Nvidia to Share New Details on Grace CPU, Hopper GPU, NVLink Switch, Jetson Orin

Seems like this will end up driving more research into very large ML models. Some of the performance benchmarks here are for trillion parameter models. I’ve observed that often Innovation directly follows silicon improvements.

&gt; Grace CPU uses LPDDR5X<p>Looks like we’re finally going with soldered RAM in the data center?

I want one. Damn hotchips always has stuff I want and then can never afford even when it gets to market

Are the vertical-specific Nvidia SDKs like DRIVE and Clara widely used?

grace and hopper? from stranger things?