Rendering with Iray GPU not using 100% GPU

[Anonymous]

Hi guys, I'm testing Nvidia Iray renderer. The thing is that I'm trying to render using my GTX 1060 GPU and.. it seems to work but.. I don't know if it's working 100% fine.

When I render using mental ray, CPU usage goes to 99%, but using Iray, GPU usage varies between 0 and maybe 10%. Shoult it be like this or am I doing something wrong? Is it the expected behavior? I'd like that CPU usage was higher and rendering time, shorter.

Thanks in advance!

[Byteman · Accepted solution]

You can disable or enable both the CPU and GPU contribution to IRAY rendering from the Hardware Resources rollout of the Renderer tab in the Render Settings dialog. A working GPU has noticable difference:

1. Just Uncheck Your GTX 1060 checkbox in the mentioned section.

2. Try rendering with let's say 100 iterations.

3. Note the time it takes to finish.

4. Now check the GTX1060 to enable the GPU.

If it doesn't reduce the rendering time considerably, or you don't see any GTX 1060 checkbox in hardware resources section, there might be a problem.

But, don't worry just because you don't see the right percentage next to 3dsmax.exe process. Some apps' GPU usage is not reported as the GPU usage of its executable, but some other host process, even not reported at all. This feature is fairly new to Windows 10 and I see changes in the way it displays GPU usage with each update. It is still being improved I guess.

One other clue to understand if your GPU actually works may be your mouse getting stuck periodically during render. I said "may be" because I'm sure they'll figure it out on a newer IRAY version or maay be a newer PC. It may be already be different from system to system. As far as I'm concerned, an i7 3930 K with 32 gb ram experiences such lags with a GTX 670.

[Anonymous]

Thanks a lot, I did the test that you suggested. For example, if I try to render for 5 minutes using only cpu, I can maybe get about.. 4 o 5 iterations, and if I use gpu, I can get much much more, maybe near 300-500 iterations in the same amount of time.

The thing is that it takes more or less the same time to render a scene using Mental Ray and Iray (althoug Iray has more photorealistic quality). I thought that time would be drastically reduced using GPU but.. it's not. I don't know if it's the expected behavior or I am doing something wrong 😞

[Byteman]

I think there is nothing wrong. Iray takes advantage of your graphics fairly well.

Unfortunately Mental Ray has been going in a wrong direction since it has been acquired by NVidia. In my opinion, NVidia tried to make the renderer more bound to graphics card power so that the users will be forced invest more on gpus. But I guess nothing went as planned. They have finally announced Final Gathering as deprecated and photon mapping has long been gone. They have introduced the gpu based GI next as new GI engine. It is easy to setup and image quality is superb, but it is unusably slow on even the Kepler based cards which are not that old. Using CPU is absolutely out of question.

Forcing the users to a renderer which imposes the use of Gpus and gpus only would have a negative effect on 3ds Max's popularity because, in order to use the software with the bundled renderer you'd have to invest considerable amount of money on graphics cards, too. I believe that is a rational reason to expel MRay from 3ds Max. Furthermore one card is never enough. Iray creates images of stunning quality but you'll need more horsepower. Always more... Cryptocurrency mining drained graphic card stocks and cards became very expensive. Not only one time sale price, but the power usage of cards is another handicap. Your power bill will be considerably high if you have 3 GTX cards and render 24/7. 

Be it known that final gathering is no longer usable, it works but you can never get satisfying results. High FG values take endlessly long to render and image still has artifacts. Values change from scene to scene and it causes the adjustments to become a pain.

As one of the loyal enthusiastic users of Mental Ray, I'm so upset to say that I feel a necessity to switch to a more reliable and efficient renderer. Mental Ray no longer fits into that description. So does Iray. You can render a car or a few objects but when you have extremely complex scenes like arch work, with large maps, you run out of gpu memory and all your gpu investment becomes in vain. Furthermore it doesn't recognize certain maps.

You make several adjustments for Iray, assign materials  and at one point it doesn't let you move on for some reason and you have to make all the assignments from scratch. That is not acceptable.

[dgorsman]

Switching to a GPU-renderer isn't a blanket "makes it go faster" choice.  It's still going to be doing some work on the CPU.  With basic rendering that's going to be the bottleneck, and you're not going to see a whole lot of improvement in time.  Scene optimization counts as well.

Also, a single GTX 1060 isn't exactly top-of-the-line in terms of clock speed or CUDA cores.  More cards running in parallel would almost certainly reduce the time on complicated renders.  Higher-end cards have more cores and higher clock speeds so may get more done but you may need to skip a couple of models before getting faster results than multiple, lower-capability cards.

[Byteman]

I have to state that the number of Cuda cores or clock speeds, just like the cores and clock speeds of Cpus, do not necessarily mean a proportional performance gain. When talking about the performance of exactly identical chips, those with higher clock speeds will definitely perform better, but these devices are always governed by the brand's financial rules. 

Just to give an example, a previous Nvidia card Gtx580 with 512 Cuda cores were performing better than its 1536 Core sibling GTX680. The Kepler based 6xx was announced to be more power efficient but there was no performance gain in Cuda based apps.

This is what the chip manufacturers always do.Increase cores decrease frequency, distribute a slight performance increase in the course of more years. Sell the same product for more years to come.

Never expect a miracleous performance gain. If AMD hasn't introduced its Ryzen chips, we'd still be waiting for Intel's next i7s.

[Anonymous]

Thanks a lot, so, after all this talk.. which renderer do you recommend to use? it's just a simple video for college, no more than 1'5 minutes, 500.000 polys and maybe.. 720 or 1080p tops.

I've been using mental ray until now but as it took too long for a single frame, I wanted to deepen into it a little before starting this new project.

Thanks again for all your help!