I’m a printmaker artist and I’m planning to export tremendously large files for big prints (100x150 cm) in inkjet. I know the theory that says “the larger the print, the less dpi”, and I usually apply it to my own work. But this time I have a printer able to make really accurate images at 450 dpi and I feel sort of sorry seeing my renders printed at 180 (the noise is really, really bad when it gets bigger). Actually, my goal IS exploring ultra HD.
As my working computer have a poor AMD Radeon 6770M without OpenCL support, I use the CPU with 10 samples (!) per tile of 16x16 with 7 threads, counting on the ultra resolution to smooth the noise off optically. It works good with images until 12000 x 8000 pixels, which I rendered in something like 12 hours.
But for larger scale, the expected time is just ridiculously long (horrifying 160 or 700 hours). I tried to switch on a friend’s computer with GPU support, but its CPU(?) doesn’t seem to be powerful enough to handle a 17k x 26k px file. It crashes. Now I’m rendering at 11k x 17k px (it gives a 300 dpi file which is an okay resolution), with an expected GPU rendering time of 13 hours, as the tiles are set at 256 x 256.
But I was wondering about the number of threads for GPU rendering. Is it better to set them at a certain value, or the auto does it well? In my current massive GPU render I can only see one tile at the time. Increasing the size of the tiles at 612 x 612 or more does have any effect?
I’m also wondering about the number of samples. At least, I quite like the noisy render when the resolution smooths it off as I said, but maybe it would be more rational to increase the samples and do the render in a smaller scale, then increase the resolution with a Smoother Bicubic in Photoshop. Doing this with 10 samples gives an awful blurred noise.
I hope you’ll forgive my messy explanations! but as often in printmaking, things have a lot of factors and it’s difficult to clearly isolate them, specially when writing in a non-native language.
I don’t expect a brilliant and immediate resolution of this question right now, because these things depend of experience. If anyone knows it well, you’re welcome. I’ll keep posting researches here, if someone is interested.