Article claims hand rolled C code can perform dot product faster than Numpy (opens in new tab)

(medium.com)

7 pointskuberlog5y ago5 comments

5 comments

I don’t know how the inner-workings of NumPy work...but this article seems quite plausible because it’s making use of parallel computation in order to achieve a faster result.

The title just seems to imply skepticism over what looks to be a legit approach.

kuberlogOP5y ago

Ya. The article seems plausible, but NumPy uses Blas/Lapack which can do parallel computation, so thats what is confusing about it to me.

tdeck5y ago

> Have you ever wondered how does NumPy perform its complex computations so fast? The secret is that it uses lower level programming language — C.

I thought NumPy used the LAPACK libraries at its core, which are written in Fortran:

https://en.wikipedia.org/wiki/LAPACK?wprov=sfla1

kuberlogOP5y ago

Same. Which is why this confuses me.

sova5y ago

Very neat idea. Seamless interplay between layers like that is nifty to think about.

j / k navigate · click thread line to collapse

5 comments

deckarep5y ago

I don’t know how the inner-workings of NumPy work...but this article seems quite plausible because it’s making use of parallel computation in order to achieve a faster result.

The title just seems to imply skepticism over what looks to be a legit approach.

kuberlogOP5y ago

Ya. The article seems plausible, but NumPy uses Blas/Lapack which can do parallel computation, so thats what is confusing about it to me.

tdeck5y ago

> Have you ever wondered how does NumPy perform its complex computations so fast? The secret is that it uses lower level programming language — C.

I thought NumPy used the LAPACK libraries at its core, which are written in Fortran:

https://en.wikipedia.org/wiki/LAPACK?wprov=sfla1

kuberlogOP5y ago

Same. Which is why this confuses me.

sova5y ago

Very neat idea. Seamless interplay between layers like that is nifty to think about.

j / k navigate · click thread line to collapse