Originally Posted by Brutuz
Yeah, it would be like having a 128 core CPU, as the shaders on a GPU are like Cores on a CPU.
Originally Posted by Licht
No not exactly since they can't handle central (for the C in CPU) data. The GPU core would be counted as a single core, then that core is only used in certain circumstances. If the data could be processed in a GPU, then it will go to the GPU core, otherwise it gos to the CPU core(s.) Really ingenious concept.
Right...not quite the same...
But perhaps that makes the GPU "CPU" more powerful.
Current CPU cores are semi symmetical units. Each core has cache, an FPU, an ALU - or perhaps multiple of each. This allows different programs to run off each core without a hitch. A GPU, on the other hand, has "shaders". I believe that's a fancy name for an FPU that has direct access to the onboard memory, or something like that. There's only one GPU - one core - but the core is so massively powerful, that it can process every task at the same time. It has so many FPUs that it could easily process data for 50 threads or processes without getting close to its capacity.
...assuming the main core can manage moving all that info around.
Now ALU operations might take a hit, but the modern GPU has more than enough FPU power for any task. A properly utilized GPU could probably encode video at 20x the speed of the current quad-cores(conservative/low estimate). That's probably why Intel won't ever give nVidia an x86 license.
Letting your competitor make a product 20x faster than yours(and cheaper) is a bad idea - especially if nobody realizes that the tech is there, so it's not like they'd complain if you didn't