[Bubbaloo]

I have used a test scene from the Viz Depot that was originally used to benchmark mental ray, which was converted to an IRAY benchmark scene by Jeff Patton. Find it here:
http://www.vizdepot.com/forums/showthre ... ge=1&pp=15

Interestingly on my computer (6-core 980X @ 3.8 and 9500GT) GPU+CPU rendering in IRAY is actually slower than CPU only. So I have included the CPU only result for my computer, and also included Jeff Patton's result on his dual GPU system.

I tried to recreate the materials in Maxwell as close as I could. Here are my results so far:



My first tests indicate to me that I would have to spend thousands on graphics hardware to match the speed of Maxwell on a good CPU with IRAY GPU rendering. Also, I have yet to see the Maxwell lighting and material quality in any other unbiased render engine, whether GPU or CPU based. I'd love to hear what others think.

[Bubbaloo]

A 30 second test:

[Half Life]

Maxwell looks better to me -- but that is no surprise.

Best,
Jason.

[sampson]

Yep, Maxwell definitely looking the best there.

[JorisMX]

Nice to see that the Maxwell Image is much clearer. Esp at the Bottom.

Also the Rough transparency on the Maxwell Render looks much better towards the Edges which is probably thanks to the improvements made in 2.0 in the BSDF model. Maybe this is due to different shaders/material setup.

I still cant help but find it a bit amusing that after all these years of idiots on cgsociety belittling Maxwell and the noise problems/long rendertimes they now all scream for Iray and want what Maxwell has been doing so fine for years now.

[Aniki]

looking forward to comparisons between Iray and Maxwell Render Realtime Preview

cheers

Aniki

[renbry]

i did a test on the weekend via 3dsmax and iRay vs Maxwell 2.1

in 8m13s Maxwell looked better, with reduced grain and a more realistic light falloff.

iRay and maxwell both had exactly the same ISO,Fstop,100W bulbs lighting the scene with very indirect light.

My machine at home is Intel Core 2 Quad 2.4, Nvidia 470 and 240.

a colleague who is very adept at GPU programming said, when i showed him these images, that CUDA is best for 'known-boundary' datasets like a grid-based fluid simulation rather than raytracing where samples have to do a lot of cross-checking accross a very slow system bus. He did say that SSS is all local calculations and could be done verrrry fast on GPU. maybe iRay will eventually have an advantage there.


Maxwell:


iRay:



Matt

[renbry]

i should state the obvious: the light falloff seems dodgy in iRay, the bounced light seems too bright and even if i crank up the exposure in Maxwell i was STILL not getting the same amount of grain in the same time. i think 8m13s is farely generous for such a resolution and simple materialed scene.

matt

[motopiku]

bubbaloo good test,
are waiting for new hardware and I am preparing a scene test practice. internal decoration, sss, displacement and caustic .. The final test for production.
I have to try this with Maxwell, Iray, Octane (now also the emitter), Bunkspeed e. .. other software.
I am very happy with how Maxwell answered with this interactive ... Very good.
I hope to soon be able to use the CPU to other computers on the network to the interactive.

good test, very encouraging. would be nice to try the interactive to the public.
Before Christmas? we hope.

[Hybaj]

It's funny to see how scene dependent the speed really is. In the end the only thing that comes to my mind is that Maxwell 1.0 would not be able to do stuff the way that Iray 1.0 does

Maxwell 2 mins (just cpu)

Iray 2 mins (cpu+gpu probably)


Gpu calculations do like simple stuff

[Bubbaloo]

It's funny to see how scene dependent the speed really is.

And hardware dependent. Your comparison is meaningless without hardware specs.