algebraic-nnhw
Hardware
accelerator architecture
SHOW HN2024
How HN took it
6 positive, 8 neutral or asking, 0 mixed, 8 critical, out of 22 comments about the project. Thread score -0.08.
Idea: 4 for, 1 againstPerformance: 0 for, 5 againstDocs: 2 for, 1 againstDesign: 0 for, 1 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Clever trick for hardware matmul with fewer multipliers
From 2 comments
"Conventional fixed-point MM" is a large suite of algorithms. It is correct that this is a 2x reduction in MULs compared to naive fixed-point matrix multiply, but there is a…
pclmulqdq on HNThe readme doesn't explain much, but the introduction to the paper itself is quite approachable. As for whether it's groundbreaking or not ... it's a neat readily-accessible…
stephencanon on HN
Objections
Floating point accuracy and numerical stability are unclear or poor
From 4 comments
On the other hand, if you tried it with floating point, you'd lose significant digits. Since the approach is to sum (a[i] + b[i+1])(a[i+1] + b[i]) and subtract the sums of…
yorwba on HNThere are a lot of matrix multiplication algorithms out there with a lot of pluses and minuses. It's always a balance of accuracy, runtime, and scaling. This one probably has bad…
pclmulqdq on HNFor floating point? Are you sure?
p1esk on HNThe paper cited is about hardware, where there is no accuracy tradeoff because you control the numerical precision completely and use fixed point. In a software implementation,…
pclmulqdq on HN
Hardware benefit requires committing to custom hardware matching dimensions
From 1 comment
It's not just a software algorithm. It's a hardware architecture optimization. To benefit, you have to build hardware that matches the dimensions of the algorithm. That's an…
lupire on HN
README is unclear about the improvement
From 1 comment
This readme does a really poor job of explaining what the improvement is or how they drop half the multiplications. What is the Big O run time on this? Is this shifting the known…
barfbagginus on HN
Requests
Comparison against OpenBLAS and cuBLAS performance
From 1 comment
Wonder how well it compares to openblas and cublas.
1letterunixname on HN
Questions
Why is this not already used in accelerators?
From 1 comment
This looks pretty cool! What's the catch? e.g. why isn't this already implemented in accelerators, is it really just a forgotten algorithm, or this has some implications on the…
halflings on HN
Does it work for floating point?
From 2 comments
For floating point? Are you sure?
p1esk on HNI just looked at the paper: the answer is no, floating point is not supported.
p1esk on HN
Alternatives named
Strassen and other sub-cubic matrix multiplication algorithms
From 1 comment
I don't know why this answer is getting downvoted. This is absolutely correct. W. Miller has a paper discussing, under conditions of numerical stability, O(n^3) multiplications…
abetusk on HN
Tropical and log semiring matrix methods
From 1 comment
This is very cool and a real interesting read! For those in the comments confused about how this is better, the paper is talking about synthesizing matrix multiplication…
michelpp on HN