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.
With FPGAs, you're paying for reconfigurability, which hurts in terms of cost, speed and power.<p>FPGAs should only really be used for:<p>Pre-silicon verification<p>Designs where power/cost aren't important (a lot of high end industrial or military for example)<p>Or low quantity designs that won't make an ROI when taping out an ASIC.<p>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)<p>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.<p>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.<p>Hbm really is the magic that makes it work (economically)
> With FPGAs, you're paying for reconfigurability, which hurts in terms of cost, speed and power.<p>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.
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.
What lack of responsibility
A fun and interesting project, but to spend $10k+ in tokens on a curiosity seems excessive! Outside of my budget for sure.
They mention it's from startup credits, so entirely free for them.
Also known as a singular 200$ subscription. Nobody sane is paying "real" api prices
I did something like this with Opus 4.5: <a href="https://github.com/cpldcpu/smollm.c" rel="nofollow">https://github.com/cpldcpu/smollm.c</a><p>It seems implementing transformers is the new "hello world" of AI assisted coding.
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.
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.
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.<p>But that still not a breakthrough or anything for such a small active set of parameters that is quantized
Fpga is software-driven, model can verify its output easily.<p>Zero surprise it did it
it only cost them $10,000 in Claude API usage. :)
Repetitive "make it faster" prompting is essentially what all these "RSI" claims are
This is cool<p>Next step is tuning your post training pipeline to optimize your model for your chosen FPGA architecture...
Good chance I'm not the only one thinking this: share the code
i wonder if we can emulate an FPGA on a rtx 3090? then test it on an AWS EC2 F2 few hours a day.
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.<p>(unfortunately for a decent model it will also be price competitive with a big nvidia card)
Now do silicon. ;)