pouët.net

you said "my framerate stays the same (240fps vs 240fps in my testCase)" so what's the other code you ran?

the same code without using the post-fx-shadercode (snippet) ! whats so hard to understand there, garg?

I used this fine device and could reproduce your measurements

sadly only for resolutions of 1x1.

So basically the performance with a 128-times-texlookup full screen blur pass is exactly the same as without the pass. That sounds like somebody royally screwed up his measurements.

It's called unlimited non-detail!

when we see ms instead of fps measured on GPU, then a sensible perf argument can be made!

I'd rather say "come back as soon as you can present us the results gathered in PIX and/or NVPerfHud and/or AMD GPUPerfStudio".

use the perf tools, luke!

i am not up to it! i KNOW i dont loose any fps, due to years of abusage! las, your turn! also: i said so, its true! ;)
i dont care about ms at all, if the fps stay the same!

kb: its a 256times-texture-lookup btw :p

Translation: I don't care about performance at all, if you don't get enough fps, buy a better computer.

thats all about what intel/pcs are about!
AMIGA ftw!

I'm almost ready for the next level.

Can I request some nice pony pictures about time measurements of convolution kernels?

I'm sorry - this thread is over. If you (the interested reader) want to read some nice information on kernel optimization - switch to page one and start reading until you reach posts about mysterious miracles and high performance hypnoglows.

i am doing 4ks! i know my last one was a fastshot and even won against yours, but i know why! my HypnoGlow was for free!

Our FLOPS beat your FLOPS

i am done here aswell, only Q left is: las, what did you measure at 720p with my snippet?

no, every platform is there to be used properly and to push the boundaries of what it can do not hide lazy hacking without proper knowledge of the hardware. You learn to abuse the hardware whatever system it is and this thread is about optimisation on PC gfx hardware :)

DVS: thats what i did! i answered mu6x on his usage of Blur...and i told him the performance-hit fps-wise is almost not measurable! all this went into francy here, just due to ppl not reading my comments correctly! i said ALMOST! i just told mu6x its unnecessary to downsample by 4 by now, with newest graphics cards. i never had onBoard-GPUs in mind, as i just answered his post, before reading the entire Thread!

no, _you_ couldn't measure it, because your methods are inaccurate.

i dont see why its inaccurate, its working, i can watch my 4ks and see the HypnoGlow working! (except i comment out "#define POST_FX")

oh, my methids to measure are inaccurate, fraps...i see!

I know that it is troll time, but seriously, I don't see why 256 texture lookups for blurring has to be super expensive. The code that is run for each pixel is the same, and those 256 texture lookups are repeated for neighboring pixels. So if the GPU plays it smart, most of the lookups could be cached. And then it is down to 256 MADs per pixel, which should not be much of a problem, or?