[QUOTE="ronvalencia"][QUOTE="AMD655"] I do not think you understand what i am talking about, no offense. Let me swing this another way, look at how GPGPU works, a GPU is highly parallel(2) and can work through any task much faster than a CPU, but can also be less accurate(3). Why do you think engines are not parallel?(1)CPU's simply are not even remotely fast enough to do the work through a parallel method(4), not real time at least. This is why we have Quadro/Firepro for parallel compute tasks(4). This metod is not out of the question by a long shot, but for the most part, is entirely not viable any time soon with games(5).
AMD655
1. I posted two reasons for multi-thread issues with the PC e.g. DX9/DX10 multi-thread overheads and NVIDIA PhysX Wintel edition. The posted chess game benchmarks doesn't render to a DX device nor it's gimped by NVIDIA PhysX Wintel edition.
2. GpGPU enforces parallelism i.e. known as explicit parallelism. AMD GCN has a scalar processor for each CU and this is not directly expose on PC's DirectX.
3. One should treat GpGPU as "many core" very wide SIMD processors.
4. CPU simply has less ALUs compared to "fat" GpGPUs.
5. When programmed correctly, a game's CPU side processing scale with CPU count.
I already mentioned games that scale with core count, they however show nothing gained with more cores, yet can use up to 8+threads. 4 faster/efficient cores will be faster in gaming than high latency/low IPC cores. A single Intel Sandybridge core can allocate unto four X86 instruction retirement per cycle rates to 1 thread while a single AMD FX core can only can allocate upto two X86 instruction retirement per cycle rates to 1 thread.
Intel Core i7 qaud core/ 8 threads has 16 X86 instruction retirement per cycle rates.
AMD FX 8xxx 8 core/ 8 threads has 16 X86 instruction retirement per cycle rates.
Intel Core i7 can act like AMD FX 8xxx during 8 threads workloads.
Log in to comment