Skip to content

Add roofline data for Apple M1 (MacBook Pro 17,1) #148

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

Closed
diegobauavi wants to merge 2 commits into from
Closed

Add roofline data for Apple M1 (MacBook Pro 17,1) #148

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
f2ae2b3
Add roofline data for Apple M1 (MacBook Pro 17,1)
diegobauavi Aug 1, 2026
8e8d41c
Regenerate ROOFLINES.md with Apple M1 (MacBookPro17,1) data
diegobauavi Aug 1, 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
@@ -1,16 +1,17 @@
<!-- generated by bench/aggregate_rooflines.py; do not edit by hand -->
# ANE roofline results

Accumulated across **1 machine(s)**, 1 submission(s). Each row is a distinct machine (grouped by hardware hash; identical silicon in different chassis stays separate by model identifier).
Accumulated across **2 machine(s)**, 2 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 [@sbryngelson](https://github.com/sbryngelson) for contributing datapoints.
Thanks to [@diegobauavi](https://github.com/diegobauavi), [@sbryngelson](https://github.com/sbryngelson) for contributing datapoints.

## Machines

| Chip | Model | CPU (P+E) | GPU | Memory | macOS | Power | Code (main merge-base) | By | Runs |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Apple M1 | MacBookPro17,1 | 4+4 | 8 | 16 GB unified | 26.5.2 (25F84) | ac | 263e02ff1898 | [@diegobauavi](https://github.com/diegobauavi) | 1 |
| Apple M5 Pro | Mac17,8 | 6+12 | 20 | 48 GB unified | 26.5.1 (25F80) | ac (high-power) | 55a1c6afdad3 (dirty) | [@sbryngelson](https://github.com/sbryngelson) | 1 |

## Numeric cliffs (correctness rooflines)
Expand All @@ -19,6 +20,7 @@ Where fp16 math on the engine silently goes wrong (issue #115, docs/cross-chip.m

| Chip | Model | matmul inf-cliff | slice-x16 (Q.4 crop-DMA) | reduce exact-sum |
| --- | --- | --- | --- | --- |
| Apple M1 | MacBookPro17,1 | ~32759 | clamp @ 4094 | <= 2048 |
| Apple M5 Pro | Mac17,8 | ~32759 | exact (no clamp) | <= 2048 |

_matmul: output magnitude that flips finite->inf (~fp16_max/2). slice: the pre-A16 crop-DMA clamps |value|>4094; 'exact' means the clamp is gone (A16/M5). reduce: largest all-exact integer row-sum (fp16 grid coarsens past 2048)._
Expand All @@ -29,5 +31,6 @@ Peak ANE numbers only; the full per-size sweeps, all-engine comparison, and per-

| Chip | Model | Peak fp16 GEMM (TF/s) | Bandwidth (GB/s) | Ridge (FLOP/byte) | Peak perf/W (GF/s/W) | Decode @b1 (tok/s) | Power |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Apple M1 | MacBookPro17,1 | - | - | - | 442 | - | ac |
| Apple M5 Pro | Mac17,8 | 10.0 | 24 | 418 | 918 | 117 | ac (high-power) |

Loading