Reducing Running Time and Time to First X: A Walkthrough

07/28/2022, 1:20 PM — 1:30 PM UTC
Purple

Abstract:

Optimizing Julia isn't hard if you compare it to Python or R where you have to be an expert in Python or R and C/C++. I'll describe what type stability is and why it is important for performance. I'll discuss it in the context of performance (raw throughput) and in the context of time to first X (TTFX). Julia is sort of notorious for having really bad TTFX in certain cases. This talk explains the workflow that you can use to reduce running time and TTFX.

Platinum sponsors

Julia ComputingRelational AIJulius Technology

Gold sponsors

IntelAWS

Silver sponsors

Invenia LabsBeacon BiosignalsMetalenzASMLG-ResearchConningPumas AIQuEra Computing Inc.Jeffrey Sarnoff

Media partners

Packt PublicationGather TownVercel

Community partners

Data UmbrellaWiMLDS

Fiscal Sponsor

NumFOCUS