Tim Sweeney: The end of the GPU roadmap Epic Games founder and 3D engine architect Tim Sweeney has presented what he calls "The end of the GPU roadmap", where he essentially says that GPU as we know them are too limited, and predicts that by 2020 developers will switch to a more flexible massively parallel programming model where all fixed functionality (texture filtering, anti-aliasing, rasterization) have been replaced with a software implementation, backed by massive computing power. Ubergizmo

Read the whole story

The end of the 3d roadmap? havnt developers been using this kind of setup for years? like the nvidia tesla systems.

"Firefox's billion claim is 'interesting math'" Can only agree.. I've personally downloaded firefox at least 20 times the past month... different computers, different testing and so on.. and I dont even use firefox currently. Everytime a reinstall is made, usually you redownload the browser to make sure you got the latest.

Not wanting to sound like a total jerk, but you don't have to be Tim Sweeney to see this coming. You just need to be him to be hear or to go on the record as saying it. For anyone who follows the tech market, even to a lesser degree, could make an educated guess.
If Intel enters the market and is successfully with their new part (I would imagine that Sweeney has a good idea about what the part can do), the amount of time he has given is more than enough to conclude that prices will go down enough given current momentum and where market leaders are going. We might even see more designs like the Cell processor, going back to a co-processing like design rather... Special drivers using the same components to provide comon system operations.
That is if the software develpoers have tools to keep up with the technologies. A lack of good education can stall hardware too from becoming mainstream. Although the entertainment industry will of course help in that respect, given games are some of the most advanced software used by consumers. Also, granted the economy is much better off sure... but then we are stating some what the obvious - perhaps, no?

"Interesting math"? I'll admit I do like Microsoft Products, but this is just embarrassing. I couldn't tell you how many times I've downloaded Firefox. it was probably highest at work when we still had Windows 98 machines so I downloaded Firefox 2 for them (3 was out, but only 2 will run on 98). I also always recommend Firefox to friends, even if they want to use another browser, if they get fed up with IE (surprisingly many aren't) it's ready for them to try out. The only thing that is "interesting" is Microsoft's lack of a real argument. Have they been hanging around the Apple marketing team lately? *zing...

i think microsoft were pointing out that firefox were using that figure as a brag implying 'users' dowloaded firefox, not total downloads over the past X amount of years. its a newspaper headline 'attenion grabber', its only when you read the full article and realise thats its total downloads of X amount of years that you realise how genuinely uninteresting that bit of news actually is.
How many copies of Win95/WinNT/Win2000/Win98/WinME/WinXP/WinVista/WinServer2003 etc have been sold??? both retail and OEM
because by definition every copy windows is a copy of IE and I bet that that figure makes Firefox's 'billion' look insignificant

Tim Sweeney's comment seems very bold. There is always some headroom associated with using software to access the hardware, and an extra software level makes no sense.
And in the case of GPUs we already have:
1. the OS
2. the graphics API (DX or OGL)
3. and the video driver which makes the calls on the GPU for the required actions
IMHO, besides Intel's Larabee (which no one has seen in action), the only way for 3D graphics stuff to be more flexible is by the graphics API level. With DX10 removing a significant amount of overhead and DX11 promising some of the same, I really don't see how you could improve on it except if you would combine the graphics API with the video driver somehow in order to make faster calls to the GPU.
Could this be how Intel is thinking?
If this is the case, then it means a whole new learning curve for devs to understand the new graphics API. Also, this will mean using a highly limited API, in that it would only work on Intel GPUs.
Either ways, this whole idea seems to reside on Intel to deliver a strong platform with Larabee.

from what i understood of larabee it doesnt need gfx drivers and its just a standard x86 pentium class processor, just loads of the buggers on one die.
so theres the GFX 'driver' removed already.

@ captain828, yes it is bold, but it is very near where intel and AMD may lead the market. Things are moving toward parallel computing. They can still write drivers to interface with the API, the cool thing is that if you have a discrete powerful parallel system onboard, you might not need to buy a new card everytime an API is updated. It might be just as easy as a driver update.
Either way the GPU is changing slowly into a gerneral purpose processing platform that could encompass the workload not only for graphics. This is the case if intel and AMD moves forward. Even Nvidia is trying to develop a platform before the space becomes processor specific like x86. This is where OpenCL and DirectX Computer should allow different processors to come into play as long as they have drivers/compilers to support the languages and APIs.