GPUs to get 570 times faster in 6 years

[GheeTsar]

So says Jen-Hsun Huang:

http://www.fudzilla.com/content/view/15242/1/

CEO Jen-Hsun Huang has claimed that GPU computing will dramatically increase over the next six years.

He thinks that it will get 570 times faster while CPU performance will only increase a staggering 3x in the same timeframe. Speaking to the Hot Chips symposium in Stanford University, Huang said that the advancement would open the door to advanced forms of augmented reality and the development of real-time universal language translation devices.

Huang predicted that advancements in GPU computation would also boost a number of applications such as interactive ray tracing, CGI simulations, energy exploration, and other "real-world" applications.

That man is full of BS in my opinion. But then I know nothing.

[kalniel]

In certain (a lot of) tasks GPU compute is not very mature. If you have a function that you're effectively having to run on the CPU anyway, and then in a future GPU chip you introduce hardware to run it, as well as being able to run it on however many 1000s of shader processors we'll have in the future, I can see a 500x increase in performance.

But that's just a function - doesn't mean you can tie it all together much faster.

[shaithis]

Why would GPU units manage to mature 200 times faster then CPU units...when they are based on the same processes and fundamental principles?

Seems odd to me.

[Pob255]

Ahh but that's the point, GPU's are set up in a different way than CPU's
GPU's are designed to do lots of very small calcuations, CPU's are designed to few but far larger more complex calcuations.
In theory graphics cards are already 50+ times faster than about 5 years ago, that just doesn't make it true in actual game perfromance.

[xPETEZx]

I am always sceptical of these kind of articles. Unless they have some way of seeing into the future, its anybody's guess how accurate they are.

[gaz256]

you should read nearing the singularity it will blow your mind

[nibbler]

I think it will be boring if you can buy a graphics card for £20 that will run Crysis on enthusiast at 1000fps. I like that it's kinda cool that its like an elite group. After all, none of us would be able to show off then

[aidanjt]

Plausible, particularly if software development practices change.

[djfarmo]

In theory graphics cards are already 50+ times faster than about 5 years ago, that just doesn't make it true in actual game perfromance.

So he expects the rate of GPU advancement to increase by a factor of 10 vs. the previous half decade?

[Jay]

lol what a load of crap... what else does his crystal ball tell him?

[JimboTB]

With NVidia starting the ball rolling by accelarating everyday applications I personally am intereasted to see whether this idea will take off and become more supported. Also to watch for is if ATI decide they want to follow NVidias lead and start to do their own version.

[DanceswithUnix]

Nvidia computing power might jump up quite a lot, but only because they have been largely stagnant since the 8800gt. Even then, 570X seems:

1/ Too precise, which is fishy (2 significant digits is marketing, not a valid estimate).
2/ Too big. By at least a magnitude.

[ArCElM]

570 times faster my hairy butt.

it say 30-50 times faster realistically, and thats HUGE.

[Duel]

Great if true, but highly speculative. What is he basing this prediction on?

[Agent]

I just hope the development software can keep up...

[shbris]

lowest spec gpu of 2009 vs highest spec gpu of 2015 could very easily get massive.

say that geforce 295 is about ten times more powerful than, say, a 9400 GT. moores law = double 18months, so 10^(60 months/18months) = about 2000 times more powerful (lowest end chip 2009 vs. possible highest end chip 2015 based on moores law and predicted difference in power between 295 and 9400)

so 570x isn't that extraordinary, unless we're told how they calculated it