A general purpose GPU won't replace a CPU, at least not an x86 CPU. You're trying to break the x86 monopoly and to be quite honest, if Intel couldn't break a monopoly they own then nVidia isn't going to be able to do it. I think this monopoly is too difficult to break because it is so entrenched. It's not about technological superiority so much as there is so much momentum and entrenchment in the x86 market.
It doesn't matter that nVidia's GPU would be slower at CPU type work. It could even be good as a replacement to the x86 CPU. It doesn't need to be great IMHO because Joe Consumer doesn't need much computing power. Just enough to browse the web, play some movie files and play some music. The x86 CPU is simply too entrenched and you'd have to be not just good but an awesome replacement to it in order to pose a credible threat.
One of the reasons AMD purchased ATI was always the convergence of a CPU/GPU on one die that would be sold as a one stop solution for vendors. Intel is also doing the same with an increased emphasis on GPU's (Larrabee).
What you'll likely see from AMD and Intel in the future is a multi-core CPU/GPU implementation and you're dense if you don't see nVidia being scared out of its pants. GPU's are getting better and better every year. Process shrinks are also progressing. While GPU's being better every year definitely benefit the ATI part of AMD as well as nVidia, it also aids the work going on towards a unified multi-core CPU/GPU. One definitely can see the day when a multi-core CPU/GPU implementation will be good enough to play games at a decent level and only the most demanding games will require a discrete GPU.
With where CPU's are advancing today and how they are advancing, I don't find it a crackpot theory that AMD and Intel can make an 8 core CPU/GPU in two to three years that provide what would be considered mid-level GPU performance (perhaps equaling the high end GPU's today) coupled with decent CPU performance. A low end CPU/GPU would be one with two CPU cores and two GPU cores and higher end ones would contain upwards of three of each cores in various configurations. These GPU cores would be run in a SLI/XFIRE type config.
OEM's will consider an integrated CPU/GPU a blessing. Less parts to keep track of and integrate into their assembly lines. Joe Computer could care less, just get them something that performs OK and works. AMD and Intel would happily take more money for these integrated CPU/GPU parts. AMD and Intel while they would probably make more on both a separate CPU as well as GPU sale, they would make more overall since they guarantee if you buy an AMD CPU (or Intel), you're also buying an AMD GPU (or Intel).
As any impartial observer can see, nVidia should be scared by the possibility of the lower mid-range GPU's being sliced off by AMD and Intel. They could be losing not just integrated graphics sales but a large majority of the video card sales from the lower mid-range on down. Their only recourse is to create technologies and software that necessitates the use of nVidia branded GPU's. That's where CUDA and PhysX comes in. However, it's no guarantee that CUDA and PhysX will be here long term either. While nVidia is king of the hill on the GPU front today, that can easily change tomorrow. They must continue to push CUDA and PhysX and they must also invest in improving their GPU performance.
Facebook
Twitter