Skymel’s core product, NeuroSplit, dynamically 'splits' individual model inferencing so an initial phase runs on end-user devices while the remainder completes on cloud GPUs. The technology continuously assesses idle device compute and balances load to avoid degrading other applications, supporting large and multiple models per application. Skymel says NeuroSplit can reduce compute costs by up to 60% and gives an example where an application that typically needs multiple Nvidia A100s at $2.74/hour can run on a single A100 or multiple V100s at $0.83/hour using NeuroSplit. Feature highlights include an inference pipeline cache and inference rotation to avoid redundant model downloads across apps. The company was founded by Neetu Pathak (CEO) and Sushant Tripathy (CTO) and is based in Sunnyvale, Calif. NeuroSplit is available in preview immediately and will enter private beta in Q3 2024.

Total raised

$1M

Funding rounds

1

Latest round

Seed

Latest activity

May 2024