Iray GPU "acceleration" - I'm not seeing it. Kerkythea vs. Iray
I made a simple test scene to compare Kerkythea and Iray speed. Disappointing but also informative. Iray keeps crashing my graphics card driver* and gives no perceptible speed increase. Is there some non-obvious thing needed to get the speed benefit of the gpu?
Test
Kerkythea: MLT+BPT, cpu only (it has no gpu mode)
Iray: photoreal with architectural mode and caustics and cpu+gpu.
I ran KT for an hour, saved picture, then ran Iray for an hour and visually compared pictures.
Results
After an hour of rendering the images were at about the same level of completeness by visual inspection. Then the driver crashed.
So for a GT 640 there is no gain in speed with Daz Iray for high levels of realism. A new 64 bit program using cpu plus gpu should be faster than an old free 32 bit cpu only program that is sort of a public beta test.
This also implies, though I didn't test, that Iray in cpu only mode is pitifully slow. That is unfortunate since due to crashing I'd have to run it in cpu only mode. I do like many things about it even without a speed boost.
Here is the KT image. I don't have the Iray image due to crash.
*Credit though that crashes could come from 4.8 being a beta, I think, or since my card is not to minimum specs recommended. Crashes are less frequent in gpu only. It did run for renders of two hours on other images.
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
Version of 4.8?
There have been 2 general builds of 4.8 released, so far.
Also, what is the size of the RAM on your video card?
Granted, that scene shouldn't be more than 1 GB, but if your card is only 1 GB and it is also pulling display duty, then there is a chance that you were rendering CPU only...
Also, what is your driver version?
HI mjc1016.
Daz version was 4.8.0.55. I installed 4.8.0.59. GT640 driver version 353.06 and is most recent non-beta update. Card has 2GB memory. This scene uses 390MB. Card is also runs the display.
Running another test with 4.8.0.59. So far it is still slow.
Interesting. A gpu monitoring program says both current and max so far gpu speed is 329MHz but the card is rated 901MHz (default, is not overclocked).
Update: Evga PrecisionX shows the gpu speed going to 901 when rendering.
Here is a better test to see if gpu acceleration really helps with my setup.
I set Max Samples to 25 so the render would stop at 25 iterations. Then I rendered the scene above with cpu, gpu, and gpu + cpu and noted the time it took. Results:
25 iterations
cpu + gpu: render time 5m 21.39s, 12.86 seconds per iteration
cpu only: render time 5m 21.72s, 12.87 seconds per iteration
gpu only: render time 22m 5.32s, 53.01 seconds per iteration
So for my set up Iray is a cpu renderer with no practical benefit from the gpu. On the bright side it is not slower than Kerkythea as a cpu renderer. However, a 64 bit cpu renderer should be faster than a 32 bit one.
Anyone know of, or performed, comparisons to other gpu accelerated renderers? Especially Cycles.
That's very odd - I have an even lowlier card than you, 450GTS with 1GB of RAM, and using the card alongside the CPU on a scene that will fit (naked Vicky on a Polygon With Not Much Else) the card roughly doubles the speed I see with just my i7 920 CPU. What are the rest of your system specs?
That is very odd...because with my 1 GB 440, I show a significant speed up when GPU only is used...in fact the times are about the exact opposite...with GPU alone being less than 5 mins and CPU only being around 20. This was done with just a few primitives and an HDR image for lighting/background. It was on a dual core 2.6 MHz or so Athlon. with 4 GB of system RAM on Win7.
And most of those who have posted in the various Iray threads also report a significant speed up.
I don't have Kerkythea on that system, so I can't run against it...but the same scene, with the HDR takes about 20 min in 3DL and an 1 hr+ with Luxrender (not accelerated, Lux 1.3, to be exact...)
As to the speed difference...in the Nvidia controls, there are settings to control whether the card throttles down or runs full speed and if it does it on demand or all the time.
Wow!, you use Kerkythea?, dunno how time consumig applying material is actually, I've used that on earlier versions and better changed to Luxion's Keyshot.
LuxRender does not benefit very much in speed rendering, I've used my GTX 780Ti and it does mere 60s/p in hybrid mode, very bad.
for Nvidia cards, the GPU Temp is very important being monitored, extensive use of GPU cores can lead to a +80°C and eventually crash/stop until being cooled again. I use Asus GPU Tweak for maximize the GPU fan at 100% of use and low my GPU temp at 45-50°C.
Thanks everyone.
Richard: system specs:
Graphics: GT 640, 2 GB memory, 901MHz gpu clock, 384 cores
CPU: AMD FX 8320, 3.5GHz, "8 cores' (AMD claims threads as cores, it is really 4 cores)
Memory: 8 GB
Windows 8.1 pro
mjc1016: Yeah, it seems strange to me too. I've read about the speed increase in posts here and expected to see obvious improvement but not super speed. I've got to be doing something wrong.
zilvergrafix: I watched the gpu temp and it didn't change much. When I tried to overclock it a little it stayed around 59C but was unstable and crashed so I went back to default. Typically I want art and imagination but when I want realism I use Kerkythea because it is free. It is not easy to use so I'm hoping to replace it with Daz-Iray or Cycles (I'm moving to Blender after the demise of trueSpace).
It hasn't crashed since I installed the latest version of Daz.
I did the tests again with architectural and caustics off. Both were on in the tests above. Results:
25 iterations
cpu + gpu: render time 1 minutes 25.90 seconds
cpu only: render time 2 minutes 35.67 seconds
gpu only: render time 2 minutes 4.92 seconds
Looks like the gpu is worthless for architectural or caustics or both but is effective without them.
The Luxrender "Hybrid" mode was an early implementation (back in 1.31 if I remember correctly). The main benefit is in complex scenes where there is a lot of calculations for light intersections, as those calculations are made by the GPU.
To see your GPU used by Luxrender to its full potential, you would need to use Luxrender in openCL (Luxrender 1.41 or above. With an exporter that exports to the openCL render engines).
Have you activated the OptiX setting in "Advanced"? I had really slow renders lately (my PC usually can just use the CPU-mode), until it dawned on me that I had reset the render settings to default, and thusly deactivated the OptiX setting for the CPU. Once I checked that, render times improved visibly.
I have seen people have problems with AMD CPUs and OptiX, I think. Certainly there is some combination of settings that works on Intel processors but is of no benefit or even counter-productive with at least some AMD CPUs.
Thanks again everyone!
Caustics is the culprit. Architectural alone is slower, as expected, but it is caustics that makes it super slow.
I read somewhere that architectural mode is metropolis light transport and does caustics but caustics mode does more detailed caustics. I can live with that. I'll try arch. mode no caustics to learn when or if I need to pay the time penalty (jewels probably).
Reading your advice that optix is important, good or bad, I tested with and without it and saw no obvious difference. Tested in photoreal gpu only. I guess AMD trouble would appear with cpu active.
Arch. mode might be MLT. What is regular render mode with no arch. or caustics? I'm guessing it is bidirectional path tracing because there are no photon controls but there are things like max path length.