zachbee

9 ms·

31.
by zachbee 3y ago
In practice, though, asynchronous systems can run a lot faster than their synchronous counterparts! Also, I'm mostly working on this project because it's cool and interesting to build oscillator-based computing systems on FPGAs, n
32.
by zachbee 3y ago
The Fujitsu device is a synchronous digital architecture. I'm using asynchronous oscillators, more like Lo et. al. ( https://www.nature.com/articles/s41593-024-01607-5 ) I personally think it's super cool that
33.
by zachbee 3y ago
This is a personal project aiming to replicate analog coupling dynamics with asynchronous digital circuits FPGAs, not a claim that we can solve NP complete problems with a fundamental speedup. Also, these sorts of Ising architectures get th
34.
by zachbee 3y ago
I've actually been working on a personal project using asynchronous ring oscillators on FPGAs to perform meaningful computation. If you make the oscillators interact in the right way, you can leverage them to solve certain graph proble
35.
Why hasn't there been a big winner in edge AI chips?(zach.be)
2 points by zachbee 3y ago|0 comments
36.
by zachbee 3y ago
They're not wrong that sampling a complex, higher-dimensional probability distribution is hard to do efficiently. I'm not sure how useful it is to do it more efficiently, though. Also, the fact that they're using ultra-cold s
37.
by zachbee 3y ago
The reason HBM "isn't deterministic" is because the memory controller sometimes stalls reads when doing a refresh operation. It's possible to pause refreshes [1] or schedule refreshes ahead of time to get HBM and DRAM to
38.
Why is everybody talking about Groq?(zach.be)
3 points by zachbee 3y ago|6 comments
39.
by zachbee 3y ago
The article says they ran GPT-2! Which isn't particularly large, but replicating a large language model with a spiking neural network seems like novel work at least.
40.
by zachbee 3y ago
Neuromorphic computing is cool, but not new tech. However, using a neuromorphic spiking architecture to run LLMs seems new. Unfortunately, there doesn't seem to be a paper associated with this work, so there's no deeper informatio
41.
by zachbee 3y ago
Neuromorphic computing basically uses individual "neurons", represented with either analog or digital circuits, which communicate using asynchronous pulses called "spikes". Unlike the human brain, neuromorphic chips are
42.
by zachbee 3y ago
I'm a professional chip designer. And as somebody who's had to build this sort of internal tooling myself, I think this sort of product is desperately needed! It makes sense to start with Verilator, because it's fast, easy, a
43.
by zachbee 3y ago
They get impressive performance, but it's not really the first "LLM-native processor". They built the thing a while ago before LLMs were that hot, and rebranded it as an LPU to pivot into Generative AI: https://www
44.
by zachbee 3y ago
You need an entire rack or two worth of Groq systems to get this performance, because the only memory the system has is on-chip SRAM (220MB). If you're trying to run on a small number of systems, you need local DRAM or HBM, which Groq
45.
How to Fork Algorand's Pure Proof-of-Stake Blockchain(zachbe.substack.com)
1 points by zachbee 3y ago|0 comments
46.
by zachbee 3y ago
What's the difference between your product and Gridspace's? I get the sense that your offering is more developer-focused, but I'm curious if there are any technical differences.