4 ms·
I was thinking more along the lines of VR experiences that are sufficiently advanced and detailed as to necessitate multiple large hybrid ML models and asset ba
by meltyness 4y ago
I was thinking more along the lines of VR experiences that are sufficiently advanced and detailed as to necessitate multiple large hybrid ML models and asset banks, consider what a Dall-E might look like in such a use case in the next 10-15 years.
It's still the Pong days for these things, and no one knows if their realistic usage will involve on-device compute resources for such a deployment.
- orev 4y agoIt’s pretty obvious that this will all be cloud based. I just don’t see people wanting to have a full rack of hardware in their houses, when many are already opting out of even small things like laptops in favor of mobile and iPads. The business models of any company providing an AI service definitely would not even consider on-prem when recurring subscription fees are so well accepted by consumers now.
- meltyness 4y agoI disagree, large broadly-based language-involved models might need that, but a performance-adaptive scope-specific application like texturing might just need part of a single graphics card's processing time, or an SD-card sized cache with narrow-register parallel processing on-board.