# Faster matmul

**URL:** <https://discourse.julialang.org/t/faster-matmul/111389>\
**Category:** Numerics\
**Created:** [March 9, 2024, 6:20am UTC](https://discourse.julialang.org/t/faster-matmul/111389 "2024-03-09T06:20:55Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![tomtom](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/tomtom/32/5106_2.png) [@tomtom](https://discourse.julialang.org/u/tomtom)\
**Post date:** [March 9, 2024, 6:20am UTC](https://discourse.julialang.org/t/faster-matmul/111389/1 "2024-03-09T06:20:55Z")

</div>

[https://arstechnica.com/information-technology/2024/03/matrix-multiplication-breakthrough-could-lead-to-faster-more-efficient-ai-models/](https://arstechnica.com/information-technology/2024/03/matrix-multiplication-breakthrough-could-lead-to-faster-more-efficient-ai-models/)

Would Julia implement new matmul algo like above?

---

<div class="post-metadata">

**Author:** ![jar1](https://avatars.discourse-cdn.com/v4/letter/j/c0e974/32.png) [@jar1](https://discourse.julialang.org/u/jar1)\
**Post date:** [March 9, 2024, 6:41am UTC](https://discourse.julialang.org/t/faster-matmul/111389/2 "2024-03-09T06:41:07Z")

</div>

See this thread from the last time a matmul algorithm came out

> [@Asymptotically faster matrix-multiplication algorithms](https://discourse.julialang.org/t/asymptotically-faster-matrix-multiplication-algorithms/96710):
>
> Does there even exist an actual implementation?

---

<div class="post-metadata">

**Author:** ![gdalle](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/gdalle/32/27854_2.png) [@gdalle](https://discourse.julialang.org/u/gdalle)\
**Post date:** [March 9, 2024, 9:49am UTC](https://discourse.julialang.org/t/faster-matmul/111389/3 "2024-03-09T09:49:00Z")

</div>

> [@Advances in the matrix multiplication algorithm](https://discourse.julialang.org/t/advances-in-the-matrix-multiplication-algorithm/111336/3):
>
> The laser method is not intended to be practical; it’s just a way to think about the ideal way to multiply matrices. “We never run the method [on a computer],” Zhou said. “We analyze it.”

---

<div class="post-metadata">

**Author:** ![abraemer](https://sea2.discourse-cdn.com/julialang/user_avatar/discourse.julialang.org/abraemer/32/51403_2.png) [@abraemer](https://discourse.julialang.org/u/abraemer)\
**Post date:** [March 9, 2024, 10:54am UTC](https://discourse.julialang.org/t/faster-matmul/111389/4 "2024-03-09T10:54:22Z")

</div>

Faster asymptotically does not imply faster in practice. One reason is that moderm CPUs are very complicated and perform lots of optimizations such that the number of operations is not a proxy for execution speed. Another reason is that the big-O notation hides the constant factors which are ginormous for these “fast” matrix multiplication making it a [galactic algorithm](https://en.m.wikipedia.org/wiki/Galactic_algorithm#Matrix_multiplication).