Shruthi Gorantala

Lots of good work on encrypted compute lately, it feels like we are at an inflection point! But can we please stop using misleading metrics like “amortized latency”? There are only 3 numbers that matter in benchmarks: - latency: how long it takes for a single operation to complete - throughput: how many operations per second you can do - cost: how much you pay to achieve these latency and throughput numbers. For FHE, we have always been transparent that we optimize for low latency, and scale throughput horizontally by adding more compute. This is particularly important for blockchain where you typically need to do a lot of operations sequentially and thus are latency bound.