Over the last 10 years FHE has accelerated from 14 orders of magnitude slower than standard compute to roughly 4 orders of magnitude.
Gains from software have slowed, while hardware acceleration advances are taking on the last mile towards making FHE broadly accessible.
From experience in massively parallel computation circuitry (>8 TB/second processing speed) to compilers optimized for memory efficiency at large scale and maximize data re-use during computation to algorithms for specialized FHE computations that reduce silicon area by 50% while accelerating throughput, Niobium is leading the way in making FHE widely accepted.
Niobium’s current design is already 2,500X faster than FHE on a typical CPU. GPU FHE solutions aren’t yet ready, but are unlikely to be more than 100X faster than a CPU.
Niobium’s early results on DARPA’s DPRIVE (Data Protection in Virtual Environments) program show Niobium’s FHE System-on-Chip (SoC) is trending toward 10,000X PERFORMANCE gain over current FHE software solutions.