytorch-cuda ok 0.000 0.002 0.000 0.000 0.000 0.007 0.021 0.001
dense_qr_f64_1536 pytorch-cuda ok 0.000 0.080 0.000 0.131 0.190 0.095 0.475 0.844
dense_solve_f64_512_rhs16 pytorch-cuda ok 0.000 0.014 0.000 0.019 0.006 0.003 0.015 0.005
dense_solve_f64_2048_rhs16 pytorch-cuda ok 0.000 0.148 0.000 0.048 0.096 0.048 0.240 0.068
dense_svd_f64_256 pytorch-cuda ok 0.000 0.003 0.000 0.087 0.168 0.084 0.419 0.096
dense_eigh_f64_1024 pytorch-cuda ok 0.000 0.066 0.000 0.084 0.149 0.075 0.373 0.399
dense_matmul_f64_3072 jax-cuda ok 1.020 0.891 5.751 0.275 0.815 0.272 1.902 1.455
dense_batched_matmul_f64_b16_32 jax-cuda ok 0.000 0.026 0.365 0.001 0.000 0.006 0.019 0.002
dense_qr_f64_1536 jax-cuda ok 0.000 0.092 0.213 0.103 0.198 0.097 0.486 1.091
dense_solve_f64_512_rhs16 jax-cuda ok 0.000 0.018 0.187 0.008 0.007 0.003 0.017 0.008
dense_solve_f64_2048_rhs16 jax-cuda ok 0.000 0.203 0.198 0.049 0.098 0.049 0.245 0.125
dense_svd_f64_256 jax-cuda ok 0.000 0.006 0.234 0.088 0.165 0.083 0.413 0.101
dense_eigh_f64_1024 jax-cuda ok 0.000 0.076 0.199 0.082 0.151 0.076 0.378 0.449
## gpu/einsum / 20261013_234113

Raw run: data/results/nvidia-gpu/gpu/einsum/20261013_234113.

tenferro-rs 9655b4e0daced17e0aa35a30d35710b069e10644; harness 9e89674e200526e294a01595152425befafd53f5. GPU: NVIDIA GeForce RTX 3060 (GPU-a78d5217-eba3-72c2-3d5b-8ae496ebbc2e).

Command through this suite: 116.029 s; collection including build/release delays: 39.178 s; report generation/copy: 0.137 s.

Scope: script clock-read through this suite report copy; devcontainer CLI startup and final wall-record write excluded.

Backend Build s Process s (startup/setup/checks/samples)
tenferro-cuda-eager 0.000 9.321
pytorch-cuda 0.000 10.278
jax-cuda 0.000 13.284

Case phases below exclude process startup/JSONL writing. Preparation includes input generation, transfers and reference preparation; lazy initialization can occur in that phase. Sampler walls include untimed batch preparation/cleanup. Command totals are cumulative across suites in one invocation, so do not sum those totals.

Problem Backend Status Backend init s Prepare s Compile s Probe/init s Warmup s Calibration s Sampling s Validation s
einsum_bin_matmul_3072_f64 tenferro-cuda-eager ok 0.000 2.329 — 0.343 0.946 0.315 2.208 2.692
einsum_bin_matmul_3072_f64 pytorch-cuda ok 1.722 1.768 0.000 0.350 0.946 0.316 2.208 2.011
einsum_bin_matmul_3072_f64 jax-cuda ok 1.002 0.847 5.913 0.274 0.815 0.272 1.902 1.345
## gpu/tensornetwork / 20261013_234113

Raw run: data/results/nvidia-gpu/gpu/tensornetwork/20261013_234113.

tenferro-rs 9655b4e0daced17e0aa35a30d35710b069e10644; harness 9e89674e200526e294a01595152425befafd53f5. GPU: NVIDIA GeForce RTX 3060 (GPU-a78d5217-eba3-72c2-3d5b-8ae496ebbc2e).

Command through this suite: 155.349 s; collection including build/release delays: 32.441 s; report generation/copy: 0.137 s.

Scope: script clock-read through this suite report copy; devcontainer CLI startup and final wall-record write excluded.

Backend Build s Process s (startup/setup/checks/samples)
tenferro-cuda-eager 0.000 7.695
pytorch-cuda 0.000 8.246
jax-cuda 0.000 10.196

Case phases below exclude process startup/JSONL writing. Preparation includes input generation, transfers and reference preparation; lazy initialization can occur in that phase. Sampler walls include untimed batch preparation/cleanup. Command totals are cumulative across suites in one invocation, so do not sum those totals.

Problem Backend Status Backend init s Prepare s Compile s Probe/init s Warmup s Calibration s Sampling s Validation s
tensornetwork_permutation_optimized_f32 tenferro-cuda-eager ok 0.000 2.102 — 3.448 0.373 0.118 1.144 0.002
tensornetwork_permutation_optimized_f32 pytorch-cuda ok 1.701 0.048 0.000 0.438 1.011 0.336 3.372 0.339
tensornetwork_permutation_optimized_f32 jax-cuda ok 0.970 0.095 3.327 0.452 0.859 0.286 2.860 0.386