Why do you think that Sweden has almost no start up scene?

According to this crunchbase data [1] it has a lot per capita.

[1] https://news.crunchbase.com/startups/countries-most-startup-...

Nice that this will be open sourced. Maybe the true performance characteristics can be better understood.

The press release seems exaggerated or written by marketing, e.g. claiming "accelerating optimization exponentially". Shouldn't the acceleration be in the order of the number of compute cores?

This 2016 podcast discusses this in depth: https://cppcast.com/alfred-bratterud/

I could only find a (legal) pdf[1] for $80. Pretty steep price when I'd probably only want to read 2-3 chapters.

[1]: Swedish site: https://www.bokus.com/bok/9781009418133/linear-algebra-for-d...

You could also check the world catalog to see if a library near you offers the ebook for lending. Universities typically allow the general public to walk in and look at books without registration.

https://search.worldcat.org/title/1409698868

Thanks for sharing that. AS400 always catches my eye after doing in internship at IBM, working with AS400 back in 2000.

Two months in and I'm a happy Beelink customer too. My kids mostly use it for Minecraft.

For the larger performance diffs, has anyone looked into why? Are there a couple of common reasons? I'd really like to know. Thanks

i have, yes. i can't speak for openblas or mkl, but im familiar with eigen and nalgebra's implementations to some extent

nalgebra doesn't use blocking, so decompositions are handled one column (or row) at a time. this is great for small matrices, but scales poorly for larger ones

eigen uses blocking for most decompositions, other than the eigendecomposition, but they don't have a proper threading framework. the only operation that is properly multithreaded is matrix multiplication using openmp (and the unstable tensor module using a custom thread pool)

Interesting. For other readers, this is about an early 80s Texas Instruments computer and not a graphing calculator as I first thought.

I skimmed your post and I wonder if mojo is focusing on such small 512x512 matrices? What is your thinking on generalizing your results for larger matrices?

I think for a compiler it makes sense to focus on small matrix multiplies, which are a building block of larger matrix multiplies anyways. Small matrix multiplies emphasize the compiler/code generation quality. Even vanilla python overhead might be insignificant when gluing small-ish matrix multiplies together to do a big multiply.

That would be cool.

Otherwise, sweepers will need both extra robust tires and to be ready change flats.