You are not logged in. Please login or register.
SmoothVideo Project → Using SVP → CPU vs GPU load
What are the settings that offload CPU and put more load on GPU? I see no difference between ATI HD6450 and nVidia 8800GTS.
GPU-Acceleration
Буфер обмена04.jpg 69.53 kb, 384 downloads since 2013-07-12
Yes, that's obvious. I mean put more work on GPU if it is capable of doing it. Right now my nvidia peaks at 40% utilization on some scenes and averages at 25% on 1080p@60FPS. I want to put more load on GPU and offload CPU as my CPU is at the 80% at the same time (not default settings).
I benchamarked hardware extensively yesterday. As you can see there is very little difference.
FX-6100 Six-Core 4000 NVIDIA GeForce 8800 GTS 512 1606 3038 1899
FX-6100 Six-Core 4000 AMD/ATI Radeon HD 6450 1579 2958 1756Instigater
Yes, that's obvious. I mean put more work on GPU if it is capable of doing it. Right now my nvidia peaks at 40% utilization on some scenes and averages at 25% on 1080p@60FPS. I want to put more load on GPU and offload CPU as my CPU is at the 80% at the same time (not default settings).
i have not english skillz to answer you
In two words: CPU algorithms not applied to the GPU
Instigater
I benchamarked hardware extensively yesterday. As you can see there is very little difference.
how much radeon 6450 is load?
I hadn't GPU monitor installed when Radeon was in my system, so no info about these metrics. I know that there most of the algorithms are not transferable to GPU processing. I just want to know what settings put more work on CPU and what more on GPU.
PS. You can answer me in russian more advanced questions. ;-) I can speak it but my writing in cyrilic is limited.
Instigater
You can answer me in russian more advanced questions.
алгоритмы расчета интерполированных кадров сделаны на CPU и оптимизированы на нем, распараллеливаются на конечное небольшое число потоков (примерно в два-три десятка). Напрямую код нельзя перенести на GPU из-за сложности этих алгоритмов и несравнимо большего числа числа исполнительных устройств на GPU (несколько сотен и даже тысяч), для этого придется переделывать сам алгоритм интерполирования.
OK, I see! Still developer will be forced to do so as single core performance is stagnating and the future is massive parallelism even in consumer computing. That's of course if he wants to support motion interpolation of 4K videos to 300fps in near future. 
SmoothVideo Project → Using SVP → CPU vs GPU load
Powered by PunBB, supported by Informer Technologies, Inc.