run through of my working method for vray 2.4 sampling in arch vis scenes.

[tricky]

So to get back on topic, the scene with the original settings (IM+LC and lots of subdivs for the dome light and maybe elsewhere) does indeed render for about 2h 10 min on the particular machine I am using, which is much longer than what I got with BF+LC. I have not profiled yet what causes the longer render time, as I want to make a few tests with V-Ray 2.4 first.

Best regards,
Vlado

In a situation like this (room with a small window) wouldn't a vray plane light set up as a portal be a quicker solution and remove the dome light altogether?

[vlado]

In a situation like this (room with a small window) wouldn't a vray plane light set up as a portal be a quicker solution and remove the dome light altogether?

Maybe yes, but then again maybe no. It depends on the HDRI itself and whether it produces sharp shadows. It has to be tested for each case.

Best regards,
Vlado

[trixian]

As in: if sharp shadow then portal = bad, or the other way around?

[glorybound]

The scene is being described as a room with a small window.... but, to me, that's a huge opening. Most of my interiors have much smaller openings.

[kosso_olli]

Vlado, could you please post a small screenshot showing your settings for these images?

[joconnell]

The scene is being described as a room with a small window.... but, to me, that's a huge opening. Most of my interiors have much smaller openings.

I think what kills it is the walls around the room too - it's really cutting down the angle of light rays that can enter the window!

[Bertjenkins]

I think the reason the tutorial is so great is that you waited all that time (must have taken ages!) purely to illustrate, and render out, what a 'bad' method is. This means as a viewer you can compare like for like; The result is a great reference that you can go back to, and compare techniques/approaches. Thanks for making it!

Of course, I for one would really like to know the theory behind the Vlado's jump from 2h 10 mins (IM/LC) to 20 -ish mins (BF+LC). I always give up on using Brute force, although i work on animation not stills. I've never once successfully used it without a noisy, slower, result. I must be doing something wrong- and i'd love to know what!

[joconnell]

That got me very curious too - when I can get on to the same render node that I used to do all the tests, I'm going to do a BF render, but then also add in sub pixel mapping and reduction of the number of bounces on the really glossy materials too on the irmap renders and see how much it'll shave off - I normally do that on a lot of my scenes and while you lose a tiny bit of brightness, a glossy reflection of a glossy reflection is so mushy at that stage, you're not going to see any detail loss in the reflections and you're going to cut down your ray count heavily. Likely to happen around monday!

[Bertjenkins]

I'd love to know the results of those tests. I tend to use sub pixel mapping as well most of the time to reduce hotspots...

[kosso_olli]

Vlado, could you please post a screenshot of your settings from 3.0? I am switching next week, and your settings seem to be a good starting point.

[fewlo]

Hi,

First off I want to say a massive thank you to joconnell, what a fantastic run through of your theory that we first talked about in the 'understanding dmc' thread, it is a powerful technique for rectifying pesky noise. However I still can't seem to get this theory working at production level.

I've been running through the steps again on a fresh interior scene animation and still hit the same problems of unacceptable render times; to get a nice smooth raw lighting pass I'm up to 512 subdivs on my dome light - a Peter Guthrie HDRI - which sends my times through the roof and isn't workable on a live job. I then revert back to my old ways and 'store with irradiance map' on the dome light which gives low render times but blotchy GI. Upping imap settings to 100/35 cleans it up but still not ideal.

So I have perfect quality render passes but 13hrs per frame at 1024 x 576 (!!!) or decent passes and just about acceptable blotchy gi with 14 mins per frame. Where's the happy medium?? Does it even exist? I can live with 'store with' on my dome light if I can get the render looking sharp but I feel the imap settings are making it look 'mucky'.

Any thoughts on what to test next?

[voltron7]

Excellent video full of great tips and info! I also love that you did not just glaze over the problems and how you solved them step by step.
It was a nice change since many videos seem to "over-summarize."

Anyway, thanks so much for making it and for free, too.
Best regards and thanks!

[joconnell]

Hi,

First off I want to say a massive thank you to joconnell, what a fantastic run through of your theory that we first talked about in the 'understanding dmc' thread, it is a powerful technique for rectifying pesky noise. However I still can't seem to get this theory working at production level.

I've been running through the steps again on a fresh interior scene animation and still hit the same problems of unacceptable render times; to get a nice smooth raw lighting pass I'm up to 512 subdivs on my dome light - a Peter Guthrie HDRI - which sends my times through the roof and isn't workable on a live job. I then revert back to my old ways and 'store with irradiance map' on the dome light which gives low render times but blotchy GI. Upping imap settings to 100/35 cleans it up but still not ideal.

So I have perfect quality render passes but 13hrs per frame at 1024 x 576 (!!!) or decent passes and just about acceptable blotchy gi with 14 mins per frame. Where's the happy medium?? Does it even exist? I can live with 'store with' on my dome light if I can get the render looking sharp but I feel the imap settings are making it look 'mucky'.

Any thoughts on what to test next?

Hmm, might be a better idea to downres the HDRI quite a bit? With a lot of the sIBL sets they supply one hdri for reflections at full res, a lower res one to drive lighting and often a light rig where any strong light sources are painted out of the hdri and replaces with actual vray lights with a similar light multiplier and colour.

[fewlo]

thanks, I'll try that.

on a side note, I had coplanar faces in that first test, sorting that out plus using the (superb) reinhard trick got the render time down to 3.5hrs, still stupid but a fair bit better than 13hrs!!

[joconnell]

Yep - big improvement alright! The Reinhard one admittedly does lose some dynamic range in the image but nothing that can't be added back in with a slight bit of levels or curves in photoshop afterwards.