And not everyone likes Nvidia

CUDA is like an lcd, fast response times are nice but there is also a hidden "input lag" that prevents anything computed promptly as on a normal cpu.

CUDA also requires procedurally linear execution patterns. When I said "simple" in the earler quote by Shadow Lady, that's what I was referring to -- you can't have a bunch of conditionals in your CUDA app or else it'll be incredibly slow. This makes CUDA very ideal for performing certain types of non-conditional algorithms on flat arrays of data, such as MPG encoding (something for which CUDA is surpassingly ideal for, hence why it's the most common thing you see it being demo'd on), and also on large-scale particle effects with 50,000+ particles.

As for things like the earlier video posted, I'll guarantee you with a chocolate glazed donut that the i7-based example that runs 2fps is bias. Implementing correct and efficient threading is something most programmers have no idea how to do (and which is probably one of the biggest advantages of CUDA, since it does it for you). I've done particle effects engine work and there's no reasonable explanation for 60k fluid particles running 2fps on a CPU with 7 effective cores (hyperthreading is especially ideal for that sort of flat-array parallel processing task, hence the core count). Consider for a second: you have what should be an effective average throughput of 60 billion instructions per second, based on a 3ghz cpu with 7 cores and 4 instruction pipelines. If you're using SIMD instructions then in theory a lot of the work is divided by 4 (and this type of particle effect should scale well to SIMD, tho on the other hand intel SIMD is pretty sucky).

So let's be conservative and say 100 billion operations per second, accounting for various pipeline stalls and poor use of SIMD features. At 2fps that's 900,000 instructions per particle, per frame. How on earth would any particle effect need 900,000 instructions per particle? Even 100,000 instructions per particle would be by my measure an extremely intense logical effect, and at that rate the i7 would be capable of 20fps, not 2.

Is it going to hit 45 fps like the CUDA example? Probably not. But 2fps is a joke. Seriously.

In any case, PCSX2 has very few areas where such linear non-conditional processes can be applied. Texture swizzling is about the only thing I can think of that could be efficiently broken down into smaller jobs spread across many threads. And yeah that's something that could benefit a few specific games so maybe someday we'll play with that. But most games don't do that much large-texture swizzling, and so they would have minimal benefits .. so it's not a high priority.

Swizzling has almost no impact on any game nowdays, improving that would not gain more than a very few fps, uploading textures and updating constant buffers are the slowest part (as far as cpu is considered, gpu can stall a lot when render targets are reused and it has to sync the results, the super fast gpu turns out to be a snail suddenly, when latencies cannot be hidden).

quantum processor is.

curses some of you beat my "long answer" in length AND awesomeness Oo

^^^^ lol , they sure did. aldough I do not understand half of the reason. the half I do understand is enough to understand its no benefit to use cuda for this type of aplication.

well,i have to say, unless process of making emulator is changed from
the condition now,it is impossible for building ps3 emu.Just because
u need atleast 10 time powerfull machine(clocked at 30ghz in total with several core cpu,10 gb ram and decent gpu).Probably u could manage ram and gpu of that but u will stop at cpu.

The worst is PC cpu is not 100% dedicated to gaming.

Be aware, their are lots of fake PS3 "emulators" out their. I made a video on youtube exposing the people who make make vids on them. haha... the swines!

nice work dude!