The advantages of more system RAM for Daz Studio renders?

rexhavit@hotmail.com[email protected] Posts: 1,178

I have 2 x Nvidia 1080ti in non SLI mode, a 24 thread i9 with 44 lanes... When I had 16GB of ram my renders would constantly run out of system RAM and DAZ would interrupt and terminate my renders.

When I went to 32GB of RAM this almost never happens.

So, it is clear to me that RAM helps the processor lanes and the graphics RAM and CUDA cores talk to one another.

It would then seem that going from 32GB of ram to 128GB would open the communication lines between 

the main processor cores, the lanes, the graphics RAM and CUDA cores and accelerate my renders even further.

At what point does having more system RAM reach a bottleneck with the 2 x 1080ti?

Would there really be much more of a noticeable difference in renders with 128GB of system ram versus 64GB of ram?

It seems the more system RAM, the faster the computer could access the CUDA cores through the PCIE bus lanes...

Any thoughts/help on this?

 

Post edited by [email protected] on

Comments

  • SimonJMSimonJM Posts: 5,942

    Well, parat from that (as far as I know) that is just wrong, no.  More RAM simply means teh computer has the facility of holding more data in direct access memory without the need to page out to a swap file.  I beleiev there IS soem correlatiion between numbers of graphics cards, their VRAM and how much RAM is needed to support them, but so long as you do not have a bottleneck (and RAM is the worst bottleneck to have) there will be no speed improvement increasing system RAM.

  • rexhavit@hotmail.com[email protected] Posts: 1,178
    edited April 2020

    Thank you Simon for your awesome comment, I definitely had a bottleneck with just the 16GB of memory and switching to 32GB remedied Daz from periodically closing on me when I would start a render (especially when a lot of geometry was in the scene).

    I read somewhere that you need at least double the system RAM for the amount of graphics RAM you have.

    I was just wondering whether if the CUDA cores would benefit from even more.

    With the CUDA cores being so numerous it seems more system RAM would be able to serve them the data in a parallel manner along the PCIE lanes.

    Maybe it is the graphics memory that serves the cores this data along the PCIE lanes.

    A schematic of this process would be helpful.

    But it is clear that there was a definite bottleneck with only having 16GB of RAM and increasing that to 32GB my renders became nearly flawlessly stable.

    I am of the mind to think that more RAM would enable the system to feed the CUDA cores with data more efficiently.

    I don't really know but it seems worthy of consideration and inquiry.

    Post edited by [email protected] on
  • SevrinSevrin Posts: 6,300

    The rule of thumb is to have twice as much system RAM available as VRAM, so double your 11 Gb on the 1080ti.  You need a little extra for Windows and whatever else you have open.  When I started using Daz, I was running an 8 Gb video card with only 16 Gb system ram, and that was always capping out.  32 Gb is much more comfortable.

  • Excellent advice Sevrin, I plan to upgrade from 32GB to at least 64GB. I am not sure spending 6-800 dollars for 128GB is in my budget.

    Having a 24 threaded CPU, I wonder how my CPU renders might be increased with more ram also.

    Could more system RAM boost the performance of using my CPU to render also in DAZ?

     

     

     

  • SimonJMSimonJM Posts: 5,942

    Again, a bit of a false premise. So long as all sceen data (and render overheads) and other system requiremnets (Operating System, other running applications) fits in physical RAM you are good. If what you are doing works well and without paging/swapping with 32 or 64GB then increasing that will not affact the outcome. Think of it like an upgradable house: you have 4 members of a family in a 4 bedroom house, will adding another 4 bedromms help in any way?

  • SpottedKittySpottedKitty Posts: 7,232
    RexRed said:
    Having a 24 threaded CPU, I wonder how my CPU renders might be increased with more ram also.

    Could more system RAM boost the performance of using my CPU to render also in DAZ?

    All it does is let you render bigger and more complex scenes. They'll run at different speeds depending on the details of your system specs, but the final result — assuming you render to the same convergence and the same number of iterations — will always be pretty much the same (unless you compare using a graphics card that can run the Iray denoiser on your render against one that can't).

  • kenshaw011267kenshaw011267 Posts: 3,805

    You do not need more RAM for Daz.

    You need to fit the scene completely in RAM, it appears, before it is transfered to the VRAM (the scene takes up less space on the GPU because of texture compression and some other stuff outside my area of expertise). So if it works, it works. You'll get no performance increase from having more.

  • rexhavit@hotmail.com[email protected] Posts: 1,178
    edited April 2020

    I like the 4-bedroom analogy. What if the 4 people who sleep in the 4 bedrooms each own a lot of junk? Then the 4 extra bedrooms might come in handy.

    Junk like, volumetric lights, lots of props, strand-based hair, boxes and boxes of hats and clothing; how about a 4-car garage? ...and volumetric clouds, the neighbor’s house out through the window and the city skyline off in the distance?

    Is the video ram going to dictate how much system ram can be used?

    Why must we have double the system ram that we have of video ram if system ram needs to exceed video ram as it is?

    It may not render faster but trying to hold a lot of geometry and at the same time as trying to assist the rendering process.

    There must a be a point of diminishing returns but being able to fill a scene with more stuff should make the renders at lease more interesting.

    I don’t mean to come off as argumentative, I am just really ignorant about how this all works and I figure I will milk the moment for as much info as I can get.  laugh

    Post edited by [email protected] on
  • SimonJMSimonJM Posts: 5,942

    The ansswers are, as usual, it depends ./.. if using Iray then all scene data must fit into VRAM or it will fall back to using CPU. Other render engines can use system RAM whilst utilising GPUs and I suspect Iray may well go that way to at some stage. 3Delight only uses system RAM so mre RAM equares to bigger scenes possible.

    In general more IS better, but you can get to a point where it is fruitless. I have 64GB of RAM (with a Titan X for rendering) and have no need for more, for example.

  • DustRiderDustRider Posts: 2,691

    Keep in mind my responses are specifically focused on using Iray with GPU rendering. If using CPU rendering, then more memory will enable rendering of larger more complex scenes. There is however, a huge caveat with rendering huge complex scenes in DAZ Studio (DS). You may run out of patience long before you come close to using 128Gb ram in DS. DS does not scale well with huge complex scenes. If you tend to have a lot of applications open at the same time, then 128Gb may be useful.

    RexRed said:

    Is the video ram going to dictate how much system ram can be used?

    Ummm ..... pretty much, if your using the GPU to render. Some scenes might take more RAM simply because sometimes the DS scene alone will take up more ram than the than the preprocessed Iray version of the scene, which will be close to what memory the scene it will consume in the GPU. For example, if you have several complex dForce simulations, your scene data will be huge (to store the animation of the dForce items), but the data sent to Iray will be less that because you are only sending one frame of information to Iray, not the entire simulation.  

    RexRed said:

    Why must we have double the system ram that we have of video ram if system ram needs to exceed video ram as it is?

    Simple, DS must pre-compile the Iray data it sends to the GPU from the DS scene data. This data will stay in system RAM while the Iray session/render is active. If you are filling almost all of your GPU memory with the render, then you will need that much free system RAM (in addition to what DS is already using) to be able to send the scene to the GPU.

    RexRed said:

    It may not render faster but trying to hold a lot of geometry and at the same time as trying to assist the rendering process.

    As others have already noted, there is no swapping of scene data from system RAM to the GPU during rendering, so once the render is started, the GPU works independently of any scene data held in system RAM. Yes, Iray is extremely inefficient with system resources. You may note that your CPU is fairly active during rendering. Iray uses one CPU core for each GPU in your system. This is Iray sending the image data to the system, and I'm guessing that the CPU is performing the convergence analysis, etc., but it is not sending any additional scene data to Iray.

    RexRed said:

    There must a be a point of diminishing returns but being able to fill a scene with more stuff should make the renders at lease more interesting.

    Yes, with regard to GPU rendering, the point of diminishing returns is when your RAM exceeds the amount needed to hold the DS scene data and the preprocessed data sent to the GPU, having more than that will simply be unused RAM (relative to DS/Iray - if you use/open other software at the same time, the extra RAM would be useful). If you add enough content to your scene that it no longer fits in GPU memory, it will default to CPU only rendering, so additional RAM will be useful, but rendering will be much slower (and your GPu's will sit idle).

    There are other GPU based render engines that use system RAM to store scene data. With those render engines, the additional RAM can be very useful for rendering larger scenes. Iray is not one of those render engines. The only one that is usable directly in DS is Octane Render. So if you wanted to use octane, you could render huge scenes on your GPU's using system RAM. But, you will also need to do your own shader work to get the same results you do with Iray in DS. For most people this extra work in not worth the additional features offered by Octane. Blender Cycles can also use system RAM for GPU rendering. There are 3 options for moving your data from DS to Blender, but to make use of this option you will need to learn Blender and how to set up shaders in Blender.

    RexRed said:

    I don’t mean to come off as argumentative, I am just really ignorant about how this all works and I figure I will milk the moment for as much info as I can get.  laugh

    Bottom line, if your using GPU rendering with Iray with one or more 1080ti GPU's, 32Gb is quite adequate. You might occasionally need more than 32Gb, depending on what you are using (like a lot of dForce content). More than 64Gb would simply be a waste (unless you have a lot of other things open on your computer at the same time). I have a system with 128Gb of RAM, but my need for the additional RAM is not due to DS/Iray, and it has no direct benefit to rendering with DS/Iray. I work with huge Lidar data sets, I use the data in Blender, and use photogrammetry software like Agisoft Metashape (Photomodeler). Those applications and data sets require a lot more system RAM than DS/Iray rendering on GPU (I only have an 8Gb GPU, so for DS/Iray only, 32Gb would be fine, and was on my previous system).

    I hope all this helps you to understand how memory is used be DS and Iray. If you want 128Gb of  system RAM, and feel you really have a need for it, that's fine. But if your primary focus is DS/Iray with GPU rendering with the GPU's you have now, then it won't help at all.

Sign In or Register to comment.