@graphty/webgpu-graph-algorithms 0.6.9 → 0.6.10

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (2) hide show
  1. package/README.md +104 -53
  2. package/package.json +11 -8
package/README.md CHANGED
@@ -683,43 +683,65 @@ measures the same crossover on a consumer's own device.
683
683
 
684
684
  ## Performance
685
685
 
686
- Regenerated from the last session of each baseline under `benchmarks/results/` (`nvidia-lovelace-driver580.json`, the
687
- dev box; `gpu-linux-t4.json`, the CI lane) by the procedure recorded in `docs/decisions/G3.md` appendix A; the targets
688
- are the T-table of plan section 10.4. A missed target is re-fixed by a recorded owner decision in
689
- `docs/decisions/G<n>.md`, never relaxed silently.
686
+ The targets are the T-table of plan section 10.4, kept in `benchmarks/results/targets.json` with one entry per
687
+ benchmark row. They bind the reference card, the Tesla T4 of the GPU lane: on that class `bench:compare` fails the lane
688
+ when a row misses its target, unless the row is one of the class's known misses, which carry the reason and the
689
+ decision record beside the target. Every other runner class reports met or missed and does not gate. The two tables
690
+ below are generated from that file and the checked-in results (`pnpm run bench:readme`, which a node test checks), so
691
+ they are never edited by hand. A missed target is re-fixed by a recorded owner decision in `docs/decisions/G<n>.md`,
692
+ never relaxed silently.
690
693
 
691
694
  ### The dev box (nvidia-lovelace-driver580)
692
695
 
693
- Measured on nvidia-lovelace-driver580 (NVIDIA: 580.173.02 580.173.2.0), session 2026-09-20T01:06:48.656Z, medians of 5 runs; Chromium: nvidia / lovelace (nvidia-lovelace-driver0, the description is redacted by Chromium), session 2026-09-16T02:18:11.896Z. The T-5 (100k), T-6 and T-7 rows are from the later session 2026-09-21T06:15:17.027Z (the file's last session, the current baseline, the one the crossover re-check reads; its Chromium session 2026-09-21T05:48:14.190Z), whose other rows are within 1.25x of this table's.
694
-
695
- | Id | What | Target | Measured |
696
- | ---- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
697
- | T-1 | Upload of the 100k / 1M weighted hot prefix (16.4 MB); 1M / 10M (164 MB) | <= 10 ms; <= 100 ms | 5.980 ms; 125.738 ms |
698
- | T-2 | `degree` + 400 KB readback at 100k (core resident), Node | <= 2 ms | 0.870 ms |
699
- | T-3 | Empty submit + 4-byte `readU32` round trip, Dawn | <= 0.1 ms | 0.181 ms |
700
- | T-4 | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k; at 16k | <= 1 ms; <= 2 ms | 0.586 ms; 1.053 ms |
701
- | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium (Node in brackets) | <= 6 ms | 2.400 ms (0.726 ms) |
702
- | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium (Node in brackets) | <= 12 ms | 3.700 ms (2.087 ms) |
703
- | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D; at 1M 2D; at 100k 3D | <= 10 ms; <= 100 ms; <= 20 ms | 0.634 ms; 5.414 ms; 1.306 ms |
704
- | T-7 | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M | <= 15 ms | 1.493 ms |
705
- | T-8 | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M; at 1M / 10M | <= 150 ms; <= 1.5 s | 17.204 ms; 198.645 ms |
706
- | T-9 | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M (100k / 1M in brackets) | <= 100 ms | 145.970 ms (12.613 ms) |
707
- | T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k; at 100k (`repulsion: "exact"`) | recorded | 0.617 ms; 16.367 ms |
708
- | T-10 | BFS direction-optimizing (`breadthFirstSearch`), wall end to end including the upload, from node 0 of the undirected RMAT at 1M / 10M (577,681 nodes reached, 6 levels; 100k / 1M in brackets; top-down 233.064 ms / 97.259 ms) | <= 100 ms | 237.388 ms (101.183 ms) in the baseline session; 89.212 ms (14.925 ms) re-measured 2026-09-25 at load average 7.3 after the direct-dispatch change and the m_f fix of issue #391, top-down 95.014 ms (14.744 ms); a sample one minute earlier at load average 10.5 read 134.674 ms (15.183 ms) auto and 99.002 ms (14.630 ms) top-down -- see below and G8-F5 / G8-F21 of `docs/decisions/G8.md` |
709
- | T-10 | BFS on the 1000 x 1000 grid from a corner (1,999 levels; 64 `mapAsync` calls = ceil(levels / 32) + 1), wall including the upload | <= 1.5 s | 5680.192 ms in the baseline session; 427.235 ms re-measured 2026-09-25 under load after the direct-dispatch change, 542.370 ms (536.980 ms one minute earlier) after the m_f fix at load average 7.3-10.5, 382.680 ms at load average 2.5 once `bfs-next-degree` took a capped grid, MET -- see below and G8-F5 of `docs/decisions/G8.md` |
710
- | T-10 | `sssp` (the near-far queue), wall end to end including the upload, random f32 weights in [0.1, 10), at 1M / 10M (100k / 1M in brackets) | recorded | 310.351 ms (118.064 ms); 271.040 ms (50.871 ms) re-measured 2026-09-25 under load; 275.474 ms (48.326 ms) after the m_f fix at load average 7.3 |
711
-
712
- Three rows miss their target in this session: the 1M / 10M upload (125.7 ms against 100 ms, the open owner decision of
713
- `docs/decisions/G1.md` section 7), the empty-submit round trip (0.181 ms against 0.1 ms: the row is measured after the
714
- `upload` group, whose CPU-heavy setup lets the SM clock fall to its idle state; the same row measures 0.041-0.075 ms at
715
- the working clock -- finding G3-F2 of `docs/decisions/G3.md` section 10), and WCC at 1M / 10M (146.0 ms against
716
- 100 ms: the row is wall end to end from a released core, so it carries the same 164 MB upload T-1 times at 125.7 ms;
717
- with the core resident the same call takes 11-16 ms). The T-9 target is therefore below
718
- the T-1 upload it includes, an owner decision for `docs/decisions/G7.md`. The `pagerank` rows run all 100 iterations
719
- (`tolerance: 0`): at the NetworkX tolerance of 1e-6 the seeded G(n, m) input converges from the uniform start in one to
720
- four iterations, which would time one pull and call it a hundred. The T-14 row is the `layout-fr` group of the later
721
- session 2026-09-20T19:25:37.311Z (the file's last session, the current baseline), whose other rows are within 1.08x of
722
- this table's; the spring-electrical preset measures 0.648 ms at 10k and 18.154 ms at 100k in the same session.
696
+ <!-- targets-table:nvidia-lovelace-driver580 -->
697
+
698
+ Generated by `pnpm run bench:readme` from `benchmarks/results/targets.json`. Each figure is the best median the row has recorded, the baseline `bench:compare` pins: Node rows across the 7 session(s) of `benchmarks/results/nvidia-lovelace-driver580.json`, Chromium rows across the 3 of `benchmarks/results/nvidia-lovelace-driver0.json`.
699
+
700
+ | Id | What | Target | Measured | Status |
701
+ | ---- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ----------- | -------------- |
702
+ | T-1 | Upload of the 100k / 1M weighted hot prefix (16.4 MB), Node | <= 10 ms | 5.734 ms | met |
703
+ | T-1 | Upload of the 1M / 10M weighted hot prefix (164 MB), Node | <= 100 ms | 122.254 ms | missed (noisy) |
704
+ | T-2 | `degree` + 400 KB readback at 100k (core resident), Node | <= 2 ms | 0.838 ms | met |
705
+ | T-3 | Empty submit + 4-byte `readU32` round trip, Dawn in Node | <= 0.1 ms | 0.062 ms | met |
706
+ | T-3 | Empty submit + 4-byte `readU32` round trip, Chromium (the <= 0.1 ms target is Dawn's) | recorded | 0.170 ms | recorded |
707
+ | T-4 | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k | <= 1 ms | 0.586 ms | met |
708
+ | T-4 | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 16k | <= 2 ms | 1.052 ms | met |
709
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium | <= 6 ms | 2.400 ms | met |
710
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Node | recorded | 0.713 ms | recorded |
711
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium | <= 12 ms | 3.700 ms | met |
712
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Node | recorded | 1.769 ms | recorded |
713
+ | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D | <= 10 ms | 0.574 ms | met |
714
+ | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 1M 2D | <= 100 ms | 5.303 ms | met |
715
+ | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 3D | <= 20 ms | 1.279 ms | met |
716
+ | T-7 | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M | <= 15 ms | 1.475 ms | met |
717
+ | T-8 | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M | <= 150 ms | 17.204 ms | met |
718
+ | T-8 | PageRank, 100 iterations, wall end to end including the upload, at 1M / 10M | <= 1500 ms | 197.154 ms | met |
719
+ | T-9 | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M | <= 100 ms | 140.984 ms | missed |
720
+ | T-9 | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 100k / 1M | recorded | 12.613 ms | recorded |
721
+ | T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k | recorded | 0.566 ms | recorded |
722
+ | T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 100k (`repulsion: "exact"`) | recorded | 16.322 ms | recorded |
723
+ | T-14 | Spring-electrical preset, GPU time per iteration (profiler) at 10k | recorded | 0.648 ms | recorded |
724
+ | T-14 | Spring-electrical preset, GPU time per iteration (profiler) at 100k (`repulsion: "exact"`) | recorded | 18.154 ms | recorded |
725
+ | T-10 | BFS direction-optimizing (`breadthFirstSearch`), wall end to end including the upload, from node 0 of the undirected RMAT at 1M / 10M (a contract on the dev box only) | <= 100 ms | 237.388 ms | missed |
726
+ | T-10 | BFS top-down, the same traversal at 1M / 10M | recorded | 233.064 ms | recorded |
727
+ | T-10 | BFS direction-optimizing, the same traversal at 100k / 1M | recorded | 101.183 ms | recorded |
728
+ | T-10 | BFS on the 1000 x 1000 grid from a corner (1,999 levels), wall including the upload (a contract on the dev box only) | <= 1500 ms | 5680.192 ms | missed |
729
+ | T-10 | `sssp` (the near-far queue), wall end to end including the upload, random f32 weights in [0.1, 10), at 1M / 10M | recorded | 310.351 ms | recorded |
730
+ | T-10 | `sssp`, the same run at 100k / 1M | recorded | 118.064 ms | recorded |
731
+
732
+ <!-- /targets-table:nvidia-lovelace-driver580 -->
733
+
734
+ Why the misses miss. The 1M / 10M upload is the open owner decision of `docs/decisions/G1.md` section 7. The
735
+ empty-submit round trip in Node meets its target at the working clock, but a session that measures it right after the
736
+ `upload` group reads about 0.15-0.18 ms, because the group's CPU-heavy setup lets the SM clock fall to its idle state
737
+ (finding G3-F2 of `docs/decisions/G3.md` section 10). WCC at 1M / 10M is wall end to end from a released core, so it
738
+ carries the same 164 MB upload T-1 times; with the core resident the same call takes 11-16 ms, and the T-9 target is
739
+ therefore below the T-1 upload it includes, an owner decision for `docs/decisions/G7.md`. The `pagerank` rows run all
740
+ 100 iterations (`tolerance: 0`): at the NetworkX tolerance of 1e-6 the seeded G(n, m) input converges from the uniform
741
+ start in one to four iterations, which would time one pull and call it a hundred. The Chromium round trip (T-3, issue
742
+ #278) is what every interactive frame of a browser layout pays; its figure is the mean of batches of 20 round trips,
743
+ because Chromium quantises `performance.now()` to 100 us, and it was measured on 2026-09-27 at a load average near 70,
744
+ so it is an upper bound.
723
745
 
724
746
  The three T-10 rows are the `bfs` group of session 2026-09-25T10:19:11.536Z (the file's last session, the current
725
747
  baseline; `webgpu` 0.4.0, driver 580.173.02, load average 1.7-2.1, medians of 5). Both T-10 targets are MISSED on the
@@ -783,28 +805,57 @@ GPU time per iteration in the last batch).
783
805
  The first run of the GPU lane (`gpu.yml`, graphty-monorepo run 35316416067, 2026-09-18) on a machine.dev T4 -- one Tesla
784
806
  T4 (16 GB), 4 vCPU of a Xeon Platinum 8259CL, driver 580.126.20 -- wrote this baseline; `scripts/bench-compare.js` fails
785
807
  a later run of the lane whose median AND minimum both exceed 1.35x the best figures this file has ever held, by at
786
- least 2.5 ms. The T-table targets were set on the dev box; the T4 meets
787
- T-4, T-5 and T-6 and misses T-1 (both uploads), T-2, T-3 and T-7 (the 1M attraction pass of the grid tier, 18.165 ms
788
- against 15 ms), which is the class difference of a datacentre card behind a cloud vCPU (host-side copies and submit
808
+ least 2.5 ms, or whose row misses a target that is not one of the known misses listed under the table. The T-table
809
+ targets were set on the dev box; the T4 meets T-4, T-5, T-6 and T-8 and misses T-1 (both uploads), T-2, T-3, T-7 (the
810
+ 1M attraction pass of the grid tier, 18.165 ms against 15 ms) and T-9, which is the class difference of a datacentre card behind a cloud vCPU (host-side copies and submit
789
811
  latency), not a regression: the exact tier's `ms / iteration` is 1.7x the RTX 4070 SUPER's at 10k and 3.0x at 65k, and
790
812
  the grid tier's is 2.8x at 100k 2D and 5.7x at 1M 2D while the attraction pass alone is 12.2x.
791
813
 
792
- Measured on gpu-linux-t4 (NVIDIA: 580.126.20 580.126.20.0), session 2026-09-20T02:29:33.210Z (run 35483512705), medians of 5 runs; Chromium: nvidia / turing (nvidia-turing-driver0, the description is redacted by Chromium), session 2026-09-20T02:28:10.676Z. The T-5 (100k), T-6 and T-7 rows are from the later session 2026-09-22T20:21:53.814Z (GPU lane run 35775999450 on commit 1d2d4dcf, the file's last session and the current baseline of this class; its Chromium session 2026-09-22T20:19:30.117Z).
793
-
794
- | Id | What | Target | Measured |
795
- | ---- | ---------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
796
- | T-1 | Upload of the 100k / 1M weighted hot prefix (16.4 MB); 1M / 10M (164 MB) | <= 10 ms; <= 100 ms | 14.338 ms; 256.418 ms |
797
- | T-2 | `degree` + 400 KB readback at 100k (core resident), Node | <= 2 ms | 2.957 ms |
798
- | T-3 | Empty submit + 4-byte `readU32` round trip, Dawn | <= 0.1 ms | 1.296 ms |
799
- | T-4 | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k; at 16k | <= 1 ms; <= 2 ms | 0.972 ms; 1.953 ms |
800
- | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium (Node in brackets) | <= 6 ms | 2.600 ms (1.359 ms) |
801
- | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium (Node in brackets) | <= 12 ms | 8.100 ms (5.314 ms) |
802
- | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D; at 1M 2D; at 100k 3D | <= 10 ms; <= 100 ms; <= 20 ms | 1.790 ms; 31.057 ms; 3.954 ms |
803
- | T-7 | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M | <= 15 ms | 18.165 ms -- MISSED: 21 % over the target on this card (1.493 ms on the dev box; G4-F16 of docs/decisions/G4.md) |
804
- | T-8 | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M; at 1M / 10M | <= 150 ms; <= 1.5 s | 45.461 ms; 1092.799 ms |
805
- | T-9 | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M (100k / 1M in brackets) | <= 100 ms | 293.079 ms (28.544 ms) |
806
- | T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k; at 100k (`repulsion: "exact"`) | recorded | 0.942 ms; 54.232 ms (the spring preset 1.051 ms; 60.706 ms) |
807
- | T-10 | BFS direction-optimizing at 1M / 10M RMAT; the 1000 x 1000 grid; `sssp` at 1M / 10M (the `bfs` group) | recorded (the contract is the dev box's) | not yet recorded: the lane has not run a branch that carries the `bfs` group; the first `gpu.yml` run of it is appended with `pnpm run bench:append` (G8-F3 of `docs/decisions/G8.md`) |
814
+ <!-- targets-table:gpu-linux-t4 -->
815
+
816
+ Generated by `pnpm run bench:readme` from `benchmarks/results/targets.json`. Each figure is the best median the row has recorded, the baseline `bench:compare` pins: Node rows across the 4 session(s) of `benchmarks/results/gpu-linux-t4.json`, Chromium rows across the 2 of `benchmarks/results/nvidia-turing-driver0.json`.
817
+
818
+ | Id | What | Target | Measured | Status |
819
+ | ---- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------- | ---------------- | -------------- |
820
+ | T-1 | Upload of the 100k / 1M weighted hot prefix (16.4 MB), Node | <= 10 ms | 14.338 ms | missed (known) |
821
+ | T-1 | Upload of the 1M / 10M weighted hot prefix (164 MB), Node | <= 100 ms | 256.418 ms | missed (known) |
822
+ | T-2 | `degree` + 400 KB readback at 100k (core resident), Node | <= 2 ms | 2.292 ms | missed (known) |
823
+ | T-3 | Empty submit + 4-byte `readU32` round trip, Dawn in Node | <= 0.1 ms | 0.155 ms | missed (known) |
824
+ | T-3 | Empty submit + 4-byte `readU32` round trip, Chromium (the <= 0.1 ms target is Dawn's) | recorded | not yet measured | - |
825
+ | T-4 | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 10k | <= 1 ms | 0.972 ms | met |
826
+ | T-4 | ForceAtlas2 exact tier, GPU time per iteration (profiler) at 16k | <= 2 ms | 1.844 ms | met |
827
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Chromium | <= 6 ms | 2.600 ms | met |
828
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 10k, Node | recorded | 1.357 ms | recorded |
829
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Chromium | <= 12 ms | 6.450 ms | met |
830
+ | T-5 | ForceAtlas2 per-frame cost, `step(1)` + the 12n readback at 100k on the grid tier, Node | recorded | 5.314 ms | recorded |
831
+ | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 2D | <= 10 ms | 1.790 ms | met |
832
+ | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 1M 2D | <= 100 ms | 31.057 ms | met |
833
+ | T-6 | ForceAtlas2 grid tier, GPU time per iteration (profiler) at 100k 3D | <= 20 ms | 3.954 ms | met |
834
+ | T-7 | Attraction gather (the `fa2-attraction` pass of the grid tier), GPU time per iteration (profiler) at 1M / 10M | <= 15 ms | 18.165 ms | missed (known) |
835
+ | T-8 | PageRank, 100 iterations, wall end to end including the upload, at 100k / 1M | <= 150 ms | 45.461 ms | met |
836
+ | T-8 | PageRank, 100 iterations, wall end to end including the upload, at 1M / 10M | <= 1500 ms | 1092.799 ms | met |
837
+ | T-9 | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 1M / 10M | <= 100 ms | 290.695 ms | missed (known) |
838
+ | T-9 | Weakly connected components (Afforest), wall end to end including the upload and the label readback, at 100k / 1M | recorded | 28.475 ms | recorded |
839
+ | T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 10k | recorded | 0.942 ms | recorded |
840
+ | T-14 | Fruchterman-Reingold exact tier, GPU time per iteration (profiler) at 100k (`repulsion: "exact"`) | recorded | 52.591 ms | recorded |
841
+ | T-14 | Spring-electrical preset, GPU time per iteration (profiler) at 10k | recorded | 1.051 ms | recorded |
842
+ | T-14 | Spring-electrical preset, GPU time per iteration (profiler) at 100k (`repulsion: "exact"`) | recorded | 59.260 ms | recorded |
843
+ | T-10 | BFS direction-optimizing (`breadthFirstSearch`), wall end to end including the upload, from node 0 of the undirected RMAT at 1M / 10M (a contract on the dev box only) | <= 100 ms | not yet measured | - |
844
+ | T-10 | BFS top-down, the same traversal at 1M / 10M | recorded | not yet measured | - |
845
+ | T-10 | BFS direction-optimizing, the same traversal at 100k / 1M | recorded | not yet measured | - |
846
+ | T-10 | BFS on the 1000 x 1000 grid from a corner (1,999 levels), wall including the upload (a contract on the dev box only) | <= 1500 ms | not yet measured | - |
847
+ | T-10 | `sssp` (the near-far queue), wall end to end including the upload, random f32 weights in [0.1, 10), at 1M / 10M | recorded | not yet measured | - |
848
+ | T-10 | `sssp`, the same run at 100k / 1M | recorded | not yet measured | - |
849
+
850
+ Known misses on this class, which `bench:compare` reports without failing:
851
+
852
+ - T-1: host-side copy cost of a datacentre card behind a cloud vCPU (the README's CI-lane section); the 1M / 10M upload misses on the dev box too (docs/decisions/G1.md section 7).
853
+ - T-2: readback and submit latency of a datacentre card behind a cloud vCPU (the README's CI-lane section); met on the dev box.
854
+ - T-3: submit latency of the cloud runner; on the dev box the row misses too whenever the SM clock has dropped to idle (G3-F2 of docs/decisions/G3.md section 10).
855
+ - T-7: the pass's 16 MiB working set does not fit the T4's 4 MiB L2, and two T4 instances read 13.8 and 18.2 ms on the same work (G4-F16 and G4-F21 of docs/decisions/G4.md).
856
+ - T-9: the row carries the same 164 MB upload T-1 times, which alone takes longer than the 100 ms target; it misses on the dev box too (docs/decisions/G7.md).
857
+
858
+ <!-- /targets-table:gpu-linux-t4 -->
808
859
 
809
860
  The exact curve (the `layout-exact` group: 2D, E = 10n, seeded G(n, m), one simulation per rung; ms / iteration from the profiler):
810
861
 
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@graphty/webgpu-graph-algorithms",
3
- "version": "0.6.9",
3
+ "version": "0.6.10",
4
4
  "description": "WebGPU-accelerated graph algorithms and layouts over the @graphty/graph-format snapshot, for Node (Dawn) and browsers",
5
5
  "author": "Adam Powers <apowers@ato.ms>",
6
6
  "type": "module",
@@ -61,7 +61,7 @@
61
61
  "homepage": "https://github.com/graphty-org/graphty-monorepo/tree/master/webgpu-graph-algorithms#readme",
62
62
  "dependencies": {
63
63
  "@webgpu/types": "^0.1.72",
64
- "@graphty/graph-format": "^1.0.7"
64
+ "@graphty/graph-format": "^1.1.0"
65
65
  },
66
66
  "peerDependencies": {
67
67
  "@graphty/algorithms": "^1.0.0 || ^2.0.0",
@@ -81,20 +81,22 @@
81
81
  }
82
82
  },
83
83
  "devDependencies": {
84
- "@vitest/browser": "^3.2.4",
85
- "@vitest/coverage-v8": "^3.2.4",
86
- "@vitest/ui": "^3.2.4",
84
+ "@vitest/browser-playwright": "4.1.11",
85
+ "@vitest/coverage-v8": "4.1.11",
86
+ "@vitest/ui": "4.1.11",
87
+ "eslint": "^9.29.0",
87
88
  "fast-check": "^4.2.0",
88
89
  "ngraph.forcelayout": "^3.3.1",
89
90
  "ngraph.graph": "^20.0.1",
90
91
  "playwright": "^1.54.1",
91
92
  "tsx": "^4.20.3",
92
93
  "typescript": "^5.9.3",
94
+ "typescript-eslint": "^8.34.1",
93
95
  "vite": "^7.0.5",
94
- "vitest": "^3.2.4",
96
+ "vitest": "4.1.11",
95
97
  "webgpu": "0.4.0",
96
- "@graphty/layout": "^1.10.2",
97
- "@graphty/algorithms": "^2.0.6"
98
+ "@graphty/algorithms": "^2.1.0",
99
+ "@graphty/layout": "^1.10.3"
98
100
  },
99
101
  "scripts": {
100
102
  "build": "node -e \"require('fs').rmSync('dist',{recursive:true,force:true})\" && tsc -p tsconfig.build.json",
@@ -118,6 +120,7 @@
118
120
  "bench:ab": "node scripts/bench-ab.js",
119
121
  "bench:compare": "node scripts/bench-compare.js",
120
122
  "bench:append": "node scripts/bench-append-session.js",
123
+ "bench:readme": "node scripts/bench-readme-tables.js",
121
124
  "benchmark": "tsx benchmarks/run.ts",
122
125
  "gpu:report": "node scripts/gpu-report.js",
123
126
  "ready:commit": "npm run build:all && npm run lint && npm run test:node"