pouët.net

what'd be faster?

category: code [glöplog]
 
transfering a highres texture or generating it with a procedural texture graph compute shader that calc the fragments. somewhat like a unrolled opstack graph. both targets meant to have a highres dxt texture as result. that just popped up in my head.

i simply need some one to have a idea or give a glimpse of what high end graphics hardware might be capable of. i can't really imagine to compute that based on specs and it's not really that comparable on my slow chip.

if you're interested to check for me... thx in advance
added on the 2012-06-10 00:06:41 by yumeji yumeji
how big is "high res"?
added on the 2012-06-10 00:08:59 by Gargaj Gargaj
yeah. some what like capable to zoom on full hd. so.. maybe 1k, 2k or 4k. i have no idea. 2k is a measure if you walk up close. could even do a second layer for more "details". also i meant generic ops like noize, glow, rect, distort, transform, colorize, normals. that kinda simple stuff.

i know it might sound weird. i don't care. i just had idea and i can't check it with my hardware. lol
added on the 2012-06-10 00:48:33 by yumeji yumeji
The correct answer is: it depends
memory copy from gpu to gpu is cheaper than calculating the texture. I'd say even a memcopy from cpu to gpu should be cheaper, depending on the complexity of the calculations.

noise, glow and such are not cheap.
added on the 2012-06-10 16:09:25 by xTr1m xTr1m
yeah. thx. i expected it to not be worth it. just thought it could make level loading in games faster. than transfering disc drive->memory->gpu texture. where the shader is smaller to load and do fast spit some generic sh*t into vram.
added on the 2012-06-10 16:49:44 by yumeji yumeji

login