The Future of Computing: AI technologies in the exascale era
summary
Explains why exaflop counts matter for throughput but not for typical AI precision, cites two systems already hitting exaFLOP and contrasts GPU matrix throughput with CPU flexibility, names Intel Gaudi architecture and EIQ as concrete alternatives and toolchains, gives device-level constraints (edge mW–W budgets) and data-center figures (50–150 MW today, training a big model could approach a GW), calls out secure boot, certificate rotation and model distillation as deployed security practices, and flags neuromorphic research like Loihi‑style chips.
AI-written summary · about
chapters (11)
- 0:00AI in exascale era
- 2:06Exascale computing for AI
- 6:02GPU vs CPU for AI
- 9:57Edge vs data center AI
- 13:07Energy-efficient AI architectures
- 15:33AI security challenges
- 16:57AI software optimization
- 21:53Beyond exascale computing
- 25:46Neuromorphic computing
- 27:53AI industry partnerships
- 29:38Q&A: Distributed vs centralized AI
related talks
58mThe Next Decade of Software Development
35mAI innovation begins with trust
04m4 real-world AI deployments at enterprise scale
21mAhead of the Plasticity Curve: AI, Biology and Technology
29mEnhancing AI with Edge Computing: A Leap Towards Affordable Performance
30mBeyond GPUs: Why AI Agents Need a Need New Hardware
01h 07mThe Next Decade of Software Development
21mAnalyst Notebook