I'd like to compare against others to see if my AO render times are reasonable. I remember seeing someone post 20-30 minutes for a simple scene but I'm getting 1.5-2 hours.

Like I said, I'm just looking for an informal comparison. I want to make sure I'm not paying a "Stupid Tax" by having the wrong settings on.

Most of the settings are visible in the screenshot.

CPU: 2GHz G5
AM: Beta 1
Frame Size: 640X480
Multipass Off
Elasped time: 1h 26m

One shadow-only light (a sun)

Thanks,
Rhett

EDIT: AO Toys project in post #14 http://www.hash.com/forums/index.php?showt...opid=175629

Only way to get a true comparison is to do a project that everyone has.

Usually that means something that came with the A:M CD.

Hey Jandals,

That seems very high. I got my Demon render time down to 1hr and that was with Hair Reflections, shadows and texture maps. I think it was at 9 or 16 multipasses as well.

Don't be afraid to try multipass with AO.. it breaks up the AO sampling over the passes.. so its not 16 times the sampling rate you choose but more like.. the sampling rate you choose divided by the number of passes.

The 1hr render time on the demon was with 1ray shadows and 20 or 30 sampling. The computer was a 2.0ghz AMD with 1 gig of ram.

Heres the pic for it...

http://www.hash.com/forums/index.php?act=A...e=post&id=15420

C

Jody:
Yeah, I know. But I'm not looking for a true comparison, just something to see if I'm in the ballpark.

Maybe I'll render Toys tonight if nobody else responds.

Colin:
Thanks! I'll try a multipass render but it sounds like that'll take just as long. Except it'll save me 7 minutes of anitaliasing...
I'll keep playing around because, apparently, something will lead to shorter render times (3 min per frame):
http://www.hash.com/forums/index.php?showtopic=20705&st=33

Cheers,
Rhett

AO does not play well witn non-multipass.

Switching to a 4-pass multipass cut the render time by 15 minutes to about 1h 10 min). Dropping the sampling to 10% shortened the time to 20 minutes.

I guess that's nothing unusual.

the time it takes to render is more effected by the patch count and surfface properties of a model. The AO settings should be maxed out in the chorography area, and the AO should be set to <=25% in the render screen (anything more is just a waste of time)

How are the AO settings in the choreography properties different from those in the render panel?

Patch count have little impact on AO render time. Render resolution have a direct impact on render time.

So it must be the surface properties or AO quality settings then, yes?


The choreography settings allow you to change the intensity and darkness of the ambient occlusion; the render options allows you to set the quality.

With such high ambiance intensity, you should normally use 100% ambiance occlusion.

ah-HA! I did not know these things.

Thanks guys. It's much clearer now. I'll try different setting later.

Rhett

Those render time seem too large for what you are trying to render. Here is a Toys render I setup: I rendered frame 98 at VGA resolution (640X480). I removed all of the decals on all of the models, I removed all of the materials, I set the main diffuse color for every model to white, I removed all of the ambiance intensities, and I deleted all of the images. After I did that, I saved the file as ToysWhite. After that I turned on AO in the choreography, set to global color, and set the Occlusion to 100%. I then went into the rendering options tab and turned on multipass, set to 16 passes and also turned on soften, and then I made sure that everything else was turned off except for Ambiance Occlusion which was set to 30% sampling.

The total time was 17:33.

[attachmentid=16530]

Thanks for that Matt

I did what you said but I aborted the render when pass 1 of 16 got up to 6.5 minutes. 6.5 min*16 passes ? 104 minutes. That's on a 2 GHz Mac G5

Here's my ToysWhite project file. Maybe someone will point out my mistake.

Any time estimates given before the first frame (or sub-frame for multipass) is complete should be treated as suspect.

My previous AO multipass renders took almost the same time for each pass and I didn't think this would be an exception. I'll let it go while I'm at the store and see what I get.

Edit: 42 minutes and counting for what took Matt Bradbury 17.5 minutes. I must be setting something wrong.

Okay, let's try something simple. Just open up Toys, and render frame 98 to 99 with these properties: Resolution VGA, MutliPass 16 with Soften, Shadows, Reflections with 2 Levels and Soften, and Motion Blur with 50%. Make sure to stop the render after frame 98 has completed because we do not need frame 99. The render took 2:56 for me and it looks like this.

[attachmentid=16541]

OK, that may explain things. My render took 12 minutes, although it looks fairly different from yours. The details on the blocks and on the trim are much more pronounced.

EDIT: Opening Toys and clicking the render button (all default settings) gives me 3:21/3:11 for frame 98/99.

Man, 2 GHz ain't what it used to be...

I suppose I should send my result to the benchmarks page. I wonder if that's still going...

Thanks for the test Matt.

Rhett

but our computer only has a 64bit 2.4 ghz processor, so why are the render times so very different. Could it be the operating system or something?

by the way, your bump maps look really screwy on that render you did and matt, you have these wierd white bands that go around the crayons

Rhett

This is an interesting thread and may uncover some interesting things but you need to ensure when making comparisons that everyone is starting with the same settings. Just opening the project does not reset everything. Try rendering Toys again but make certain that you setup everything as shown here, or provide your own settings and I will match them. I would be interested in seeing what everyone's results are.

Here are the settings that I used:
[attachmentid=16601]

All frames were rendered to TGA images, 640x480 with 5 passes. Here are the results which I have converted to JPEGs with PS compression set to 7 to save space. If you want the actual targas for comparison I will post them. Here are the jpegs for frame 99:

A:M 11.1i: [attachmentid=16602]
A:M 12t: [attachmentid=16603]
A:M 13 beta: [attachmentid=16604]

And here are the render times: [attachmentid=16605]

Interesting results. The first thing that is apparent is that you and I are rendering different frames yet if you check my render settings you will see that I am setup to render frames 98 and 99 as you specified. I don't know why A:M12t renders completely different frames and 11.1 and 13 display incorrect information for 'Current frame'. This could mess you up if you started a project in 11 or 12 and moved it up a version.

Machine specifications: 3.2 GHz Pentium 4, 1 GB RAM, Win XP Pro.

Okay, Rhett. Lets see what you get.

That is odd. Can you check the FPS settings? It might just be coincidence that eight thirtieths of a second is equal to 6.4 twenty-fourths of a second. (OK, so similar arithmetic doesn't work with 78 and 98.)

Thanks for setting the standard Paul,

I matched your settings for this render and got 3:27/3:20. 170% longer than your Pentium which is 60% faster than my G5.
[attachmentid=16607]
I still have something weird going on with the bump maps in my image.

Once or twice, Toys loaded as a 24 FPS project and frame 98 was different (probably the same mystery frame you got). but it was just a matter of setting the project frame rate...

Anyway, thanks again. It's good to see the comparison even if I'm not bragging about the result

Rhett

I loaded the Toys project in each version of A:M and then set everything to ensure that all settings matched. You can see in the bottom of the Render time panels that in all versions the frames are 98 and 99, the ones set in the render options. The frame numbers at the top, (frames 78 and 79), are not what I set up. This is an A:M glitch.

Your times seem about right compared to mine. What version are you using, Rhett? I will do a render with AO in that version and provide the settings.

Also note that the image you rendered is different from the ones A:M chose to render for me. ???

Remember that bump maps, normal maps, and hieght maps have been adjusted in V13. 100% now means 1 cm.

---Added---
Paul, I used your settings and nothing was out of the ordinary. Here's the render.

[attachmentid=16608]

Perhaps it's just the FSB that decreases render times.

Okay. I have updated to 12 v+ and here are the results:

[attachmentid=16609]

And the times:
[attachmentid=16610]

It seems that A:M 13 is the fastest of the three!
A:M 11.1i and 13beta both give the incorrect 'Current Frame' information or rather I think they render those frames but display at the bottom that they have actually rendered frames 98 and 99. Perhaps that was fixed in the recent beta release though.

Matt, looking at your render time made me curious. I'm sure that I used to get render times of around 45 seconds for frame 99 so I went looking for my old Toys tests. I couldn't find them but I did find Steve Sappington's benchmark test results: Toys tests. There is no mention of which version of A:M these were tested with, and I expect it was various versions, so the info is kind of irrelevant but it does secure my memory of faster render times in a previous life. Of course settings make a difference too and I may have rendered a single pass or no multipass. I tested this and sure enough here are the results:

[attachmentid=16616]

It must be your 64bit CPU that is giving you those great results, Matt.

I opened the toys project in V13 Beta 1 and I too got the strange white banding on the crayons. Something to do with the decals. Also, it seems the front decal on the cardboard ship is off center. ??? Decal shift?

Anyway, I rendered with the settings posted by Paul Forewood (in post #20) and got a render time of 1 minute 12 seconds. Didn't have any problem actually rendering frame 98.

Here is my render (Converted TGA to JPG just as Paul did).

[attachmentid=16619]

Here are my machine specs...

Pentium 4 (3.4 Ghz, Hyperthreaded, Socket 370, NOT mobile verison), 1 Gig of RAM, mobile SATA hard drives (80GB, 5400 RPM) RAID striped pair. GeForce FX 6800Go module with 256MB RAM. All of this is running Windows XP Pro (SP2 + all critical/security updates)

Hope that helps.

Thanks Jody,

I started collecting v13 times in case anyone else wants to share and compare...

http://homepage.mac.com/mordredc/hash/bench/AMMark.xml

Rhett

Mr. Bravo, now that you have a better grasp for AO, try doing the Space ship render again.

Heh, forgot that was there Probably time to take that down...

Anyway, I turned Ambience Occlusion up to 100% but turned the sampling/quality setting down to 10%. Now it's a 20 minute render instead of 2 hours.

I hate paying the Stupid Tax

Just wanted to clear up a misconception: you don't want to stop frame 99. Frame 98 will take more time because it needs to load a bunch of images from disk. Frame 99, after all the images are loaded, will be the true render time.

I knew about the loading time, but I felt that the time to load off of the disk added only an insignificant amount of time to the render; for me, there was no time added between the two frames.

It's good to hear that your times have decreased.

I think you are a tad bit confused on how AO is really utilized. There are four settings for AO:

Ambience Color: The color you want your environment to be.
Ambience Intensity: Is just like the light property of "Intensity." Does nothing to affect the render time.
Ambience Occlusion: Is basically the same as the light property "Cast Shadow>Darkness." Does nothing to affect the render time.
Occlusion Sampling: Found in the render settings, it sets the quality of the Ambience Occlusion. Affects the render time.

For your particular project and renders i would suggest you use these settings:
Ambience Color: w/e
Ambience Intensity: 100% (100% intensity light, can go higher if desired)
Ambience Occlusion: 100% (100% darkness shadows)
Occlusion Sampling: 25% (25% quality)

Alright, I hope that explains a few things.

Not for me, with my puny 2X CD-rom drive.