The future of hardware and software

[jducoeur]

The hardcore programming geeks may want to check out this really neat slide deck from Tim Sweeney, one of the long-time gods of game technology.

In it, he essentially argues that special-purpose GPUs are a dead end at this point: we should instead be moving back to software renderers, since CPUs are getting nearly as powerful as GPUs and are far more flexible. Along the way, he joins in the chorus arguing that we need to rethink how we're doing our programming, focusing on ways to make parallelism really hum. He specifically calls out pure functional programming languages, software transactional memory, and strong software/hardware support for vectorization as the key elements.

It's fascinating, watching a consensus build within the industry. I'm seeing approximately the same argument coming from a lot of directions right now. I believe it's getting to be steam-engine time for functional programming, at least in some form...

Comments

[dsrtao]

The Wheel of Reincarnation keeps on turning. I had thought it was due a couple of years ago, but AMD and Intel kept postponing things.

http://catb.org/~esr/jargon/html/W/wheel-of-reincarnation.html

[jducoeur]

Huh -- hadn't come across the term before, but it's exactly right in this case.

Anyway, it was a few years ago that they started making all the right noises, but it's taken a while for things to materialize: this particular shift required the companies to throw in the towel on the direction they'd pursued for decades and change course, and that's not quick. It's just starting to get Real now; I expect it to be a few more years before it *seriously* takes off...

[mindways]

Oh, neat - thanks for the link!

(I did enough graphics/3D coursework in grad school that I'm interested and can - barely - follow some of the details, though my practical knowledge of is exactly zilch.)