| Pemalite said: Then again, it's all mathematics in the end, just the CPU can handle allot more complex math than a GPU, whilst the GPU can handle more mathematic problems at once. |
GPUs & CPUs can handle the same math, it's just a matter of how quickly you need the result. GPU "threads" are each executed very slowly compared to a modern CPU core.
For example a CPU core could execute several instructions from the same thread, including perhaps SIMD, each cycle at say 3 GHz. A single thread may hypothetically run at say 3 Giga-instructions-per-second.
A GPU like GCN runs at say 1GHz or less so it's already 3 times slower per-thread, then add up to 40 warps sharing a core (similar to how CPU hyperthreads share a core) and each instruction is 'scalar' (no SIMD). You will realize that the GPU may run a single thread hundreds of times slower than the CPU.
Therefore if you need the answer to a single math solution quickly then the CPU will win every time. On the other hand the GPU can run 10s of thousands of threads at the same time, so if you need 10 thousand math solutions then the GPU will win every time, and more power efficiently too.
That's why doing graphics on CPU is a terrible idea, and becomes even more terrible over time. The future is more specialized silicon. Graphics on GPU hardware. Video encode/decode in hardwre. Computer vision in hardware. Motion processing in hardware. etc. etc.
