Spark the EVA!
Match the frozen EVA‑02‑L reference as fast as possible. Any model or general inference approach is allowed. Pass at 9,900 exact top-1 matches, then win on end-to-end wall time.
Example wall time
228.734317s
Pass gate
99.000% 9,900 of 10,000 exact top-1Daily budget
2attempts Resets 00:00 Asia/SeoulMachine
1×DGX Spark Fixed harness, one clockExample performance over time
Fabricated personal-best wall times follow five competitors with different optimization specialties; no benchmark runs were performed.
Preparing wall-time history…
| Current rank | Competitor | Date submitted | Submission | Wall time |
|---|---|---|---|---|
| 1 | user3 | 2026-07-26 11:05:00 KST | #2003 · baseline-loader | 886.929774 s |
| 1 | user3 | 2026-07-27 11:05:00 KST | #2007 · nsys-profile | 835.204119 s |
| 1 | user3 | 2026-07-28 11:05:00 KST | #2010 · nvjpeg-device-decode | 790.510665 s |
| 1 | user3 | 2026-07-29 11:05:00 KST | #2014 · fused-normalization | 698.714320 s |
| 1 | user3 | 2026-07-30 11:05:00 KST | #2018 · custom-resize-kernel | 535.227992 s |
| 1 | user3 | 2026-07-31 11:05:00 KST | #2020 · fused-stem-kernel | 387.604553 s |
| 1 | user3 | 2026-08-01 11:05:00 KST | #2025 · persistent-cuda-pipeline | 271.882109 s |
| 1 | user3 | 2026-08-02 11:05:00 KST | #2030 · final-fused-cuda-kernels | 228.734317 s |
| 2 | user4 | 2026-07-26 11:30:00 KST | #2004 · first-container | 879.772634 s |
| 2 | user4 | 2026-07-27 11:30:00 KST | #2008 · memory-mapped-manifest | 704.116982 s |
| 2 | user4 | 2026-07-28 11:30:00 KST | #2011 · pinned-buffer-pool | 620.994511 s |
| 2 | user4 | 2026-07-30 11:30:00 KST | #2019 · async-jpeg-decode | 515.220774 s |
| 2 | user4 | 2026-07-31 11:30:00 KST | #2021 · double-buffer-prefetch | 418.662106 s |
| 2 | user4 | 2026-08-01 11:30:00 KST | #2026 · overlapped-copy-infer | 309.117449 s |
| 2 | user4 | 2026-08-02 11:30:00 KST | #2031 · final-async-io-pipeline | 252.981762 s |
| 3 | user5 | 2026-07-26 12:00:00 KST | #2005 · safe-baseline | 894.117302 s |
| 3 | user5 | 2026-07-28 12:00:00 KST | #2012 · bf16-first-pass | 820.449884 s |
| 3 | user5 | 2026-07-29 12:00:00 KST | #2015 · selective-fp16 | 603.115227 s |
| 3 | user5 | 2026-07-31 12:00:00 KST | #2022 · int8-calibrated-head | 427.770661 s |
| 3 | user5 | 2026-08-01 12:00:00 KST | #2027 · hybrid-precision-backbone | 319.822448 s |
| 3 | user5 | 2026-08-02 12:00:00 KST | #2032 · final-mixed-precision | 274.605884 s |
| 4 | user2 | 2026-07-26 10:35:00 KST | #2002 · eager-fp32 | 872.614093 s |
| 4 | user2 | 2026-07-27 10:35:00 KST | #2006 · torch-compile-first-pass | 849.880441 s |
| 4 | user2 | 2026-07-29 10:35:00 KST | #2013 · compiled-fp16-graph | 640.233990 s |
| 4 | user2 | 2026-07-30 10:35:00 KST | #2017 · static-shape-capture | 491.887264 s |
| 4 | user2 | 2026-08-01 10:35:00 KST | #2024 · cuda-graph-replay | 351.066804 s |
| 4 | user2 | 2026-08-02 10:35:00 KST | #2029 · final-compile-graphs | 296.412995 s |
| 5 | user1 | 2026-07-26 10:10:00 KST | #2001 · python-reference | 899.284951 s |
| 5 | user1 | 2026-07-28 10:10:00 KST | #2009 · batch-size-search | 762.488105 s |
| 5 | user1 | 2026-07-30 10:10:00 KST | #2016 · parallel-jpeg-loader | 618.772663 s |
| 5 | user1 | 2026-08-01 10:10:00 KST | #2023 · prefetch-queue-tune | 481.339702 s |
| 5 | user1 | 2026-08-02 10:10:00 KST | #2028 · final-loader-batching | 379.847221 s |
Example results
Fabricated standings · no benchmark runs| Rank | Entrant | Submission | Agreement | Images/s | Wall time |
|---|---|---|---|---|---|
| 1 | user3Example leader | #2030 · final-fused-cuda-kernels | 9901 / 10000 ✓ | 43.72 | 228.734317 s |
| 2 | user4 | #2031 · final-async-io-pipeline | 9912 / 10000 ✓ | 39.53 | 252.981762 s |
| 3 | user5 | #2032 · final-mixed-precision | 9906 / 10000 ✓ | 36.42 | 274.605884 s |
| 4 | user2 | #2029 · final-compile-graphs | 9920 / 10000 ✓ | 33.74 | 296.412995 s |
| 5 | user1 | #2028 · final-loader-batching | 9950 / 10000 ✓ | 26.33 | 379.847221 s |