Logix on the ESICUP nesting benchmark
Ten standard nesting tests, 60 seconds each. Logix, the PolygonLogix nesting engine, averages 86.9% density across them. Watch it run, then see the score on every test.
The overall score
Mean density across the ten tests. All three engines had the same 60 seconds per test, on the same machine. sparrow is the open-source research solver from KU Leuven, built from its public code and run by us. Turbo is the engine PolygonLogix shipped before Logix.
-
LogixThe PolygonLogix nesting engine.86.9%
-
sparrowOpen-source research solver, version 0.3.0. Run by us.86.7%
-
TurboOur previous engine. Nine of the ten tests.73.4%
Turbo’s mean covers nine tests, because its shirts layout failed the check. Given 20 minutes per test instead of 60 seconds, sparrow’s authors report a mean of 88.2%.
Test by test
Each row is one test and every mark is a 60-second result. The further right a mark sits, the denser the nest. Logix is the green dot, sparrow the diamond and Turbo the square.
- Logix
- sparrow
- Turbo
Density in percent at 60 seconds per test. Logix is the median of five runs and sparrow the median of three. Turbo returns the same layout on every run.
| Test | Parts | Logixmedian of 5 runs | sparrowmedian of 3 runs | Turbosame every run |
|---|---|---|---|---|
| albano | 24 | 88.16 | 87.49 | 72.31 |
| dagli | 30 | 88.07 | 87.02 | 76.38 |
| fu | 12 | 91.92 | 91.47 | 77.02 |
| jakobs1 | 25 | 89.07 | 89.07 | 75.37 |
| jakobs2 | 25 | 82.11 | 83.90 | 64.31 |
| mao | 20 | 84.99 | 84.70 | 71.55 |
| marques | 24 | 89.74 | 89.31 | 78.53 |
| shirts | 99 | 87.46 | 87.40 | – |
| swim | 48 | 76.42 | 75.56 | 62.52 |
| trousers | 64 | 90.95 | 90.95 | 82.31 |
| Mean | 86.89 | 86.69 | 73.37 |
Logix’s scores run from 76.42% on swim to 91.92% on fu. Seven of the ten finish above 87%. Against sparrow at the same 60 seconds, Logix is ahead on six tests, level within a tenth of a point on three (jakobs1, shirts and trousers) and behind on jakobs2 by 1.79 points. It is ahead of Turbo on every test Turbo completed, by 9 to 18 points.
What the clock buys
The same ten tests at three time limits. Most of the density arrives in the first ten seconds, which is the kind of time a quote can afford. A production nest can wait the full minute.
- 10 seconds
- 85.9%
- 30 seconds
- 86.8%
- 60 seconds
- 86.9%
Mean density across the ten tests at each limit.
How the scores were taken
- The tests. The ten standard irregular nesting instances kept by ESICUP, the EURO Special Interest Group on Cutting and Packing: albano, dagli, fu, jakobs1, jakobs2, mao, marques, shirts, swim and trousers.
- The rules. Each test’s own set of allowed rotations, no mirroring, and no spacing between parts.
- The score. Density is total part area divided by the strip width times the length of strip used.
- Logix. 60 seconds of wall-clock time per test on 8 threads. Five runs per test, each from a different random seed, with the median reported.
- sparrow. Version 0.3.0, built from its public code and run by us: 60 seconds per test on 16 threads, three runs per test with the median reported, scored by the same check. In one of its thirty runs, on dagli, two parts overlapped by 0.001 square units on a strip 1,000 units wide. We counted that run at the density sparrow itself reported for it.
- Turbo. Version 1.4.5, under the same rules and the same check. It stops on its own after a few seconds and returns the same layout every run. Its shirts layout failed the check, so that test has no score and its mean covers the other nine.
- The machine. One 16-core workstation (AMD Threadripper PRO 5955WX) for all three. Logix ran four tests at a time. sparrow ran one at a time, and seven of its thirty runs overlapped with an unrelated 4-thread job.
- The check. Every layout is scored by a validator that is separate from the engine and works from the true part outlines. A layout with any overlap, or any part outside the strip, gets no score.
- The date. Recorded 8 October 2026.
What density means on your floor
Density is the share of the material used that ends up as parts. These ten tests are strip problems: a fixed width, as much length as the engine needs, no sheet edges and no gap between parts. Several of them come from garment cutting, not sheet metal.
So a score here will not match the utilization you see on a 60 × 120 sheet with half an inch of spacing. What the benchmark gives is the one comparison where every engine is handed identical parts and identical rules.
The number that matters in a shop is sheets used on your own jobs. That is what a demo is for.
Sources
- The tests. ESICUP datasets.
- sparrow. Code at github.com/JeroenGar/sparrow. The 20-minute figure is from Gardeyn, Vanden Berghe and Wauters, “An open-source heuristic to reboot 2D nesting research”, arXiv:2509.13329: expected density over 100 runs of 20 minutes each on its authors’ hardware.
- If you see an error in a number on this page, write to [email protected].
Count the sheets on your own job
Send us your parts and we’ll nest them with Logix, live, on your sheet sizes and your spacing. At FABTECH 2026 in Las Vegas, October 21–23? It’s running at booth N9813.