SIML Compute
DemonstratedFit a scientific model into available memory and measure the cost of advancing it.
- Current work
- The lab has compared CUDA implementations of cellular signaling and binary lattice updates against tuned baselines. The two studies count different kinds of work.
- Connection
- A model supplies its state, update rules and correctness checks. The compute experiment varies the representation or execution while preserving that workload.
- Limits
- The reported gains apply to specified populations on one GPU. Transfers, setup, observation frequency and scheduling can change application performance.
- Next test
- Repeat a matched comparison on a partner workload and another GPU with a declared accuracy tolerance.