Machines rendering differently?

[RErender]

What if you set the dmc sampler adaptive amount to zero?

[tarn]

What if you set the dmc sampler adaptive amount to zero?

OK, I mis-spoke above. I was doing so many tests that I lost track. When I set AA to Fixed with 1 sample, the problem actually did go away. So then I tried AA Fixed 4 and the differences returned. I also tried AA Fixed 4 with Adaptive Amount at zero, and I also had differences between the same frame rendered twice on the same machine.

So, to summarize, the only thing so far that has cause the problem to go away entirely is setting AA to Fixed with 1 sample (essentially the same as no AA at all).

What I don't understand is, the manual says that DMC should produce the same result each time:
"deterministic Monte Carlo uses a pre-defined set of samples (possibly optimized to reduce the noise), which allows re-rendering an image to always produce the exact same result"

http://www.spot3d.com/vray/help/150S...s_dmc.htm#refs

Also, under "Time independent", the manual states:
"Note that re-rendering the same frame will produce the same result in both cases."

So I am still a little lost.

Tim

[tarn]

Here's an example of what I was talking about above:

http://members.lycos.co.uk/tarn878/D...dapt-Amt-0.gif

This is a cropped section, showing a piece of the same frame of animation rendered twice on the same machine. I have combined these into an animated GIF to show the difference. I also enlarged this 300% to make it easier to see.

This was rendered with AA Fixed 4, no AA Filter, and the DMC Adapt Amount at zero.

In contrast, if I set the AA to Fixed 1, then the two frames are identical.

Tim

[RErender]

Strange, a few more tests I can think of are turning off filtering, both in AA and bitmaps. Trying it without GI, saved or otherwise and then looking at the hardware not sure if math errors could cause something like that, I guess the boxes are not OC'd? In researching I found this http://support.microsoft.com/?kbid=936357

[tarn]

Thanks for all the suggestions. I really appreciate it.

I downloaded the Windows update you mentioned, but it looks like it was included in SP3 for XP 32-Bit. When I tried to install the update, a message said that the update was an older version of what is already installed.

I have also tried a test with all the suggested items turned off:
1. GI Off
2. DMC Adaptive Amount set to zero
3. Bitmap Filtering Off
4. AA Filtering Off
5. But still with AA on set to Fixed 4

I still get the same results: the same image rendered twice on the same computer does not come out identical. If I set AA to Fixed 1 (no AA at all), the problem goes away.

Tim

I did notice one very interesting thing: some buckets render the same, and some render differently! I drew a grid on the rendered image at 64 pixel intervals (my bucket size), and some buckets have variation, while others don't. To me it appears as if the DMC generator is out of sync in different threads. This particular computer is Core2 Quad, so it renders with 4 buckets, but other computers we have which are only Core2 Duo (two buckets) are doing the same thing as well.

[tarn]

More testing!

Now I have found that if I use Adaptive Subdivision AA, the problem goes away.

So to summarize:

1) Adaptive Subdivision AA with min 0 max 3 renders frames identically.
2) AA Fixed 1 also works (but of course I need AA with more than one sample)
3) AA Fixed 4 causes a potential difference between the same frame rendered on the same computer.
4) Adaptive DMC also causes this problem.

In the above tests, the AA filtering didn't seem to make a difference, nor did the GI settings.

In the past, I have typically used Adaptive DMC, but I guess I will stick with Adaptive Subdivision AA for now.

Tim

[duke2]

It's fantastic that you are doing these tests. I get this too but I always thought it was just part of it/indication of low sampling.