Gadi Singer (Intel) interviewed at the O'Reilly Artificial Intelligence Conference NY 2019
Interviewat O'Reilly Artificial Intelligence Conference 2019 - New York, New York · 2019-04-15..04-18·
summary
Intel describes VNNI augmenting AVX vector instructions to give some deep‑learning ops up to 14x boost on CPUs, targeting inference-heavy loads where current deployment sees about 4:1 inference:training moving toward ~10:1. Cascade Lake Xeon adds low‑precision and latency/large‑memory support so recommendation systems and mixed tensor + sequential pipelines run on Cascade Lake Xeon without moving data off CPU. For heavier tensor math Intel offers dedicated NPUs: NNP L 1000 Spring Crest for training and NNP I 1000 Spring Hill for inference.
AI-written summary · about
chapters (4)
- 0:00Introduction
- 0:32What is Deal Boost
- 3:32How is it being used now
- 7:43How do these fit into the requirements for cascade lakes
related talks
50mAI acceleration at the edge
30mAI: unleashing the next wave
09mThe AI stack from chip to analytics
24mTools & Techniques for a cost optimized Analytics & AI pipeline
13mAccelerating AI algorithms
05mAccelerate AI development from cloud to client
30mOptimizing vLLM for Intel CPUs and XPUs
06mUnleashing intelligence and data analytics at scale