I am testing the IPP Asynch examples on the new i7 4770 CPU and I observe that the speeds of the CPU and the GPU acceleration are equal. Shouldn’t I expect a speedup with the GPU accelerated implementation? Should one configure something at the beginning of the GPU accelerated implementation (like for the CPU num threads previously)? Can anybody share some of her/his experiences with the new GPU library?
Nice to hear, that you're experimenting with Async. Could you tell what Async functionality do you work with and what image resolutions are?
Faster than i7-4770 @ 3.4 GHz? I doubt :).
Non-linear resizes (cubic and Lanczos) could be of equal performance because of more computational-intensive algorithms than simple linear resize. At least, on my i7-4770MQ @ 2.4 GHz only Lanczos resize is faster on GPU.
Nevertheless, one of potential benefits of GPU execution (in the case of integrated graphics the GPU can hardly be called as "accelerator") is CPU offloading - i.e. moving of part of job off the CPU - and power efficiency. The most visible goal of integrated gfx execution is perceptual computing, which if being executed on CPU, will drain the battery very quickly.