Nevada Spring

Started by zaxxon, May 06, 2014, 11:58:58 AM

Previous topic - Next topic

zaxxon

I really appreciate the kind feedback from all of you. This scene is a 'bit' heavy and at this resolution took about 30 hours on my 6 core intel box with 32 gigs of ram (using 24 gigs during render). Clearly I'm not very good at 'optimizing' the render scene  :). I had an earlier larger version cooking at det .85, AA 16, but I turned it off at 68 hours and counting (be nice to pause/save/resume render...).

Thanks Oshayn: experimentation is still what my ST stuff is.

David: Thank you, but I still consider myself a novice ST user; probably not noticeable, but the fallen and two upright trees have blackberry vines growing up around the limbs, just trying out the 'grow around' functions in ST. I'll post a close up shot later.

mhaze: Thank you. Looking forward to seeing some more ST work from you and David as well.

Speedtree is aptly named. It really is a productive and intuitive interface. I've also been messing with the Plant Factory (demo), and while I think it may have more bells and whistles for specialized plant creation, I find it a bit more more demanding and slower to finish a model. It would be interesting at some point to compare the two in the context of TG.

Thanks again.

Dune

30 hours for this size and your machine specs  :o That could surely be brought down, I suppose. And perhaps detail 0.6 and AA6 would work just as well for such small size.

archonforest

I have a feeling that 6 core Intel cpu can be Hypherthreaded. Thus u would have 12 proc cores and that will bring the render time down so we can see more beauties from you more often :)
BTW what CPU u got there?
Dell T5500 with Dual Hexa Xeon CPU 3Ghz, 32Gb ram, GTX 1080
Amiga 1200 8Mb ram, 8Gb ssd

zaxxon

archonforest: the cpu is an intel 4930 i7 3.4 ghz (no o/c), and indeed TG treats the render as 12 cores. I have another i7 (4 core 2.6 ghz), but it will need some upgrades (mainly a liquid cooler) to put it in a distributed render set-up, maybe someday.

Dune: I'll give that a shot. My current 'high' trial renders are at .65/8 and in all honesty they look pretty decent. Still, I remain a 'pedal to the metal' junkie, but I kinda like the .8/8 output; BTW, that is a combo that I gleaned from comments that you made in a prior post as being an optimal/efficient setting. My flirtations with extreme settings are probably over (for now  ;)), as the .85/16 didn't really provide discernible differences (certainly not worth the absurd increase in render times!) from .8/8. About 50 populations and another 20 something individual objects, all a tad over-weight; atmosphere quality at 128 (I know, I know...), soft shadows at 32 samples, clouds at 4 quality, a 32 meg .ter file from World Machine, two suns - one for sky and another for the terrain (another tip from you, gracias mi amigo!). Lots of room to economize I suspect... The real star here is Terragen. I suspect this scene would have choked Vue to death, yet TG just soldiered on cranking out pixel after pixel. Kudos to Planetside for this outstanding piece of software!






Oshyan

Kudos appreciated! But stop abusing poor TG like that, you could almost certainly optimize quite a lot from what you describe. :D I could elaborate if you like, but I'll let you decide. ;)

- Oshyan

zaxxon

That would be most appreciated Oshayn!  I've certainly read many threads here regarding setting's, but an across the board set of optimal settings would be really great (if I'm reading you correctly?).
I've tended to error on the 'generous' side, partly because I just love watching TG render the scene (long time render screen junkie, lucky to have a life), but the scenes are growing more complex of late and some optimizing suggestions would be most welcome!

Oshyan

Well, I'm happy to give my high-level, from-a-distance opinion. But optimal settings do vary from scene to scene. I've seen people create things that really did render better at Detail *2* instead of 1 (fortunately this is a rarity, and last I saw it was so many years ago it may no longer even be true with subsequent renderer improvements). So it's hard to make ideal recommendations without actually looking at the scene "live" and testing a few things. But I think giving good general ranges is at least helpful, so here goes (based largely on what you already mentioned as your settings, and recommendations for changes). Note that this is not a complete "set everything to this for all goodness all the time!" guide. :D

I think your .8/8 baseline is generally good for high quality, "final" rendering. Some vegetation may benefit from higher AA, but in general you're good at that level. It's the other settings that seem a bit unnecessary to me in some cases.

You already hinted that you understand 128 atmosphere samples is too high, and indeed it is. By a lot I'd guess. For the shot that is shown I'd be surprised if more than 32 would even show a noticeable difference, and the default of 16 is even likely to be sufficient (it's designed to provide good quality in normal daylight scenes, after all). Certainly there are cases where increasing atmosphere samples is necessary to reduce noise, but very, very seldom is 128 actually needed for any notable quality gain, and at that point it is generally giving diminishing returns. For complex *localized* light source situations, particularly with fog near the camera, higher atmo samples become more important, but this is a far cry from that. So I'd suggest backing down to 16 samples to start, then render some crops in the shadow areas of both terrain and clouds. That's where you're most likely to see atmo sample noise. In this scene, in the shadow side of the hills on the right, and the underside of the clouds. If you don't see noise, leave it at 16, if you do, raise to 32 and re-render your crops. I would be surprised if you need to go above 32.

Soft Shadow samples 32... OK, so the default is 9 here. IF you haven't changed the soft shadow diameter, then it's largely unnecessary to increase the samples. In most cases any small amount of noise due to undersampling at 9 samples is going to be hidden by rough terrain, vegetation, etc. However if you absolutely must have max quality, then you could increase to 12 or 16. 32 is absolutely unnecessary, in fact I"m not sure I've ever used a value that high. Soft shadow samples seem to be impactful numerically than others, evidenced by the default of 9 vs. 16 for atmosphere for example. So upping to 12 or even 16 has a big impact both on render quality (in difficult soft shadow situations, like larger shadow diameter), and on render time. Use 16 at most unless you're seeing significant noise in a clean and obvious shadow area, which is just not visible in this scene.

Clouds at quality 4... now you're just yanking my chain. ;) OK, so I can see why someone might do this sometimes... but your clouds do not appear to be that challenging and complex. It really should not be necessary. *Especially* with a detail of 0.8. Keep in mind that when Defer Atmosphere is off, the Detail setting interacts with the Cloud Quality setting to produce the final quality. A cloud quality of 1 with a Detail 0.8 should give pretty high quality final results. Quality 2 *might* be necessary with Detail 0.8 when you have more complex, high density, sharp-edged clouds. I would seldom, if ever, go to 4, and then only when main detail was lower, most likely. I also use Defer Atmo/Clouds sometimes, though it is more complex to tune correctly and I don't recommend you do so in a scene like this (too much vegetation that requires high AA for it to be really beneficial).

Finally, the 2 sun approach. Ok, I know this is advocated by some notable artists here, so I'm not going to argue against it on principle. I just continue to be unconvinced that judicious use of other options such as Enviro Light "Strength" sliders cannot produce the same or similar effect, but with more realism and "correct" results. Hopefully you're not using soft shadows on that 2nd sun, regardless. ;)

Also, more generally, consider that rendering at higher resolution can often be more beneficial overall vs. super high settings, and this is more feasible the better your render times. So you could, for example, get faster render times by allowing a bit more noise in the render, but then render at higher resolution and get overall benefit. Just something to consider since your image here is a tad low resolution. Tying into this is the consideration of what is a test or experiment vs. "final" image (I know many of us seldom actually have "final" images, hehe). Where possible, it's best to keep the really top-quality settings for your "final" output, basically. Intuitive stuff, but I know the temptation to crank up settings can certainly be strong. :)

Regardless of all that it's a gorgeous-looking scene. Hopefully the above will help you iterate faster and crank out new versions, updates, improvements, and variations with ease. Let me know if you have any remaining questions.

- Oshyan

oldm4n


zaxxon

Thanks for the settings overview Oshayn.   Many of your recommendations I've seen in this forum, nice to see them gathered into one place. I'll do some crop renders using those levels as a baseline, although even at the outset I suspect I'll still be tweaking them on the high-side  :).  The relational complexity of TG is a fascinating puzzle, I still am just scratching about on the surface and I do appreciate all of the expert feedback offered up, thanks!

Oysterman

Really good work!

Makes me wanna buy a copy of Speedtree.

May I ask which version you use?

Cheers

Tangled-Universe

#25
I think Oshyan's directions on optimizing are quite spot on.

I disagree a little bit on the default 16 atmosphere samples, as I find that always a bit too less with vegetation and shadows of displaced terrains, but despite that I actually do try 16 atmosphere samples first before I increase them.
So like Oshyan recommended, I would do that too.
If I eye-ball this picture 32 should be really fine.

As Oshyan stated it can sometimes be more effective to render the image a bit larger with more modest render settings and then scale it down to the intended size. It is difficult to estimate/judge the trade off, so it's a bit of a matter of experience/experimentation and how serious you are on saving rendertime.
For instance, in this particular case with the finely detailed (foreground) vegetation AA8 is doing a very nice job, but personally I render with little compromises and would render it with AA12 with 1/16 first samples and keeping the PNT at the suggested 0.025 (I'd never change that to a lower value as suggested by TG itself, as Matt has figured these out for us).
On the other hand, if you'd render this a bit larger and keep it at AA8 and then resize it, the result might actually be quite the same.
One of the two is faster, but it's difficult to say in advance.

Furthermore, like Oshyan, I would also like to advocate for using the Environment Light node to get detail in shadows instead of using more sunlights. A second sun (or more) works, but it's not like the real deal and can result in a flattish look quickly if you overdo it.
One might say "don't overdo it then", but it's just not that simple. I'd let the renderer do the work for you and only give it directions on how to "weigh" the indirect lighting in relation to the surfaces.
So indeed, you may like trying to increase the "strength on surfaces" setting in the Enviro Light node to 1.5 for starters, then increase further with 0.5 increments if it's still not to your liking.
If you have TG3 you can even render the whole thing out into render elements and never even have to worry that much about this, as long as it looks close to what you want.
Simply duplicate the "surface indirect" render layer into your comp and voila...you have more detail in your shadows and you can even play with the opacity to get it exactly how you like it without rendering it again! ;D

Cheers,
Martin

Nacer Eddine

wow , nice and clean image
but really i need to some one make video tutorial to explain clearly the best setting to do the render ,
i mean the best time/quality
thanks

zaxxon

Thanks for the additional input TU, much appreciated. Tweaking render settings is very time consuming and to have the experienced folks here weigh in is very helpful. The second sun in this image is a exact duplicate set only for the atmosphere, and set at a lower level, so that shouldn't have much if any effect on the terrain, if any (?).  I'm already using some of the prior suggestions with good effect.  Terragen is becoming a little less Terragen incognito thanks to all the help from the forum!