You're right but my understanding is that Groq's LPU architecture makes it inference-only in practice.
Like Groq's chips only have 230MB of SRAM per chip vs 80GB on an H100, training is memory hungry as you need to hold model weights + gradients + optimizer states + intermediate activations.
Re:Nvidia's $20B antitrust loophole
by refibrillator on Saturday December 27, 2025 @03:31PM
(#46404975)
H100 has 80 GB of HBM3. There’s only like 37 MB of SRAM on a single chip.
Re:Nvidia's $20B antitrust loophole
by bionhoward on Saturday December 27, 2025 @02:53PM
(#46404663)
Would SRAM make weight updates prohibitive vs DRAM?
Also tenstorrent could be a viable challenger in this space. It seems to me that their NoC and their chips could be mostly deterministic as long as you don't start adding in branches