Loading
Research · Computing
Many of the systems we build are limited by computation: real-time rendering, model inference, quantitative workloads, large-scale data. This direction studies how to use modern hardware well — distributing work, exploiting parallelism and keeping complex computation predictable.
Active direction
Where does moving work to the GPU — in the browser or on the server — change what an interface can do?
How should real-time systems degrade under load without misleading their users?
Which simulation workloads benefit from distribution, and which are dominated by coordination cost?
No findings have been published in this direction yet.