Nvidia turns up the AI heat with 1,200W Blackwell GPUs | Hacker News Reader

Nvidia turns up the AI heat with 1,200W Blackwell GPUs

theregister.com

44 pointsPaulHoule13 comments

Perhaps the claim is that Moore's law hasn't ended, but it's sweating!

Maybe I should get a few and make a water heater around them and rent out the compute power for a double whammy.

In the keynote speech Jensen joked about selling a jacuzzi add-on

its nice it's funny, climate crisis in full swing and here we now have 1200W graphics cards which are in ridiculously high demand.

> in ridiculously high demand.

It depends what you mean by "high demand". It's extremely unlikely you'll be able to buy one for less than $20k.

[dupe]

More discussion: https://news.ycombinator.com/item?id=39749646

https://news.ycombinator.com/item?id=39764988

I'm thinking we shouldn't call these "GPUs" anymore. Would help me in filtering out uninteresting headlines if nothing else.

We'll just start saying the "G" is for "General"

It should just be Greatly Parallel Unit at this point.

I am curious on which applications require/prefer FP64?

It's critical for many scientific simulations: fluid dynamics, finite-element analysis, that sort of thing. A few years back I was helping try to track down and buy "outdated" K80s (or were they K40s?) to run a big FEA project because they still had the best FP64 performance reasonably available despite being otherwise outclassed.

It is much easier to list the very few (but very important) applications for which FP64 is not necessary: machine learning/AI, graphics and video and audio processing, other digital signal processing.

The list of applications for which FP64 is required is huge, but they are less known because they belong to scientific/technical computing. Any design engineer would use such applications, but most programmers and non-professionals would not be familiar with them.

And the military:

https://www-users.cse.umn.edu/~arnold/disasters/patriot.html

Reply on news.ycombinator.com