HackerTube

THE VIDEO SHELF · deep learning hardware

Trends in Deep Learning Hardware: Bill Dally (NVIDIA)

Bill Dally of NVIDIA gives a University of Washington lecture on trends in deep learning hardware. It covers numeric formats, specialized instructions, energy per operation and co-design of hardware and software.

Watch & explore the libraryYouTube ↗

Why HN readers recommended it

HN commenters cite it for NVIDIA's perspective on AI hardware efficiency, including logarithmic number formats and energy use.

This is a summary from the source dataset, not an endorsement by Hacker News or the video creator.

Video & recommendation details

Creator / channel
Paul G. Allen School
Topic
deep learning hardware
Format
lecture
Level
advanced
HN recommenders in dataset
3
HN mentions in dataset
5
Recommendation years in dataset
2023–2024

These counts and years describe the included Hacker News records; they are not YouTube views or the video’s publication date.

From the original HN conversations

Excerpts from the recorded comments. Follow each source for the full comment and surrounding discussion.

  • It looks like both Pat Gelsinger and Lisa Su have no clue about software and can't even delegate this. They expect "the community" to do the software for their extremely complicated hardware. If you see the presentations by Nvidia's Bill Daly [1] it shows they've …

    alecco on Hacker News · 2023-12
    In Intel CEO: 'The entire industry is motivated to eliminate the CUDA market'

  • Depends what you consider as part of chip production. If you consider new instructions and the related software as new chip production, then the performance improvements have continued quite dramatically. Single chip inference has gone 1000x over the last 10 years according to Bill Dally …

    huijzer on Hacker News · 2023-12
    In ASML and Samsung seal deal on 2nm chips

  • In a recent talk[1], Bill Dally of Nvidia suggests using actual logarithms, as they're more consistent in relative error than floats, which is especially relevant at low precision as increasingly used in AI. Multiplications then become simple additions, but then addition becomes more complicated, as …

    raphlinus on Hacker News · 2024-01
    In Why does integer representation of a floating point approximate its logarithm?

Summaries, classifications and recommendation records are from the HackerTube source dataset. Read about this library and its limitations.

Keep exploring