Howling Husky
14K posts

Howling Husky
@Howling__Husky
🦸♂️ Show Host & Communication Wizard 🪄 | Community Advocate 📢 | Husky Trainer ⛏️🐺 | Spatial Computing Enthusiast 🌐
Katılım Nisan 2021
3K Takip Edilen3.5K Takipçiler
Howling Husky retweetledi

Will Fast Matrix Multiplication ever be practical?
Strassen’s 1986 discovery of fast matrix multiplication (FMM) – asserting that the product of two 𝑛×𝑛 matrices can be computed in sub-cubic time 𝑛^𝜔 ∼ 𝑛²·⁸⁷ ≪ 𝑛³ – had a profound impact on theoretical computer science and algorithm design.
Since then, mathematicians improved on Strassen’s algorithm, and some experts believe that, eventually, it will be shown that 𝜔 ≈ 2, which would mean that the time to compute AxB is essentially the time it takes to merely read the inputs: ~O(𝑛²) (!) Needless to say, such result would have a major impact on the AI compute age we’re entering…
Unfortunately, FMM algorithms only work for enormous matrices--on the order of the number of atoms in the universe (“galactic algorithms" [1])--and it is currently hard to imagine them being practical on any imaginable hardware. Besides their asymptotic runtime, a core practical issue with FMM algorithms is that they all inherently rely on recursive divide-and-conquer, which creates memory and IO-bottlenecks, and is numerically unstable; This is likely the reason why the largest hardware manufacturers in the world are not developing chips for FMM. Even Strassen’s original algorithm, which gives nontrivial FLOP speedup for relatively small matrices, struggles to beat the sheer parallelism of naiive MatMul on GPUs or TPUs.
Some interesting progress on practical FMM seems underway [2] and would be interesting to follow, but it remains to be seen whether divide-and-conquer can be implemented in both silicon and memory to deliver wall-clock speedups for realistic dimensions of matrices in LLMs.
What is means for @prlnet. That’s the reason we designed the Pearl proof-of-work protocol (cuPOW) with the underlying baseline being “naiive” matrix multiplication O(𝑛³), which is what NVIDIA, AMD, Cerebras and all other AI hardware accelerators implement today.
Nevertheless, it is important to stress that Pearl’s protocol doesn't rely on naiive MatMul remaining SoTA -- if FMM becomes practical some day, Pearl's protocol can easily adapt to the 𝑛^𝜔 baseline (since the next version of cuPoW will only verify the output AB).
In fact, one of the intriguing aspect of @prlnet is that it creates incentives (for both humans and machines) to develop faster MatMul algorithms and hardware (as had happened in Bitcoin with SHA256). Of course, without proper modification, such breakthrough would break the security assumption of Pearl-GEMM, so such algorithmic breakthrough would better be public.
FMM and FFT. Our recent paper [3] shows that it is possible to achieve fast matrix multiplication without using Strassen-like divide-and-conquer, using only the Fast Fourier Transform, which is omnipresent in countless industry-scale applications. This paper presents a simple algorithm running in 𝑂(𝑛²·⁸⁹) time, which only sums a few convolutions in 𝕫ₖᵐ, using FFT (see figure below for illustration of the algorithm).
Despite being highly parallel (no recursion), this FFT algorithm for MatMul remains asymptotic, as it still requires many parallel repetitions on submatrices in order to obtain noticeable speedup over naiive MatMul (𝑛³). Whether FFT can lead to subcubic time MatMul
for reasonably-sized matrices is a fascinating question!
I believe FFTs are the most promising tool in this direction...
[1] Lipton, Richard J., and Kenneth W. Regan. “David Johnson: Galactic Algorithms.” In People, Problems, and Proofs, 109–112. Springer, 2013. doi.org/10.1007/978-3-….
[2] Karstadt, Elaye, and Oded Schwartz. “Matrix Multiplication, a Little Faster.” Journal of the ACM 67, no. 1 (2020): 1:1–1:31. doi.org/10.1145/3364504.
[3] Uffenheimer, Yahel, and Omri Weinstein. “Improved Sparse Recovery for Approximate Matrix Multiplication.” arXiv:2602.04386, 2026..

English

Pearl is going to cause a GPU shortage
Likely already having an impact from what I can tell..
@prlnet
English

@BarrySilbert Easy $PRL @prlnet FDV hit already 1 billion and only on OTC!
English

What VVV is doing represents a small fraction of what the winner will be doing.
Private, permissionless inference is huge.
I am betting that winner will be @prlnet
And this isn’t considering the fact that inference is just 1 Ai workload of the many that Pearl supports as a protocol… so whatever you think the value of inference is, understand that is one slice of the pie Pearl will capture.
And we aren’t even discussing the fact that PRL is a superior SoV asset for the Ai era…
We’re just discussing some of the uses. But as we know from Gold & Silver, use-cases are not the reason for their monetary premium… they are just a nice reinforcement in being a consistent flow of demand.
BTC showed us what its economic incentives can do in creating the world’s largest supercomputer.
Run it back with Pearl becoming the world’s largest supercomputer for Ai workloads.
English
Howling Husky retweetledi
Howling Husky retweetledi
Howling Husky retweetledi

@CryptoSGiants I missed the war... you should say, "Enough room in the space!"
English

Ok someone had to do this. I'm doing it. Bringing some much needed clarity to this $NOCK vs. $PRL war
Crypto Twitter is split in a heated $NOCK vs. $PRL debate. Many holders argue which is better without grasping the real mechanics
This thread breaks down the shared vision, core differences, unique strengths, and massive adoption potential of each project
LFG


English
Howling Husky retweetledi

Call me a conspiracy theorist, I really don't care anymore, but the way this UFO disclosure is being rolled out is not organic, at all. When major news outlets are reporting on alien races, famous skeptics are changing their beliefs, old stories are re-appearing as if they are new across the huge network of podcasts and news organizations, and new stories are being pushed every day...
This decision was made from the deepest part of the deep state because the engine bringing it to us is their primary propoganda and disinformation network.
In other words, whatever they are about to tell us or show us, I will be, and I suggest you be, very, very cautious about believing. Something is afoot here...
We are getting AI with ubiquitous data centers, increases in global conflicts, energy and agriculture is about to take a huge hit, and new viruses all about the same time we are going to be told aliens are real.
We need to use discernment on everything right now.
English
Howling Husky retweetledi

I do not care what you say if you don’t wear that shit on your chest. This is about Nock holders who have no idea what they talking about.
Let the numbers talk. We’ll see how much traction that “well-timed” market does by count of GPUs. My guess is it won’t go that well. And then have to wait H1 2027 unless it is able to just be beta (like it has been its whole existence so far) and ride other coattails in the meantime (like trying to do now with @prlnet and did in past with @Zcash )
Character doesn’t lie and even things like a crypto have character 😁
English
Howling Husky retweetledi
Howling Husky retweetledi

Nock is essentially taking an approach of, we are going to be a network that aims to attract all of the global compute. Whether they acknowledge or not, that is essentially the path they are taking. A general compute layer.
Objectively difficult to achieve. My curiosity is what the incentives are to actually make this happen ? How do you incentivize X type of compute in the same way you incentivize Z type of compute in a 1:1 / economically sensical manner ?
How do you incentivize the capturing of all that supply (of compute) ?
Because you do in fact want a MOAT to be established and incentives are the help and driver in bootstrapping this to occur. Hence my fascination with @prlnet and using BTC economic incentives to not only boostrap security (creating superior SoV asset in world/age of Ai) but also bootstrap the network becoming a global layer for inference & training; a global layer that cannot be competed with because of mentioned economic incentives. BTC is the world’s largest supercomputer by many multiples… Apply that to Ai workloads. The rest works itself out. Supply & Demand.
*genuinely want to hear answers to questions on Nock as I like the idea (of course) but just skeptical of reality and path in getting there as I’ve been around long enough to have seen these general ideas be too obtuse in making it unable to make any meaningful strides or traction towards making said idea a reality*
English
Howling Husky retweetledi




