New Build (291)

[Roy Hirshkowitz]

This one introduces two new engines.  Both of these are very experimental at the moment. I may not retain either of them if they don't prove useful.  Although they appear very stable, there may be some hidden bugs.  Be careful using these for production until we have some confidence in their stability and characteristics.  The engines are only accessible via the nXtEngine AutoLisp command use (nXtEngine 2) -OR- (nXtEngine 3).

Both engines include more light pathways than the current product does and can therefore potentially provide more accurate solutions.  Two "tricks" which are used in the first two engines, transparent shadow rays and daylight portals, are not used in these new engines.  This gets rid of two big sources of inaccuracy.  Caustics pathways are automatically included.

Engine 2 can be much less noisy than the Path Tracer for most interiors.  It's convergence, on the other hand,  is not as graceful and will include some discrete artifacts.  It's usually a lot faster than the Path Tracer for interiors.

Engine 3 is a little more accurate than Engine 2, particularly when window coverings are involved.  It converges more gracefully, but slower for some interiors.  It is considerably faster on some exteriors.

Some things to keep in mind with these engines:

• Light leaks in the corners are possible when doing interiors.  Make sure walls, floors and ceilings are modeled as slabs with thickness.
• Both engines start slowly and accelerate.  Pixels/sec should increase as they proceed.
• Using accurate IORs is very important.  IORs of 1.00 should not be used for glass.
• Designating glass as Thick/Thin is more important with these engines.
• Translucency is not implemented.  Allow Surface Roughness is.
• These engines use more memory than the previous two.  64 bit is recommended.  Start your testing and low resolutions.

Let me know what you think-- I'm looking for feedback here.

[Roy Hirshkowitz]

Here's Alejandro's at about 35 minutes-- 640 x 480.  Very difficult lighting:

[Roy Hirshkowitz]

Just released 293-- fixes the problem with > 32 lights for engine 2.

[qweqwi]

Roy very great job...nxt 1.0.292...

[Miguel Tano Adanes]

Thanks Roy. I'll give it a try.

[Barry Thomas]

[IMG]http://i292.photobucket.com/albums/mm8/bwthomas/nXt%20RENDER/packet...[/IMG][IMG]http://i292.photobucket.com/albums/mm8/bwthomas/nXt%20RENDER/engine...[/IMG]

The first image  was rendered with build 292 packet tracer in 8 minutes, 68 passes. The second image was created with engine 2, 508 passes in an hour. The results for engine 3 were looking promising, but at about 64 passes the render screen went completely white.

Once being engaged in engine 3, there seems to be difficulty getting out of it and back into engine 2 or engine 1?  or is it 0?  None the less, I discovered that selecting a saved view switches the engine back to the packet tracer that had been in use when that view was saved. Is this an expected result....the engine saved with the view?

Loading 293 now

[Roy Hirshkowitz]

The engine is definitely saved in the model-- I don't think it's saved with the view.  You shouldn't have any problem using autolisp to change engine.  You can also use the Path Tracer checkbox, just check and uncheck should work to get back to 0. Send the model if you want me to try e3 here.

The differences between the two examples is very subtle but telling-- engine 2 capture the (small amount of) light that bounces in a specular fashion from the floor.  It's the reason the shadow under the side panel is more subdued.  Did the e0 one use any indirect light?

[Barry Thomas]

Here's the result using e3. Very similar to e2 in the way that the reflected light off the floor is treated, compared to e0. I am concerned about the long rendering time required to get this to a useful state. 502 passes shown here. My e0 versions of this entire series only required 30 to 50 passes at the same 1100 x 1100 resolution to produce an acceptable image for publication. The resulting image, though, makes it worth the wait....I think. I'm still on the fence.

I've got another image cooking on another machine in e2 that is showing som interesting results. I'll post it in a minute.

[Roy Hirshkowitz]

This stuff really doesn't compete with e0 unless you're talking about some pretty involved scenes.  You probably should compare results to the path tracer.

The e0 floor looks different, BTW, because that image is not converged.  If you allow it to run for a long time the floor will get a lot fuzzier.

[Barry Thomas]

This is a model I did some time ago. I was fairly satisfied with the result. The image on the left was produced with e0, the other with e2. The marble floor seems to have more depth. The color is much different between the two images. Settings in both renderings are identical, same background hdri.

[Roy Hirshkowitz]

Not a really good test.  This isn't the sort of image which is going to show much difference.  There are almost no inter-reflections in this one.  (I desaturate the sky a little for glossy reflection, obviously more aggresively in the first case-- that's something that can be easily tweaked.)  The image below, for example, shows a big difference.  There is a reflected caustic on the ceiling from the shiny floor which is totally absent in previous versions.

[Jorge Javier Lara Domínguez]

Adjuno dos imagenes, la primera fue realizada en version nxt286 con Path trace. La segunda con nxt293. Ambas tienen el mismo tiempo de proceso 165 minutos. La version 286 tiene 138 pases, la 293 tiene 85 pases. Que opinan?

[Roy Hirshkowitz]

Which engine?  Interesting perhaps from a performance point of view-- otherwise there's not much there.  (I'd expect an almost identical result.)

Or this one-- where accurate light is cast by sources behind frosted glass covers: