Best rendering option for me: Iray vs Server?
greg.madro_83c2fe6748
Posts: 1
A bit newer to all of this. I used poser way back when and have recently gotten back into rendering.
Currently I have Daz Studio running on my main PC. It is a i7-6800K with 6 cores, 128GB RAM and an nvidia 980GTX. I have been using Iray for my renders but wanted to see if my server may be a better option for me. It is a dual socket Xeon E5-2450 with 16 total cores available. I have seen a few mentions of external rendering engines like Reality and Octane and wanted to know if anyone can help lead me in the right direction. I would like to keep it on the cheaper side to start (I see that octane would likely run $500+) if using my server will be a better option for me? I would like to have renders that would be of quality similar to Iray and hopefully faster if run on the server.
Thanks in advance
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
Your 980 is going to be significantly fast than that server for Iray. I'm willing to bet that extends to Octane as well but don't know about Lux since I stopped watching it long before they got their GPU stuff underway.
...while CPU rendering on the Server will be slower than on the 980 it will not need to fit under the 4 GB VRAM limit (actually slightly less depending on your OS). If a fair amount of your scenes exceed 4 GB, they will dump to CPU mode anyway and all you'll have is 12 CPU threads instead of 32 and 12 (roughly around 10.5 GB after Windows and system utilities) of physical memory . I have only 12GB and keep experiencing render jobs dumping to Virtual Memory which is even slower (as it hits your HDD) as the open scene file and Daz programme also take up system memory resoruces as well as the render process. So not only are you down to slower CPU/memory speed but even slower platter rotation speed and seek time for the HDD.
It really boils down to how complex of a scene you generally plan to create. A good number of mine are around 7 GB on average which leaves only (3.5 GB for rendering) and some are almost 9 GB with the Daz programme open. I have been looking to upgrade the memory to 24 GB but sadly my system is an older generation one that uses DDR3 memory which is getting harder to find. Most of what I am seeing is ECC server memory which won't work in it and what few commercial kits are still available have been going up in price almost weekly. .
How much memory does the server have?
The server has 192 GB of RAM. I certianly understand the power of GPU rendering but I was curious if 16 CPU cores would be able to perform better. Based on what I am seeing I would be better off with another GPU to up the render speed locally than try to do an offload to the server.
...if I am reading this correctly, the server has dual 8 core Xeons?
Compared to your 980, not at all.
...true from a total core number standpoint.
If indeed the server mentioned has dual core Xeons, that means a total of 32 processor threads,. Yeah, a ripple in the pond compared to the 2048 CUDA cores of the 980., But then the reverse holds true for available memory, 4 GB on the 980 vs 190 on the server's MB (taking 2 GB out for OS and system utilities).
Again were it myself in this situation, I'd go with the server as 4 GB would be OK for character proof and in progress test rendering, but at best support about 30% to maybe 40% of my finished scenes. For my purpose that is diminishing returns especially as I already mentioned with 12 GB of physical memory, render jobs often dump to glacial swap mode. With that kind of memory horsepower, render jobs that exceed the 980's memory would still finish in less time than on the 12 GB system as we are looking at a difference of 20 CPU threads between the two. With 190 GB of available memory it would ensure no Daz render job would ever dump to virtual memory. Xeons are also more efficient at calculating big number sets compared to i7s as that is what they're designed for (which is why they're preferred in professional workstations).
When my retroactive benefits finally come through, then yes I will be looking to build my dual 12 core Xeon monster. I'll only be running with 128 GB to take advantage of four memory channels instead of three or two (need W8 Pro to support 256 GB but just can't stomach that hideous UI). Oh, it will have a 6 GB 1060 to run the displays and as I mentioned for test rendering, but the system's real horsepower will be provided by the Xeons and physical memory. With 48 threads I can dedicate 16 from each CPU to rendering (which is four times what I have now) and use the other 8 from each for production work and not experience sluggishness. This is where having a high number of CPU cores has an advantage. Basically this single system can handle the tasks of several (this is what servers are meant to do), so I can be rendering a scene, setting up another, along with creating/modifying textures or doing postwork in a 2D programme.
Should Nvidia see AMD's bet and fire back with a 16 GB HBM-2, 4096 core prosumer GPU card of their own that is priced competitively, then they'll get my attention. I'll have plenty of room to drop one in.
I still don't quite understand the need for massive amounts of RAM to render scenes. Yeah, I can see if you load a bunch of stuff into your scene with some high rez textures it can require a lot of memory. But there are ways around it that can vastly speed up your renders and require far less memory, and give similar (or much better) results. I'm certainly no expert, but from what I've seen in videos of professional studios and such they seem to prefer breaking their scenes into layers and compositing them together in the end. I think it gives far more control over the final product.
For example, the simplest version is if you have a character and a background. Render the background ONLY (without the character in the scene), then use that rendered image as a background in a separate scene with just the character doing its thing. Then just render the character with an alpha channel in its own scene. Then those two rendered images/sequences can be composited together, and you can do all kinds of tweaks in the compositor (depth of field, color adjustments, and so on).
Another method, and one that I actually use, is to break the scene into parts. Like if you have a character in a house, break the house into parts that are separate scenes, and when the character is in the living room just use the living room scene.
Yeah, it might take some more thought and effort, but once you get the hang of it it really makes life a lot simpler. If you've ever rendered just a character with no background you quickly realize how fast and little memory it is. And the whole process of compositing makes you think more about controlling the parts of your image to improve the overall quality, rather than just taking whatever you get with the entire scene in one render.
BTW, there are some free compositing apps out there that are professional quality. I can't recall off the top the one I was using, but some are pretty amazing.
...if a scene is complex (like the ones I usually create) with many characters, props and other scenery elements in different positions on the "stage", multi plane compositing becomes much more cumbersome, particularly if shadows from one render "plane" fall on elements in others. Another matter is reflections. if you have items with reflective surfaces, you need the surrounding scene to provide accurate reflections. For example, in my old bus stop scene with the two girls, I actually placed scene elements off camera so there was something to be reflected in the shelter's windows There are also times when objects need to be off camera to produce shadows that you want to appear in the final rendered scene (in that same scene I have a couple trees off camera to produce shadows that I wanted on the roadway).
Keep in mind also I plan to create large format images which will require a higher resolution and amount of detail.
As I have mentioned, postwork has never been my strong suit, particularly when digital painting is involved, which is why I need to get the most I can out of a single render pass. Therefore I need all the CPU and memory horsepower I can get. Also, as I mentioned above, having a high amount of CPU thread and memory overhead, means I can perform multiple tasks at the same time without them having a detrimental impact on each other.
But it is doable why limit your scenes to ones you aren't satisfied with just because of a bit of extra work.
I've always had to composite my larger scenes because of hardware restraints, composing layers doesn't really take any postwork skill as ebergerly says its just planning things out and thinking ahead, I don't let it stop me creating the images I want to create. If you can't afford the hardware the you have to find workarounds.
...I've tried to get compositing down and it just doesn't look right compared to rendering the complete scene in a single pass. Something is always off, be it shadows or refelctions that I need or lighting on some elements doesn't look right. I always make sure my rendering camera is "locked" so the positioning never changes, even when working on single pass renders and make use of multiple "work" cameras to focus in on specific elements for positioning & such (this was the reason I dumped Reality4/Lux as it would randomly select one of the work cameras as the render camera even though I had already set the primary one).
I already do a fair amount of pre-planning for scenes. Having been involved in the fine arts (oil and watercolour painting) I also have a very clear vision of what I want the final scene to look like from the beginning.
For characters, I create them as individual subset files which I merge in and paste to a placeholder in the location where I want them after I am satisfied with the morphs, clothing, hair, skin, and basic pose. I do the same for more complex scene elements as well. This saves a good deal of time in not having to deal with increasing viewport sluggishness as more polys and textures are added to the scene. I also switch to a less detailed viewport mode during setup (I never use Iray View mode as it is excruciatingly slow and ends up crashing the programme after a while). I used to work on stage and set design which included blocking of the actors, various props, set elements, etc, and use the same techniques here to aid in the workflow.
With my current system, I schedule my rendering overnight or when I'm away from home for the day. That way I don't have to sit around bored "watching the paint dry" so to say.
Fortunately I am retired so I have more time to work on my 3D projects. However I sitill like my worktime to be productive rather than just sitting around waiting long for evne a test a render process to complete.