Announcement

Collapse
No announcement yet.

How much vram do i need

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • How much vram do i need

    Hi all-

    I've just built a custom PC for modeling and rendering and want to add a GPU next. Following the stats on my current project renders I see I'm using on average about 10GB of RAM (out of 64GB) with CPU rendering. When switching to GPU rendering in the future, would the amount of memory used be about the same? I want to make sure the GPU I buy can handle my projects, but this puts me right on the cusp between a 3080 and a 3090. I know that VRAM doesn't stack so even with 2x 3080's I'd still be at 10GB, but the 3090's CAN actually stack and get up to 48GB.

    In other words does CPU and GPU rendering each use the same amount of their own respective V/RAM?

    Also, would bucket rendering cut down on the amount of VRAM used?

    DESIGNS RENDERED, LLC

    Current Hardware Setup:
    • Ryzen 9 5095x
    • 64 GB DDR RAM
    • 2x Nvidia RTX 3090 Ti w/ NVLINK 48GB vram total

  • #2
    Hi charles_messenger and welcome to the forum!

    Originally posted by charles_messenger View Post
    When switching to GPU rendering in the future, would the amount of memory used be about the same?
    Short answer is no, applications don't use system memory and GPU memory in the same way
    If V-Ray CPU uses say 20 GB of memory for a specific scene, this doesn't mean V-Ray GPU will need to 20 GB of GPU memory
    V-Ray GPU is quite efficient on using GPU memory, but you still need to be careful about displacement.

    For instance, some of the scene that take 40 to 50 GB of system memory will be able to fit in 12 GB of VRAM. This is normal to happen
    That being said, a 3090 will be perfect in your case


    Originally posted by charles_messenger View Post
    In other words does CPU and GPU rendering each use the same amount of their own respective V/RAM?
    No, as explained above

    ​​​​​​​
    Originally posted by charles_messenger View Post
    Also, would bucket rendering cut down on the amount of VRAM used?
    Bucket mode will save system memory, not GPU memory. As the rendered pixels get saved to disk directly instead of being kept in system memory
    for V-Ray GPU right now progressive mode is more efficient on GPU memory, here are some useful advice on optimizing memory,

    -Enable "Hardware-accelerated GPU scheduling" in Windows 10 specially in the recent Windows builds, this frees up more VRAM on the GPUs
    It should be under Graphics Settings, and will require a system restart
    https://cdn.discordapp.com/attachmen...mage_66836.jpg
    https://cdn.discordapp.com/attachmen...mage_66837.jpg

    -Use the latest recommended driver for V-Ray GPU, this is very important and affects performance and stability. The current recommended driver is 497 Game Ready driver, this needs to be used on all machines.
    We usually have the current recommended driver in this page here
    https://www.chaosgroup.com/vray-gpu

    -Use progressive sampler mode in V-Ray GPU, it uses less VRAM overall, and makes use of V-Ray's smart sampling and scene adaptivity(more on render settings later)
    For rendering animations you can set a maximum render time in minutes per frame, this works well for rendering animations and avoids any stuck buckets near the end of rendering
    On average progressive mode will consume 3-4 GB less GPU memory than bucket, on high end GPUs

    -On an artist's machine, you could use a separate GPU for Maya and Windows performance(could be a lower tier GPU like the 1650 super for example). This way all VRAM is dedicated to GPU rendering on the second card
    You can then control the GPU used by Maya and other apps in Nvidia's control panel, under manage 3D settings
    https://forums.chaos.com/filedata/fe...9&d=1630320507
    In this case here, the 2060 super is used for Maya and 3060 is used for rendering. This saves another 3-4 GB of GPU memory on average. Maya/Max use a lot of GPU memory

    -Max subdivs of 48 is going to work for 99% of the GPU scenes, it is recommended to leave it as it is (V-Ray GPU needs more samples by nature compared to CPU rendering, this is expected behavior) you could stick to Noise Threshold for controlling quality vs render time. It is the only slider you use in V-Ray GPU to control sampling
    Noise Threshold of 0.01 to 0.03 should work, depending on if you use denosing or not.

    -In case of multiple GPUs, it is best to render one frame per GPU using Deadline for example or writing multiple .bat files
    This works well if the frame resolution is around 2k to 3k, you will need to use progressive mode and set a time limit per frame for best efficiency
    *If you are rendering in 6k or higher, you can use all GPUs on the same frame safely. And here bucket mode is better, this
    **if you are using Distributed Rendering on the same frame, Bucket mode is also better in that case.

    Best,
    Muhammed
    Muhammed Hamed
    V-Ray GPU product specialist


    chaos.com

    Comment


    • #3
      Muhamad, rtx3060 with 12gb arrived today for testing. I found the windows"Hardware-accelerated GPU scheduling" But i cant find min/max and noise threshold to set it as you described when I switch from CPU to GPU rendering..... both progressive and bucket
      Matthias

      Comment


      • #4
        Originally posted by digitarch View Post
        Muhamad, rtx3060 with 12gb arrived today for testing. I found the windows"Hardware-accelerated GPU scheduling" But i cant find min/max and noise threshold to set it as you described when I switch from CPU to GPU rendering..... both progressive and bucket
        Matthias
        Hi digitarch

        If you are using 3Ds Max and V-Ray 5 you will find Samples Limit setting(instead of Min/Max Subdivs)
        Set Samples limit it to 3000

        Best,
        Muhammed
        Muhammed Hamed
        V-Ray GPU product specialist


        chaos.com

        Comment


        • #5
          using 3dsmax 2022.3 & vray 52002, rtx3060 with 12gb: we opened one of our scenes, switched from cpu to gpu rendering, switched to rtx render device, set progressive with sample limit 3000, threshold to 0,03
          error: device not compatible, driver 512.15 is newer than verified 497.09, aborting render

          Comment


          • #6
            That is just a warning, the scene shouldn't crash here. It will continue to render.
            Could you please share a screenshot of your V-Ray Log in the frame buffer?
            Make sure you delete any unused render elements in the scene, for best performance

            Best,
            Muhammed
            Muhammed Hamed
            V-Ray GPU product specialist


            chaos.com

            Comment


            • #7
              Muhamed, now we installed the correct nvidia 497.09 driver, now there is no "driver error" but still rendering ends wit "abort rendering" I will post a vray log screenshot....

              Comment


              • #8
                vray log attached
                Attached Files

                Comment


                • #9
                  Please use my shaderball scene here, does it render for you ?
                  Do you get any errors?
                  Muhammed Hamed
                  V-Ray GPU product specialist


                  chaos.com

                  Comment


                  • #10
                    ok, Muhammed, your gpu shaderball scene is running fine, yeah!
                    RTX3060 at about 90% workload, cpu about 30% workloud.

                    Comment


                    • #11
                      The error could be about memory usage in your other scene, please start a new thread and we can troubleshoot there
                      Muhammed Hamed
                      V-Ray GPU product specialist


                      chaos.com

                      Comment

                      Working...
                      X