Skip to content

bench: add M2 Max 38-core GPU roofline #297

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
sbryngelson merged 2 commits into from
Oct 7, 2026
Merged

bench: add M2 Max 38-core GPU roofline #297

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d119f85
bench: add M2 Max 38-core GPU roofline
Jeff271828 Oct 6, 2026
f64e790
bench: regenerate rooflines.json for the M2 Max submission
sbryngelson Oct 7, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 5 additions & 2 deletions bench/results/ROOFLINES.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,11 +1,11 @@
<!-- generated by bench/aggregate_rooflines.py; do not edit by hand -->
# ANE roofline results

Accumulated across **10 machine(s)**, 14 submission(s). Each row is a distinct machine (grouped by hardware hash; identical silicon in different chassis stays separate by model identifier).
Accumulated across **11 machine(s)**, 15 submission(s). Each row is a distinct machine (grouped by hardware hash; identical silicon in different chassis stays separate by model identifier).

Add yours: `PYTHONPATH=. python3 bench/roofline_suite.py --contributor <your-gh-handle>`, commit the JSON it writes under `bench/results/rooflines/`, regenerate this file with `python3 bench/aggregate_rooflines.py`, and open a PR. See [`rooflines/README.md`](rooflines/README.md).

Thanks to [@751K](https://github.com/751K), [@Rayan-and-beyond](https://github.com/Rayan-and-beyond), [@axiom-of-choice](https://github.com/axiom-of-choice), [@danlee2002](https://github.com/danlee2002), [@diegobauavi](https://github.com/diegobauavi), [@hyfjjjj](https://github.com/hyfjjjj), [@n0madic](https://github.com/n0madic), [@sbryngelson](https://github.com/sbryngelson) for contributing datapoints.
Thanks to [@751K](https://github.com/751K), [@Jeff271828](https://github.com/Jeff271828), [@Rayan-and-beyond](https://github.com/Rayan-and-beyond), [@axiom-of-choice](https://github.com/axiom-of-choice), [@danlee2002](https://github.com/danlee2002), [@diegobauavi](https://github.com/diegobauavi), [@hyfjjjj](https://github.com/hyfjjjj), [@n0madic](https://github.com/n0madic), [@sbryngelson](https://github.com/sbryngelson) for contributing datapoints.

## Machines

Expand All @@ -14,6 +14,7 @@ Thanks to [@751K](https://github.com/751K), [@Rayan-and-beyond](https://github.c
| Apple M1 | MacBookAir10,1 | 4+4 | 7 | 8 GB unified | 15.6.1 (24G90) | ac | caeef8edf13b | [@Rayan-and-beyond](https://github.com/Rayan-and-beyond) | 1 |
| Apple M1 | MacBookPro17,1 | 4+4 | 8 | 16 GB unified | 26.6.2 (25G83) | ac | 263e02ff1898 (+101) | [@diegobauavi](https://github.com/diegobauavi) | 2 |
| Apple M1 Max | MacBookPro18,2 | 8+2 | 32 | 32 GB unified | 26.5.1 (25F80) | ac (high-power) | 1f8b477a5397 | [@sbryngelson](https://github.com/sbryngelson) | 2 |
| Apple M2 Max | Mac14,6 | 8+4 | 38 | 64 GB unified | 26.7.1 (25G241) | ac (high-power) | 4114158bb7cf | [@Jeff271828](https://github.com/Jeff271828) | 1 |
| Apple M2 Pro | Mac14,12 | 6+4 | 16 | 32 GB unified | 26.5.2 (25F84) | ac | 64108472d01f (dirty) | [@axiom-of-choice](https://github.com/axiom-of-choice) | 2 |
| Apple M4 | Mac16,12 | 4+6 | 8 | 16 GB unified | 26.2 (25C56) | ac | 23ef10a6c370 | [@sbryngelson](https://github.com/sbryngelson) | 1 |
| Apple M4 Max | Mac16,6 | 10+4 | 32 | 36 GB unified | 26.7 (25G229) | ac | 654b1d772321 (dirty) | [@n0madic](https://github.com/n0madic) | 1 |
Expand All @@ -31,6 +32,7 @@ Where fp16 math on the engine silently goes wrong (issue #115, docs/cross-chip.m
| Apple M1 | MacBookAir10,1 | ~32759 | clamp @ 4094 | <= 2048 |
| Apple M1 | MacBookPro17,1 | ~32759 | clamp @ 4094 | <= 2048 |
| Apple M1 Max | MacBookPro18,2 | ~32759 | clamp @ 4094 | <= 2048 |
| Apple M2 Max | Mac14,6 | ~32759 | clamp @ 4094 | <= 2048 |
| Apple M2 Pro | Mac14,12 | ~32759 | clamp @ 4094 | <= 2048 |
| Apple M4 | Mac16,12 | ~32759 | exact (no clamp) | <= 2048 |
| Apple M4 Max | Mac16,6 | ~32759 | exact (no clamp) | <= 2048 |
Expand All @@ -52,6 +54,7 @@ Peak ANE numbers only; the full per-size sweeps, all-engine comparison, and per-
| Apple M1 | MacBookAir10,1 | 4.3 | 5.4 | 802 | 622 | 278 | ac |
| Apple M1 | MacBookPro17,1 | 2.7 | 0.85 | 3171 | 484 | 276 | ac |
| Apple M1 Max | MacBookPro18,2 | 4.6 | 7.4 | 616 | 897 | 244 | ac (high-power) |
| Apple M2 Max | Mac14,6 | 5.4 | 8.4 | 640 | 844 | 277 | ac (high-power) |
| Apple M2 Pro | Mac14,12 | 3.4 | 0.98 | 3429 | 744 | 276 | ac |
| Apple M4 | Mac16,12 | 5.4 | 7.7 | 708 | 1112 | - | ac |
| Apple M4 Max | Mac16,6 | 8.9 | 23 | 385 | 1103 | - | ac |
Expand Down
24 changes: 24 additions & 0 deletions bench/results/rooflines.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,6 +68,30 @@
"reduce_exact_sum": "<= 2048",
"timestamp_utc": "2026-08-05T20:14:25Z"
},
{
"chip": "Apple M2 Max",
"model_identifier": "Mac14,6",
"hardware_hash": "8c6229361be7",
"p_cores": 8,
"e_cores": 4,
"gpu_cores": 38,
"ram_gb": 64,
"macos_version": "26.7.1",
"macos_build": "25G241",
"aneforge_version": "0.1.dev362",
"power": "ac (high-power)",
"contributor": "Jeff271828",
"peak_fp16_gemm_tflops": 5.38,
"bandwidth_gbps": 8.41,
"ridge_flop_per_byte": 640.0,
"peak_perf_per_w_gflops": 844.0,
"decode_tok_s": 277.0,
"matmul_inf_cliff": "~32759",
"slice_x16_cliff": "clamp @ 4094",
"reduce_exact_sum": "<= 2048",
"timestamp_utc": "2026-10-06T05:29:29Z",
"peak_perf_per_w_source": "powermetrics"
},
{
"chip": "Apple M2 Pro",
"model_identifier": "Mac14,12",
Expand Down
Loading
Loading