Vibecoding an inference chip

102 points by ankitvgupta 6 hours ago on hackernews | 35 comments

datadrivenangel | 5 hours ago

Does not say what type of model or size they were serving, but one of the traces mentions Maple, which is a 20b model. Still, getting ~70+ tok/s is quite good.

datadrivenangel | 5 hours ago

it only cost them $10,000 in Claude API usage. :)

hmate9 | 5 hours ago

You get that much usage in a monthly $200 subscription.
And a kitten.
So who's paying the cost?

exe34 | 4 hours ago

Corporate customers

ralph84 | 3 hours ago

API pricing vs. individual plan pricing is just price discrimination. They're both positive gross margin.

surgical_fire | 3 hours ago

People believe in some wild stuff.

drob518 | an hour ago

If that’s true then the api is waaaaaay overpriced.

stronglikedan | 3 hours ago

Everyone else that doesn't use the entirety of the plan they pay for.

nyc_data_geek | 3 hours ago

Retail and society as ANT shares get pulled into index and wealth funds

hoppp | 2 hours ago

The IPO they just raised can subsidise it for a while.

risyachka | 4 hours ago

Fpga is software-driven, model can verify its output easily.

Zero surprise it did it

ilaksh | 3 hours ago

Not really. It's only 1B active, and he was probably using the 5 gb quant. Also he said it was over 200 tokens per second.

But that still not a breakthrough or anything for such a small active set of parameters that is quantized

codekansas | 4 hours ago

This is cool

Next step is tuning your post training pipeline to optimize your model for your chosen FPGA architecture...

chalkycrimp | 4 hours ago

Good chance I'm not the only one thinking this: share the code

cpldcpu | 4 hours ago

I did something like this with Opus 4.5: https://github.com/cpldcpu/smollm.c

It seems implementing transformers is the new "hello world" of AI assisted coding.

saejox | 4 hours ago

i wonder if we can emulate an FPGA on a rtx 3090? then test it on an AWS EC2 F2 few hours a day.

Archit3ch | 3 hours ago

Now do silicon. ;)

IshKebab | 3 hours ago

marvinborner | 3 hours ago

They seem to target general-purpose non-homogeneous parallelism, not inference, right?

spwa4 | 3 hours ago

I must say I do see one application for this, that you can't really match reasonably otherwise: you can bake weights into an FPGA and then run them for 5W at a speed that will be competitive with a big Nvidia card.

(unfortunately for a decent model it will also be price competitive with a big nvidia card)

Repetitive "make it faster" prompting is essentially what all these "RSI" claims are

hoppp | 2 hours ago

This is exciting, I have been wishing for good ternary models for a while as it seems more effective and combining it with FPGA I think this is a super promising path forward.

jluxenberg | 2 hours ago

A fun and interesting project, but to spend $10k+ in tokens on a curiosity seems excessive! Outside of my budget for sure.

ricardobeat | 2 hours ago

They mention it's from startup credits, so entirely free for them.

mixtureoftakes | 2 hours ago

Also known as a singular 200$ subscription. Nobody sane is paying "real" api prices

itissid | 2 hours ago

Someone mentioned that FPGAs though amazing to create bespoke chips that are fast, run into a fab capacity issue. Someone explained briefly to me that even if you could make an fpga that could do inference better than current GPUs it was economically possible to make them at scale. I can't recall whether that was for technology related issues or just because fab capacity is completely filled up.

ShadowBlades512 | an hour ago

One issue is that FPGAs are very fine grained and they need to gaurentee an image the tool confirms works for a chip must work across all produced chips. There is a level of binning for speed and binning for some disabled portions of the chip but it is a lot less flexible then disabling shader cores, memory controllers, and portions of the cache on a GPU.

mikeayles | an hour ago

With FPGAs, you're paying for reconfigurability, which hurts in terms of cost, speed and power.

FPGAs should only really be used for:

Pre-silicon verification

Designs where power/cost aren't important (a lot of high end industrial or military for example)

Or low quantity designs that won't make an ROI when taping out an ASIC.

That said, I'm currently designing one into a product because at the moment, the cost of an ASIC is too much of a gamble (and also makes it too much of a black box for functional safety certification)

I also got a couple of different LLM's running on a commodity FPGA, and hit 59,000tok/s with t=1, and 23,000tok/s with t=128. I say LLM, but the model was 3M params, so pretty dumb! But, to hit that speed, I couldn't use the DDR4 as it's bandwidth was the bottleneck.

I currently have an 8b model running with the DDR in the loop, but it's stuck at 14ish tok/s, but the model is usable.

Hbm really is the magic that makes it work (economically)

femto | an hour ago

> With FPGAs, you're paying for reconfigurability, which hurts in terms of cost, speed and power.

In that vein, what happened to HP's Field Programmable Nanowire Interconnect (FPNI)? At the time, the claim was that it had the potential to outperform ASICs in cost, speed and power.

numpad0 | 22 minutes ago

Isn't it like FPGAs use more transistors per equivalent transistor it simulates?

And therefore only useful when the real CPU/GPU/ASICs don't exist or are highly inefficient for some reason, like you have to have a super wide bus signal to apply a known set of parameters in batch or something?

What lack of responsibility

seanjhkwon | 29 minutes ago

It shows the importance of automated indicators when giving AI huge tasks. Maybe more important than the intelligence of the model.

reuben364 | 11 minutes ago

Over the week or so I have vibe coded a Lean-like language with

- quantitative types

- a Rust-ish DSL that lowers into the base language

- a C, LLVM, WebAssembly backend and a Verilog backend for a subset of the language

- A RISC-V implementation written in the Verilog subset that runs Doom on my Arty FPGA.

- A category theory and real number library

- A graph storage engine with queries written using profunctor optics

- A datalog with an egglog in progress

- A Joern-like code analysis engine for C and C# using tree-sitter to parse

- A React style framework

- Editor plugins for my language

All at varying stages of completeness.

What you can do now is wild. My fear is a sort of Gell-Mann Amnesia effect where my lack of expertise in the area is causing me to miss major problems with it. But at least some part of it has to be real. I plan to announce it in a more official capacity once I've allayed those concerns (https://wiz.badlydrawnwizards.com for a sneak peak, but sorely need to update the website)

My bottleneck currently is CI capacity. I've turned every spare computer in my household into a Slurm node so I can run more testing locally to avoid wasting precious CI capacity.