Metraipolis › archive › dc_4241f2dcfbdd

Bandwidth, Not FLOPS: FFT Kernels, Matrix Units and SAR Imaging on Apple M6 — arXiv:2609.32237

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.32237
License (as archived)
CC0-1.0
Archived
2026-10-07T22:13:03.091Z
SHA-256 of the archived text
23a2c338501f0b0eb3c33d691339e839fe5a3e8b623a31126fafcb79530af7f4
Stable URL of this page
https://metraipolis.com/documento/dc_4241f2dcfbdd

How to cite

Bandwidth, Not FLOPS: FFT Kernels, Matrix Units and SAR Imaging on Apple M6 — arXiv:2609.32237 — arXiv (en), https://arxiv.org/abs/2609.32237 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4241f2dcfbdd (Metraipolis archive dc_4241f2dcfbdd, sha256 23a2c338501f0b0eb3c33d691339e839fe5a3e8b623a31126fafcb79530af7f4)
BibTeX
@misc{dc4241f2dcfbdd,
  title = {Bandwidth, Not FLOPS: FFT Kernels, Matrix Units and SAR Imaging on Apple M6 — arXiv:2609.32237},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.32237},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4241f2dcfbdd; sha256 23a2c338501f0b0eb3c33d691339e839fe5a3e8b623a31126fafcb79530af7f4},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.32237. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

The fast Fourier transform (FFT) underlies radar, imaging and scientific computing. A classic rule for fast GPU FFTs is to compute the largest block that fits on chip and compose larger transforms from such blocks. We test this rule on Apple's M6 chip, whose GPU performs 29 arithmetic operations in the time it reads one byte from main memory, and which adds matrix units to both its GPU and CPU. Comparing our kernels with Apple's MPSGraph and vDSP libraries and MLX, we find that data movement, not arithmetic, sets FFT speed. Large batches run at or near the main-memory bandwidth limit in every GPU library, and benchmarks that keep data in cache overstate real throughput by up to $3.7\times$. The on-chip rule still predicts where speed collapses: MPSGraph and MLX lose half or more of their speed once a transform outgrows the GPU's 32\,KiB local memory. Keeping such transforms in registers avoids an extra trip through memory: $2.2\times$ faster than MPSGraph, and $4.4\times$ with half-precision storage. The GPU's matrix units do not help, because recasting the FFT as matrix products adds as much arithmetic as they save. The CPU's matrix unit, 47 times faster than its vector units, does: our kernel for it beats vDSP by up to $5.3\times$. End to end, a $4096\times4096$ synthetic aperture radar image takes 8.1\,ms on the GPU (6.0\,ms with half-precision intermediates), $14$--$19\times$ faster than a CPU reference. Running the CPU's matrix unit alongside the GPU gains nothing: both share one memory bandwidth.

Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AR).
Categorie arXiv: Performance (cs.PF) ; Hardware Architecture (cs.AR).
Pagina del lavoro: https://arxiv.org/abs/2609.32237 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T21:54:31.012Z dalla pagina https://arxiv.org/catchup/cs.AR/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).