1) Spent quite a bit of time with Habana and was *super* impressed. Taped out two fantastic chips while raising dramatically less money than most competitors. Exceptional team, absolute execution machine. Only startup in recent MLPerf benchmarks.
2) Also a great lesson in the merit of starting on the n-1 node, which forces a startup to actually develop a true architectural advantage and conserves money vs. spending vast amounts for a risky, bleeding edge tapeout that doesn’t force architectural innovation.
3) IMO, acquisition all about RoCE which enables performance to scale linearly across many, many processors.
Also raises several questions. Will these chips ever be fabbed internally? Wherefore now for Xe GPU? And will ONEAPI ever work? Regardless, good move.
4) Will be interesting to see how Habana fares over time vs. more exotic approaches. Stands to reason that all/most startups were evaluated. Would love to know if anyone is using the Graphcore instances in Azure.
5) And am also super excited abt @MythicInc. As AI “eats the world” there will be a huge need for low power edge inference, where there has been less investment vs. the gladiator combat for the datacenter crown. Doing analog compute in memory brilliant from 1st principles.
6) Outside of Habana and @mythicinc, most curious to see where @CerebrasSystems goes from here. Have to admire the vision and tech. "Go big or go home."
Sidenote: I heard "Go big or go home" a lot when I worked as a housekeeper in the ski industry at Alta. Excited for ❄️⛷️
You can follow @GavinSBaker.
Tip: mention @threader_app on a Twitter thread with the keyword “compile” to get a link to it.
Enjoy Threader? Sign up.
Threader is an independent project created by only two developers. The site gets 500,000+ visits a month and our iOS Twitter client was featured as an App of the Day by Apple. Running this space is expensive and time consuming. If you find Threader useful, please consider supporting us to make it a sustainable project.