PDA

View Full Version : Antialiasing LW11+



Mr Rid
02-08-2013, 11:55 PM
I am way behind in LW and just now trying v11.5 AA. I watched the tutorials and read, trying hard to resist my brain glazing over at mathy explanation and four more values to try to figure and balance.

But using the attached test scene, I cannot begin to get the same quality and render time in LW11 that I get in LW10. I understand that leaving the Lighting, Shading, Min, & Max samples at "1" is equal to Antialiasing "1" in LW10(?). So by only trying various Min & Max sample values to get the same quality, I only wind up with longer render times in v11.5.

What LW11.5 AA values would get the same quality and render time that I get in LW10?
111294

Since I think v11.5 handles reflection blur differently, turning all reflection blurs to "0" did not make any difference.

gordonrobb
02-09-2013, 02:37 AM
I'm a little confused by it too. I'm sure there'll be good documentation out there. Anyone got any pointers as to where to look?

madno
02-09-2013, 02:52 AM
In case you have not seen it.
This is by Rebel Hill explaining the difference between AA in LW 10 and LW 11.

http://www.youtube.com/watch?v=UgNB9tZWUmM&list=PL8A1C0DB658775A63&index=2

zardoz
02-09-2013, 03:16 AM
well I tried your scene and try to insert 4 in the light samples and keep 1 in the shading samples, then for the camera min=1, max=4 (I tried 8 too) and the AS=0.001
and got nice results. see if you can compare these

bobakabob
02-09-2013, 03:42 AM
I'm a little confused by it too. I'm sure there'll be good documentation out there. Anyone got any pointers as to where to look?

Good thread. I'd also appreciate some straightforward advice on generic render settings in LW along the lines of Zbrush e.g. 'basic', 'medium', 'good', 'excellent', as a starting point.

I've tended to start by setting lights / shading to 1; AA set to min 5; max 20; threshold 0.01; over sample 0.5 then work from there for good results. I'm still not confident I'm getting the best out of LW though and don't fully grasp the threshold and over sample values. Could someone confirm which values equate to best quality exactly? The default radiosity settings in Lightwave produce really nice results but again it would be great to see some suggested generic settings from Lightwavers for good quality renders.

gordonrobb
02-09-2013, 04:28 AM
In case you have not seen it.
This is by Rebel Hill explaining the difference between AA in LW 10 and LW 11.

http://www.youtube.com/watch?v=UgNB9tZWUmM&list=PL8A1C0DB658775A63&index=2

Now that was informative. Thanks very very much.

madno
02-09-2013, 06:01 AM
I rendered your scene as well. Seems to me that the final render size is 600x600 but the preview window in your sceenshot is larger.
Tried it and noticed that LW 11.5 scales the small render into the bigger preview. This upscaled preview shows less quality then the real render.
Regarding the reflection blur - set it to 1% in your GREEN material. That matches the LW 10 value better.
Another thing I tried: I set the Threshold in Camera setting to 0.01. As I understood it, it defines when LW considers pixels to be to different and therefore thows another AA pass (lower values give better quality but higher render time) - but on the other hand I reduced Maximum Samples to 32.
Changed also some other settings (no double sided area lights, and some settings in global illumination (less bounces, no volumetrics etc.))

I don't know if this makes your render faster, as I am by no means a LW expert.

111299111300

111301

Tobian
02-09-2013, 07:42 AM
I started to fix this, and then I realised I neither know how fast your system is, or have LW10 installed any more so oops :D

As for optimising, I find that you need a lower oversample in LW11, especially if you're using LWF, as it looks blotchier. For blurrier scenes, I find higher than 1 shader samples are needed, or it's too much noise to clean up. I also find you need higher than 1 min samples, if you've got anything in the scene which is brighter than 100%, because otherwise you'll get fireflies, which won't go away.

I did a quick optimise to get it down to about 30 seconds here. (35) I put in DB&W's schlicks approximation node to kill reflections during the radiosity pass (though this works also in 9.x up) which sped it up, and did a 4/8 min/max which ballanced out the speed/noise to acceptible levels. Also you weren't quite playing on a fair playground there... The reflection blur in LW 11+ causes WAY higher blur per value (which is observable in your comparison images) So likewise, I lowered the ammounts, to make it fairer (as aa/sample passes have to do more work for blurrier surfaces). See how this looks on your system?

jwiede
02-10-2013, 01:10 AM
I rendered your scene as well. Seems to me that the final render size is 600x600 but the preview window in your sceenshot is larger.
He's viewing the render at 200% in both LW10 & LW11.5 so that alone doesn't explain the render quality difference.

Mr Rid
02-11-2013, 02:56 AM
I understand that LW10 Camera Antialiasing value of "1" is equal to LW11+ Shading Samples"1", Light Samples "1", and Camera Minimum Samples to "1" (no?)... then it should only be a matter of raising the Camera Maximum Samples... but any variation of both Min and Max values are not getting a render of equal AA quality and render time as in LW10. One should be able to leave the AS at ".03" and only adjust the Camera Min & Max values in order to get the same result as in LW10. The whole point of the newly added AA values is to optimize renders. So I dont know what I am missing in the example scene.


In case you have not seen it.
http://www.youtube.com/watch?v=UgNB9tZWUmM&list=PL8A1C0DB658775A63&index=2

Yes I saw that looong explanation. As always, I appreciate RH' taking the time, but I think the pertinent info could be more concise. And it does not shed light on why the example scene looks worse, or takes longer in LW11.


well I tried your scene and try to insert 4 in the light samples and keep 1 in the shading samples, then for the camera min=1, max=4 (I tried 8 too) and the AS=0.001
and got nice results. see if you can compare these

I find that your suggested settings take almost twice as long to render and still looks noisy and jaggie in LW11. I attached a new version of the Box object with no reflection blur. But LW11 still wont compare to Q and render time.


I started to fix this, and then I realised I neither know how fast your system is, or have LW10 installed any more so oops :D

As for optimizing, I find that you need a lower oversample in LW11, especially if you're using LWF, as it looks blotchier. For blurrier scenes, I find higher than 1 shader samples are needed, or it's too much noise to clean up. I also find you need higher than 1 min samples, if you've got anything in the scene which is brighter than 100%, because otherwise you'll get fireflies, which won't go away.

I did a quick optimise to get it down to about 30 seconds here. (35) I put in DB&W's schlicks approximation node to kill reflections during the radiosity pass (though this works also in 9.x up) which sped it up, and did a 4/8 min/max which ballanced out the speed/noise to acceptible levels. Also you weren't quite playing on a fair playground there... The reflection blur in LW 11+ causes WAY higher blur per value (which is observable in your comparison images) So likewise, I lowered the ammounts, to make it fairer (as aa/sample passes have to do more work for blurrier surfaces). See how this looks on your system?

No one needs to compare processors. The point is just to render in LW10, then try to get the same quality and render time in LW11. One should only have to adjust the Camera Min and Max samples. I find that lowering the ".6" oversample only takes longer, and doesnt make a significant difference in the example render. The reason I use ".6" is just because it is (or was) very close to classic "Enhanced" which serves general purposes fine for me, along with ".03" AS.

I saved a new version of the Box object (and scene attached) with no reflection blur, to keep comparisons even (and no DP plug that others may not have). Your settings produce a comparable render for me quality-wise, except for the area around the light reflected in the ball, but it still takes longer in LW11, than in LW10.
111350

So, I still dont see the 'optimizing' benefit of the new LW11 AA values, using the example scene.

Hieron
02-11-2013, 03:39 AM
While you are not the first to mention LW11(.5) may be slower, in your first post ( regardless of lengthy OS and AS consideration) the reflection blur is way way different since it was revised. That combined with any AS will slow down rendering by a lot. So, the difference in the first renders you show is directly related to that for a large part.

When comparing these things, it is better to go to the basics and turn of AS first, and see whether the actual calclualtions are slower or not. Without AS, blurry reflections play a lot smaller role too. In your last example, there seems a difference still, but less pronounced. I would still suggest to ditch AS first and go from there.

BTW, there are very little production scenes of ours that benefit from AS. I suppose we do not have enough "flat" areas to make up for the extra time spent to define areas that need more samples, since AS almost always leads to slightly higher render times and still cover almost all pixels. In alot of our renders, there is an inherent variation between pixels and AS will just keep pushing in samples. Afaik, it is not yet smart enough to keep track of whether samples are actually changing or not, between passes.

AS will also kill repeatable noise patterns, making it a tad trickier for a tool like Neatvideo to take out. Just saying.

ps: some settings I noticed, and seemed not ideal:
-forcing threads to 4, Is that better than auto mode?
-"Reduce noise", will give horrible edge artifacts and is legacy? Turn it off and see how all edges improve vastly. Never ever use it (imho), do get Neatvideo for AE etc. Honestly, this function should be flat out removed. If not, I'd like to be enlightened. Right now it lures people in. It was the first thing I noticed about those renders.
-AS will be helpful on such a scene with quite flat color areas indeed.
-double sided area lights, will blow out the white poly above and make GI splotches due to that a lot harder to control

Ended up with:
111353

Don't have LW10 anymore, so no idea how it compares. Wouldn't use LW10 anymore anyway.. so a bit of a moot point for me... Ow and not sure if it is optimized in the sense of it being much faster, but it is a tad more controllable now since it is easier to maximize the samples used.


... I put in DB&W's schlicks approximation node to kill reflections during the radiosity pass...

Didn't know that one, got to try it out! Thanks.

Hieron
02-11-2013, 05:09 AM
Nice and useful! didn't spot the new DB&W tools, if you click the name on their site it actually links to the 1.8 version. Clicking on "1.9" does bring you to 1.9 version though... cool:
111354

Rendertime seriously reduced to 9.9 seconds.

The difference between refl included or excluded is imho very minor in this case:
111355

(see left above, Photoshop difference mode: mostly dark and no difference)

Thanks for pointing it out Tobian and Mike for coding it!
Is there a scene wide global switch for this btw? Seems useful....

Mr Rid
02-11-2013, 05:41 AM
While you are not the first to mention LW11(.5) may be slower, in your first post ( regardless of lengthy OS and AS consideration) the reflection blur is way way different since it was revised. That combined with any AS will slow down rendering by a lot. So, the difference in the first renders you show is directly related to that for a large part.

Once again, the last example I posted uses no reflection blur. It does not make a difference in the comparison I am making. I think you missed my point that LW11 AA appears to be more complicated while taking longer to render for the same quality in LW10.



When comparing these things, it is better to go to the basics and turn of AS first, and see whether the actual calclualtions are slower or not. Without AS, blurry reflections play a lot smaller role too. In your last example, there seems a difference still, but less pronounced. I would still suggest to ditch AS first and go from there.

Again, this defeats the point of comparison I am making.


BTW, there are very little production scenes of ours that benefit from AS. I suppose we do not have enough "flat" areas to make up for the extra time spent to define areas that need more samples, since AS almost always leads to slightly higher render times and still cover almost all pixels. In alot of our renders, there is an inherent variation between pixels and AS will just keep pushing in samples. Afaik, it is not yet smart enough to keep track of whether samples are actually changing or not, between passes.

I've experimented extensively with AS (and squandered WAY too much time debating about it), and have experimented with many example scenes that others post, and I consistently find the best balance of quality and time by using AS. I found only one rare exception scene.




ps: some settings I noticed, and seemed not ideal:
-forcing threads to 4, Is that better than auto mode?
-"Reduce noise", will give horrible edge artifacts and is legacy? Turn it off and see how all edges improve vastly. Never ever use it (imho), do get Neatvideo for AE etc. Honestly, this function should be flat out removed. If not, I'd like to be enlightened. Right now it lures people in. It was the first thing I noticed about those renders.
-AS will be helpful on such a scene with quite flat color areas indeed.
-double sided area lights, will blow out the white poly above and make GI splotches due to that a lot harder to control

Ended up with:
111353

I dont know who setup this scene originally, but these settings have no bearing on the AA comparison I am making between LW10 and LW11. The same scene should not be taking longer to render in LW11, when the new AA values are supposedly about saving render time.


Don't have LW10 anymore, so no idea how it compares. Wouldn't use LW10 anymore anyway.. so a bit of a moot point for me... Ow and not sure if it is optimized in the sense of it being much faster, but it is a tad more controllable now since it is easier to maximize the samples used.

Well, you definitely missed my whole point. If you dont have LW10 to test with then you cant really make a useful comparison. The new AA values are more controllable, but the practical benefit eludes when applied to the example scene. I tried your camera settings and it is taking much longer than the settings we've thus far been over. So far, no one has come up with settings in LW11 that produce the same AA quality in the same render time as LW10. So the new sampling values are not optimizing anything in the example scene.

Mr Rid
02-11-2013, 05:49 AM
UGH, the new forum is really annoying... it keeps hanging for minutes at around every last quarter of each hour, and oddly double posting on me.

Hieron
02-11-2013, 06:33 AM
Once again, the last example I posted uses no reflection blur. It does not make a difference in the comparison I am making. I think you missed my point that LW11 AA appears to be more complicated while taking longer to render for the same quality in LW10.

Err yes, it does matter when AS is involved since it will be using a ton of samples to try and fix the blurring. More changed than merely AA samples btw.





Again, this defeats the point of comparison I am making.

I've experimented extensively with AS (and squandered WAY too much time debating about it), and have experimented with many example scenes that others post, and I consistently find the best balance of quality and time by using AS. I found only one rare exception scene.


Hey man, just posting our results. Apparantly they differ to yours, doesn't make it useless and I did put in "afaik" and "imho"'s. Apparantly I squandered my time here for sure.




I dont know who setup this scene originally, but these settings have no bearing on the AA comparison I am making between LW10 and LW11. The same scene should not be taking longer to render in LW11, when the new AA values are supposedly about saving render time.

More things changed between LW10 and 11 than just the AA, the reflection blur being the obvious thing in the first post but it won't be the only one.
While the other settings have no bearing on AA comparison (actually, they do but hey.. ), they do seem quite sloppy and I would think it normal for me to point them out. I suppose not since it doesn't support your AA case.


Well, you definitely missed my whole point. If you dont have LW10 to test with then you cant really make a useful comparison.

I did not miss the point and did not make any comparison at all with LW10, I just added my input to things spotted in that scene. While all of those seemed good input, you do not seem to care as it is not in your exact concern.


The new AA values are more controllable, but the practical benefit eludes when applied to the example scene. I tried your camera settings and it is taking much longer than the settings we've thus far been over. So far, no one has come up with settings in LW11 that produce the same AA quality in the same render time as LW10. So the new sampling values are not optimizing anything in the example scene.

Some people here improved a ton in that scene, yet you do not seem to care nor appreciate. I have no LW10 and LW10 will change more than just the actual AA settings used so it is not a fair comparison per se. I did not only change the camera settings and I'm pretty sure that the 9.9 seconds end result in LW11.5 is quite nice on this simple quad core 2600k. Heck, the GI pass alone took 8 seconds in that LW10 scene when first opened.

Your render shows 16 seconds LW11 GI time so I suppose your CPU is 2x slower than mine (if the difference is less, the improvements shows are greater) so your 30 seconds LW10 time would be 15 sec here. So the 17 seconds I posted first is not THAT bad (and it has little noise like yours) and the 9.9 seconds is a vast improvement (it could be done in LW10 as well)..

So dunno man, but you seem to have your opinion set no matter what. Imho, I would be amazed if LW11.5 is actually really slower to render per sample. And you first comparison with AS and blurry reflections shows me that if you did spend tons of hours, you missed the point. What is your cpu at least? Cinebench points are very representative and that way it saves me installing LW10 to figure out a comparison with your rendertime.



UGH, the new forum is really annoying... it keeps hanging for minutes at around every last quarter of each hour, and oddly double posting on me.

Yes it does, always. Would be nice if they fixed that. For some magic reason I end up trying post around that exact time very often.... :/


ps: I am not saying the LW AA system is amazing or improved. I think just too many things changed from LW10 to compare directly and the max samples amount will be helpful. Regarding general open AA issues: for example the moire patterns of certain procedural textures are annoying to get rid of. "low discrepancy" doesn't help at all and was bugged together with OS for a while until 11.5. Not sure if it is correct now or just ignoring OS.

Tobian
02-11-2013, 11:25 AM
Mr. Rid, I don't have LW 10.x installed either, (which I already said) so I can't directly compare the two in your example, hence I uploaded my scene for you to try in 10.x as a comparison? Which is why I made the comment about your CPU speed, as CPU speed is a big issue on render speed... DUH.

I expect that yes, in your very specific example the render was slow but I have never in my life used an AA of 1, so I am not going to miss those particular speed losses. As a rule I have found in like for like scenarios LW11 is both better and faster, and the reflection blur is simply orders of magnitude improved, but expensively slow to render. You gain some you lose some. Since as a rule I don't render things with a couple of samples per pixel, I haven't noticed a major slowdown.

At the end of the day I CAN NOT do anything about the speed of the renderer, but I CAN help with some tips and tricks which I use to speed up renders, such as the Schlicks approximation node. If you're interested in seeing complex scenarios where I use this, check out my 3D World tutorial http://www.3dworldmag.com/2012/08/23/build-surfaces-in-lightwave/ and you're welcome Hieron, I asked Michael to code that for me, to optimise my workflow, and a huge thanks to him for doing it for me! I already have ideas for version 2, but that's pushing my luck :D

Celshader
02-11-2013, 12:51 PM
Random tips for LW11.x:


Minimum Samples set the floor for antialiasing quality. No pixel will receive fewer samples than what you specify here. Try a Minimum Samples of 3.
Maximum Samples set a ceiling for quality if Adaptive Sampling is enabled. No pixel will exceed the number of samples than what you specify here. Try a Maximum Samples setting of 192.
Threshold specifies the maximum difference in brightness allowed between two pixels, when Adaptive Sampling is enabled. Try starting with a Threshold of 0.01.
Light Samples defines how many light samples will be taken for each Camera sample. You should start with a Light Samples value of 1. If the camera fires 100 rays for a given pixel, each of those rays will fire 1 Light Sample, totaling 100 light samples for that pixel.
Shading Samples defines how many shading samples (reflection/refraction blur) will be taken for each Camera sample. Start with a Shading Samples value of 1. If the camera takes 100 antialiasing samples for a given pixel, it will take 100 shading samples for that pixel (one shading sample for each camera sample).
Rays Per Evaluation are also the number of rays fired per camera sample. Consider starting with a non-interpolated RPE of 4. If the camera takes 100 antialiasing samples for a given pixel, it will also collect 400 RPE for that pixel (100 x 4 RPE per camera sample).


Differences between LightWave 10.1 and LightWave 11.x:


LightWave 10.1 doubles the number of camera rays per Adaptive Sampling pass, making it possible for LW10.1 to overshoot the quality Threshold. LightWave 11.x only fires one additional camera sample per pixel per Adaptive Sampling pass until each pixel meets the Threshold exactly (or until the ceiling defined by Maximum Samples gets hit). LightWave 11.5 can never overshoot the target Threshold quality.
LightWave 10.1 has no user-defined limit on the number of rays it can fire to clean up a stubborn pixel. LightWave 11.5 does (Maximum Samples).
LightWave 10.1 has samples per light, samples per surface and samples per node. LightWave 11.x consolidates those multiple quality settings to two scene settings (Light Samples and Shading Samples).

Spinland
02-11-2013, 01:05 PM
That's pure gold, Jen, thanks for sharing!

jasonwestmas
02-11-2013, 01:22 PM
wow, those settings seem oddly high Jen. what on earth are you rendering that you need a maximum of 192 samples?

Celshader
02-11-2013, 02:41 PM
wow, those settings seem oddly high Jen. what on earth are you rendering that you need a maximum of 192 samples?

Nothing here. I just want Mr Rid to successfully get a clean render, first. Then he can lower his Maximum Samples ceiling until he finds the right balance of speed and quality.

For production shots, I think most of my co-workers cap Maximum Samples at 64 or less. A co-worker with Max/VRay experience says he sets his min/max samples in VRay to 1/100, so he defaults to a Minimum Samples of 1 and a Maximum Samples of 100 in LightWave 11.x.

192 samples is still less than the thousands of samples (http://forums.newtek.com/showthread.php?133294-Adaptive-Sampling-Info-Old-And-New&p=1298054&viewfull=1#post1298054) that LightWave 10.1 would waste on a difficult pair of high-contrast pixels.

jasonwestmas
02-11-2013, 02:57 PM
Nice to know all that, thank you. :)

Mr Rid
02-12-2013, 05:51 PM
Its curious how everyone seems to be missing the point. I cant make this any simpler. If anyone cares to download my examples scene http://forums.newtek.com/attachment.php?attachmentid=111351&d=1360576606 in post #10, load in LW10 and hit F9. Then load in LW11 and try to get the same AA quality in the same or less time. If you succeed, please let us know exactly what AA settings you used, so we can all understand how the AA Max and Min values can optimize renders.

Given the explanations by RebelHill, and Jen, you should only have to adjust the Camera Min and Max sample values to get something akin to LW10's fixed AS sampling. The LW11 lighting and shading sample values seem to be more for a global adjustment of any lighting-related noise, or shader-related noise accordingly. For example, you dont have to increase examples for each of every area light, but instead may just raise the global Lighting samples. Ir if you have noisy reflection blurs, you can just increase the global Shading samples instead of having to adjust each surface sample value.


Err yes, it does matter when AS is involved since it will be using a ton of samples to try and fix the blurring. More changed than merely AA samples btw.

Strange, your above sentence is replying to my sentence "Once again, the last example I posted uses no reflection blur." There is no reflection blur involved, so how would a "ton of samples to try and fix the blurring" come into it? Explaining for the fifth time (beginning with my first post), removing all reflection blur from the example makes no difference in the fact that LW11 is only rendering slower for me than LW10 with comparable AA quality.



Hey man, just posting our results. Apparantly they differ to yours, doesn't make it useless and I did put in "afaik" and "imho"'s. Apparently I squandered my time here for sure.

Of course, each renderintg situation has its own variables. But using the example scene here, if you or anyone can get the same AA quality in less time by disabling AS (or with any AA settings), then please post scene or settings. If I disable AS and set Antialiasing to "3" I get the same render time, but slightly crunchier edges. If I set Antialiasing to "2" I get less render time, but distinctly crunchier edges. So again, I see no advantage to disabling AS.


More things changed between LW10 and 11 than just the AA, the reflection blur being the obvious thing in the first post but it won't be the only one.
While the other settings have no bearing on AA comparison (actually, they do but hey.. ), they do seem quite sloppy and I would think it normal for me to point them out. I suppose not since it doesn't support your AA case.

If you want to disable '2 Sided Area Lights,' and 'Noise Reduction,' and use any multithreading option you care to, thats fine as long as you use the same settings in both LW10 and LW11 for quality vs speed comparison. It doesnt matter.



I did not miss the point and did not make any comparison at all with LW10, I just added my input to things spotted in that scene. While all of those seemed good input, you do not seem to care as it is not in your exact concern.

I would think it would be the concern of every LW user if LW11's Min & Max settings are only serving to take longer to render, and to befuddle.


Some people here improved a ton in that scene, yet you do not seem to care nor appreciate.

Your suggestions may make the scene prettier or render faster which may be helpful to someone reading here, but it is irrelevant to the point of my example that you are responding to. Regardless of other settings, the point is about getting at least the same AA quality and render time in LW11 as in LW10. Otherwise, I see no advantage in having Min and Max sampling values.


I have no LW10 and LW10 will change more than just the actual AA settings used so it is not a fair comparison per se. I did not only change the camera settings and I'm pretty sure that the 9.9 seconds end result in LW11.5 is quite nice on this simple quad core 2600k. Heck, the GI pass alone took 8 seconds in that LW10 scene when first opened.

Your render shows 16 seconds LW11 GI time so I suppose your CPU is 2x slower than mine (if the difference is less, the improvements shows are greater) so your 30 seconds LW10 time would be 15 sec here. So the 17 seconds I posted first is not THAT bad (and it has little noise like yours) and the 9.9 seconds is a vast improvement (it could be done in LW10 as well)..

So dunno man, but you seem to have your opinion set no matter what. Imho, I would be amazed if LW11.5 is actually really slower to render per sample. And you first comparison with AS and blurry reflections shows me that if you did spend tons of hours, you missed the point. What is your cpu at least? Cinebench points are very representative and that way it saves me installing LW10 to figure out a comparison with your rendertime.

If you care to post the example scene http://forums.newtek.com/attachment.php?attachmentid=111351&d=1360576606 with your settings, I will look forward to comparing render time and AA quality in LW10 to LW11. As it stands, I dont see the benefit of having the added Min and Max sampling values, that I cannot at least disable and switch back to a nice 'auto/fixed' sampling rate that worked more efficiently before in LW10.

Mr Rid
02-12-2013, 06:06 PM
... I just want Mr Rid to successfully get a clean render, first. Then he can lower his Maximum Samples ceiling until he finds the right balance of speed and quality.



Thanks Jen, as always. But a balance of speed and quality that are comparable to LW10 is what no one seems to be able to produce. Using your suggestion of Min-3, and Max of 64, the LW11 render takes 24.6s, and LW10 is 18s. When I bring the Min or max values down to even the render time with LW10, then the AA quality gets worse than LW10. So what is the advantage of the Min and max values in LW11?

I propose that the Camera Min and Max sample values could be an OPTION that can be disabled. Then LW could default to the old 'fixed' sampling (would that be a fair term?) if you dont care to spend hours trying to figure out how the Min Max thing works (or if they arent really helping). Am reminded of the options of 'Fixed' or 'Adapative' Min and Max substeps in RealFlow. Or the Min and Max samples should at least default to a generally practical setting, comparable to what users are used to from LW10.

My frustration is that the first time I run LW11, I load a scene, hit F9, and the AA looks awful. Naturally I compare to how the scene looked fine in LW10 (who wouldnt?). Days later, after lengthy tutorials, reading, and forum queries, I still cant get as good a render in the same time.

Rendering is the most fundamental function to LW. I dont know why so often I seem to be the only one to notice these things. I am about to start a project, and I dont care to use LW11 if it is going to take significantly longer to render.

Mr Rid
02-12-2013, 07:17 PM
-forcing threads to 4, Is that better than auto mode?

For anyone trying to compare, when I load the LW10 saved scene into LW11, LW11 is automatically changing the Multithreading to 'Auto,' and setting the Lighting and Shading samples to 8.

dwburman
02-12-2013, 08:53 PM
My main LW computers are rendering now, so I won't be making a comparison.

Maybe this scene isn't one that benefits from the changes in the renderer. I don't know why it's taking longer for comparable quality.

I made an 8-minute quick start video for unified sampling for those who don't want to invest the time in Rebel Hill's video or Jen's LALWUG presentation. Those videos are definitely worth watching, though. http://youtu.be/_vHNd17aSiw

Of course, that doesn't help the discussion at hand. :)

Hieron
02-13-2013, 01:06 AM
For anyone trying to compare, when I load the LW10 saved scene into LW11, LW11 is automatically changing the Multithreading to 'Auto,' and setting the Lighting and Shading samples to 8.

your opening images show 8 vs 16 threads in the render info, a scene opened at 4 here. Just saying.

a simpler scene would prove your point better, so mistakes like the refl blur from your post 1 can be ruled out. There are other possible issues in nodes.

regarding the rest, nvm. no time. Will perhaps try it myself, if time allows. You were not the first one to suggest this slowdown btw. You managed to reply a book but forgot to post your cpu btw? 9.9 s on a 2600k from this thread info. But if you want to drop LW that is fine with me too...

dwburman
02-13-2013, 10:43 AM
I did the test. I think I got fairly close to the LW10.1 render by setting the light samples to 3 and setting the min canera samples to 6 and turning off AS entirely. I also set the sampling pattern to Low Discrepency, though I don't know if it made a difference or not.


For some reason, having AS on - even with a threshold of .005 and a high max sample, the edges of the ball ended up jaggy looking.

Celshader
02-13-2013, 11:07 AM
Thanks Jen, as always. But a balance of speed and quality that are comparable to LW10 is what no one seems to be able to produce. Using your suggestion of Min-3, and Max of 64, the LW11 render takes 24.6s, and LW10 is 18s. When I bring the Min or max values down to even the render time with LW10, then the AA quality gets worse than LW10. So what is the advantage of the Min and max values in LW11?

Oh -- I forgot to mention one more thing -- consider using the Fixed or Classic Sampling Patterns instead of Low-Discrepancy in LW11. In LW11.0.x I noticed that Fixed/Classic got fussier about cleaning up noise than Low-Discrepancy. Fixed/Classic would take longer, but they would look cleaner.

I do not know if this changed for 11.5, but I start with Fixed juuuuuuust in case.

Mr Rid
02-13-2013, 10:25 PM
Am trying various scenes...

This one has worse GI leaks in LW 11
111518
111519

Cook Torrance spec difference
111520


I did the test. I think I got fairly close to the LW10.1 render by setting the light samples to 3 and setting the min canera samples to 6 and turning off AS entirely. I also set the sampling pattern to Low Discrepency, though I don't know if it made a difference or not.

For some reason, having AS on - even with a threshold of .005 and a high max sample, the edges of the ball ended up jaggy looking.

It looks like your LW11 example shows Min Sample 5, but I see slightly crunchier edges, looking at the bottom edge of the red wall, and the ball edge. I get slightly better AA by using AS, but in the same render time.

After applying your settings, I get the same result, and it takes one second longer than in LW10 to get a comparable render. Not a big deal but I still see no speed advantage to using the Min/Max samples.

geo_n
02-13-2013, 11:33 PM
LightWave 10.1 has samples per light, samples per surface and samples per node. LightWave 11.x consolidates those multiple quality settings to two scene settings (Light Samples and Shading Samples).
[/LIST]


we can all understand how the AA Max and Min values can optimize renders.



Afaik its not the purpose of unified sampling and lw 11 new AA system to speed up renders as its first priority.
Its to speed up getting to a good render time/quality ratio without needing to optimize the scene heavily that only technical rendermonkeys will have time for. We have a vray rendermonkey optimize our scenes but if vray had the locked lw approach of simplifying global settings he would be less useful. Both have advantages ofcourse. Unified sampling and simpler global settings make the life of average joe easier. More settings to optimize scene will benefit render pros who need to squeeze render time as much as possible.
Will try the scene but I won't be surprised if lw 9.6 might even be faster.

Hieron
02-13-2013, 11:58 PM
I did the test. I think I got fairly close to the LW10.1 render by setting the light samples to 3 and setting the min canera samples to 6 and turning off AS entirely. I also set the sampling pattern to Low Discrepency, though I don't know if it made a difference or not.


For some reason, having AS on - even with a threshold of .005 and a high max sample, the edges of the ball ended up jaggy looking.


"reduce noise" does that, see my earlier post. Do not use it, it is crap. (imho :))

it should be a focus of this thread :)

mike_stening
02-14-2013, 09:29 AM
i think on such a small render as this the differences are moot. The tests i have just been doing have given a better quality render overall in LW11.5 to LW10 but take a little extra longer. i think if you started pumping the res up/quality there may be less of a difference time wise/quality wise between the 2 versions
I have been turning reduce shading noise off as you essentially loose detail, think of it as a post render effect like PS surface blur.
Other things to look at are, i find Monte Carlo quicker in general than final gather and the tests i did were FG LW10.1 compared to MC LW11.5.

OK did a few tests

Increasing the light samples in LW11.5 to 3 to match the LW10 scene actually sped the render up by about 6 seconds.
I'm still getting my head around this but personally i feel its easy to get a high quality render at a quicker time in LW11 than it is to try and bring LW 11 down to a fast OK quality render in LW10.

111532111532111533111534111535

to add my renders are always over 1.5hrs as they are for print so i have been playing alot with settings. If you are only using luminous poly's/HDRI then you can turn light samples down to 1 but your shading sample will have to come up to compensate. Lowering either shading/light samples to 1 when you are using them only seems to lead to longer render times as AS try's to reduce noise. Its a balance between Light/Shading samples and MIN/MAX AA + AS and over sampling though I'm not a fan of that as it again seems to ad time and smooth things too much but then oversampling for me is personal preference.

Mr Rid
02-19-2013, 12:04 AM
Where did the individual 'Light Quality' setting disappear to in 11.5?

bobakabob
02-19-2013, 12:40 AM
There are now unified settings for lights and shader quality, set to 1 by default (3 in LW 11). In the LW 11 addendum there's a really good explanation of the change. It puts everything in one place. I'm sure RebelHill explores the changes in his rendering video.

Mr Rid
02-19-2013, 01:09 AM
If every light is stuck with the same global Light Samples value, how does this offer 'more control' for optimizing? I am used to using multiple area lights at different sizes, where a smaller light may not need the same shadow quality as a larger one (or at varied distances). Why not likewise eliminate all individual surface sample values, and adjust only the global Shading Samples for all surfaces?

Seems like a step in the opposite direction from adding the 'user-defined' management that Jen mentioned.

mike_stening
02-19-2013, 01:58 AM
If every light is stuck with the same global Light Samples value, how does this offer 'more control' for optimizing? I am used to using multiple area lights at different sizes, where a smaller light may not need the same shadow quality as a larger one (or at varied distances). Why not likewise eliminate all individual surface sample values, and adjust only the global Shading Samples for all surfaces?

Seems like a step in the opposite direction from adding the 'user-defined' management that Jen mentioned.

surface samples are set by shading samples, again unified in render globals. so no more individual shading samples to be set in nodes.
all lights having the same sample settings , to me at least, means that you my not be trying to fight a lights poor sample setting with AA/AS/OS, or having to hunt down which sample settings for which LIGHT/NODE etc are causing the problems. Also i think it stream lines the renderer by it not having to calculate lots of different sample settings and combining that with ray samples and raydiosity samples.

generally my renders now are much more defined and render quicker than similar settings in LW10

OFF
02-19-2013, 03:26 AM
Good compare session, Mike!

dwburman
02-19-2013, 01:06 PM
BTW, the DP_Light set has individual light samples as an option.

Mr Rid
02-19-2013, 08:21 PM
So, NT added the Min & Max sample values in order to allow more AA control, but consequently more values to monkey with. But then they eliminated individual light and surface sample values in order to have less values to monkey with but consequently less control. Makes complete sense.

Tobian
02-20-2013, 02:07 AM
well it does in some ways make a lot of sense: put all sample settings in one place, instead of every light, shader, surface, material etc. It's slightly counterintuitive ONLY for those who are too used to the old system, because some are unwilling to change. Likewise I still see people using Classic cam, not because it's 'better', but because they were unwilling to change, to spend time to figure out these new-fangled aa settings :-). Ideally 'presets' would be good, to help speed the learning curve, but it's not always useful anyway, since different types of scene need very different handling. Ideally I'd also like a sampling 'multiplier' for surfaces, shaders and lights, with nodal control, so we can fine-tune some situations when a light or shader is being tricky (like very soft shadows, or glancing reflections, which are blury), and hopefully this will come. :-)

Mr Rid
03-26-2013, 04:33 PM
I see no sense whatsoever in the lumping of all light and surface samples under one global value just to make users less confused, while sacrificing advanced control over optimization. Meanwhile, lets convolute by increasing the number of AA values in order to have greater optimization. :question:

A global sample value is fine as an additional control, but not as a replacement for precise optimization of each light and surface. It is more intuitive to lump all sample values into one place, but not at the loss of advanced control. This isn't like Daz aiming only at hooking novices.

I am currently working on a typical example scene of why a single light sample value stinks, where I have six area lights and one dome light. A sample value of 6 would work fine except with three of the area lights that introduce much noise since they are scaled up quite a bit while being close to some geometry. Light samples must be cranked to 30 to reduce noise down to 'acceptable,' but I am instead balancing a value of 20 with the hiding of remaining noise under post-applied grain. But those three problem lights switch off about a third of the way thru total frames, after which I only need a sample value of 6 for remaining frames- going from 20 to 6 drops render time by half here. It would help if I could envelope the sample value thru the lighting change, but the real problem is that I only need increased samples for a few lights, not all of them. I do not see how anyone could view this situation as being more efficient.

And btw, I continued using the Classic cam for a long time because it just plain rendered faster than the Perspective in most situations. 'The Perspective Cam is faster!' was a myth in 9.6. I had too many discussions about this with co-workers and on the forum, and each time someone claimed they had a scene where the Perspective cam rendered faster, upon close inspection they were really just unknowingly using lower AA quality, since AA works differently in each cam. After countless tests, reflection blur was the only situation we could find where Perspective rendered faster. Now, the Perspective cam is faster. But this is part of the same optimization confusion that most users dont seem to test closely enough.

RebelHill
03-26-2013, 06:29 PM
Light samples must be cranked to 30 to reduce noise down to 'acceptable,'
No wonder you're having rendertime issue... there's no need to push light samples up that high (even 6 is overkill in a lot of situations). Keep base samples low and exploit the adaptive sampling to do the cleanup...

http://www.youtube.com/watch?v=UgNB9tZWUmM

Sanchon
03-27-2013, 05:20 AM
Keep base samples low and exploit the adaptive sampling to do the cleanup...

Yes... This works well for simple, basic scenes but where I have scene with a lot of thin instances like grass, plants, lot of geometry - cranking up adaptive sampling will increase rendering times to unacceptable values with not as much quality improving which isn't worth it. I noticed that sometimes is better to turn off adaptive sampling at all and do rendering with high minimum samples values only. Much better render times and better overall quality - especially in dark areas where you must use very low AS to get rid of the noise. The best option would be able to indicate areas or objects where more samples should be directed or more intelligent adaptive sampling system to get rid noise and aliasing in areas where exactly they are. Or maybe separate shading and AA system for instances.

Tobian
03-27-2013, 08:05 AM
Mr. Rid: since I render almost exclusively WITH reflection blur I can see the confusion there. It's been demonstrably faster for me. If you're working mostly with oldfi and or diffuse lighting only setups then yes, it may well be slower.

I do however agree we need 2 more settings within the unified sampling:

1) customisable multipliers for surfaces/lights/materials/shaders. There are times you need a certain light or shader to simply have more samples than it's neighbours. I do agree there should be this option for power users trying to get render times down. I'd also like granular control over the multiplier for some scenarios: surfaces with reflection blur caught on glancing angles smear the reflections quite a lot. Increasing the global samples/as to get these to be smooth means firing pointless samples at all surfaces which probably don't need it. no it's no use saying that you can fire a crap load of AS samples at the scene, because often you have to raise the min samples to get the grain to the point where it will actually go away... ever...

2) Importance sampling. If a surface is 90% reflective and 10% diffuse, then 90% of the samples should be reflection samples and 10% should be shadow rays/diffuse samples. Likewise if a surface is only 2% reflective it really doesn't need as many samples for reflection as one that's 98% reflective. Especially in that first min samples pass this could dramatically speed up some calculations, especially since adaptive sampling can clean up any of the resultant noise anyway.

RebelHill
03-27-2013, 10:41 AM
Yes... This works well for simple, basic scenes but where I have scene with a lot of thin instances like grass, plants, lot of geometry - cranking up adaptive sampling will increase rendering times to unacceptable values with not as much quality improving which isn't worth it.

Its certainly true that if your scene contains lots of small detail that going low on base samples, high on AS can increase rendertime... however there's almost always a balance point, and since AS will direct additional samples only where needed, you can almost always get a faster render using AS. The thing to do, as I show in the vid, is to do a few test renders (limited region of a specific area if u dont wanna wait for full frames)... where u trade off base samples for AS passes till you find the sweet spot. Generally, start with LOW base samples and high AA, then go through, increasing base, and reducing AA (in a multiplicative manner so to as to get ~the same number of total samples used)... you'll note render times start to drop, then at some point, they'll start to go up again, and there u have it.

gerardstrada
04-09-2013, 09:04 PM
I am way behind in LW and just now trying v11.5 AA. I watched the tutorials and read, trying hard to resist my brain glazing over at mathy explanation and four more values to try to figure and balance.

But using the attached test scene, I cannot begin to get the same quality and render time in LW11 that I get in LW10. I understand that leaving the Lighting, Shading, Min, & Max samples at "1" is equal to Antialiasing "1" in LW10(?). So by only trying various Min & Max sample values to get the same quality, I only wind up with longer render times in v11.5.

What LW11.5 AA values would get the same quality and render time that I get in LW10?
111294

Since I think v11.5 handles reflection blur differently, turning all reflection blurs to "0" did not make any difference.
Haven't had LW 10 installed here, but have tried the scene with LW 9.6 and LW 11.5.

At first glance v9.6 looks blazingly faster than v11.5 for getting the same results, but it seems the cause of discord is Noise Reduction. It's producing different results in LW 11.5 (and v11) than in LW 9.6 (and probably that's what's happening in v10 too). In recent versions of LW, Noise Reduction produces more AA artifacts than in previous versions - I don't really use it frankly (prefer DP Denoiser filter for such cases) but it would be good to have it working properly since it's one button solution.

So for these tests I disabled Noise Reduction. Then, things get in the same terms. At least comparing LW 11.5 with LW 9.6 we are able to get pretty much the same render time and render quality by setting the lights quality to 16 and Max samples to 6 (1 for Min samples).

http://s7.postimg.org/o0cptp2nf/LW9_6.jpg http://s21.postimg.org/s4wtx887b/LW11_5.jpg

In this case haven't touched anything but people that work in linear light might want enable the output CS too, and save in linear as commented here (http://forums.newtek.com/showthread.php?119485-Why-is-the-quot-Adaptive-Sampling-quot-pass-so-slow&p=1315224&viewfull=1#post1315224).

Agree about the Shading/Light quality options. User should be able to set up these options globally or locally per surface/light.



Gerardo

p.d. Attaching the scenes in case things were different in LW 10.

Ma3rk
04-10-2013, 05:31 PM
Coming in late to this thread but as I haven't seen it mentioned I thought I'd pass on this link of a presentation that Jennifer H. made a few months back at the LA LW User group that addresses several of the questions raised here.

http://www.youtube.com/watch?v=8JdONpR6dR8&list=UU6NL3EUX90auYzSB_VTUkYA

sadkkf
04-20-2013, 12:01 PM
Also jumping in a little late, but by doing that I was able to take advantage of some of these posts and optimize my renders.

Still, I'm getting what I think are unacceptable jaggies on some text I'm rendering.

113753

I've been tinkering with the settings quite a bit and nothing really seems to help this. Starting with min/max samples of 32/50 with classic recon and fixed sampling pattern I've used .004 for the AS. This looks exactly the same as the highest settings of 256/2000 and .001 AS. Am I missing something?

My shading and light samples are both set to 1 since this is just text I'm worried about right now.

Any thoughts?

ivanze
04-20-2013, 02:26 PM
Just for testing, sadkkf, try rendering that same text with Classic Camera and AA PLD-5 or better and please post if it looks better.

sadkkf
04-20-2013, 04:04 PM
Um...yeah. Much better!

Here's PLD21, but PLD5 didn't look noticeably different. And seriously better render times, too.

113760

Thank you!

sadkkf
04-21-2013, 07:53 AM
In all the videos and examples I've seen demonstrating the new AA methods the perspective camera was used. For me, I need to use the classic camera to achieve the same results, even though they don't look much better posted here.

There was time I would have asked why.