Announcement

Collapse
No announcement yet.

How much GPU RAM would be enough?

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

  • How much GPU RAM would be enough?

    Here is an interesting question for you guys. Based on your experience with V-Ray and V-Ray RT GPU, how much GPU RAM do you think would be enough for the average scenes that you are working on (assuming that instancing is available on the GPU)?

    Best regards,
    Vlado
    33
    8 GB
    15.15%
    5
    16 GB
    51.52%
    17
    24 GB
    24.24%
    8
    More than 24 GB
    9.09%
    3

    The poll is expired.

    I only act like I know everything, Rogers.

  • #2
    Hi Vlado,

    I'm thinking that it may be important to note here that RT/GPU typically seems to render a scene in significantly less RAM than when you render normally. I'm guessing that this is because the GPU gets sent only scene rendering information, and doesn't have to load the scene file itself like Max does.

    As a quick example, when I render say a 275k face scene normally, Max reports it took 1.2+GB of RAM to render. But when I render the same scene in RT/GPU, the GPU monitoring software reports that I am only using 648MB of RAM. In this case, that is in the neighborhood of half.

    I'm also guessing that if you are rendering to higher resolutions and need large bitmaps, then that would make some difference as well. Also, larger files may have less of a ratio between platforms, but larger files also take up more RAM loading into Max.

    So I guess that what I am saying is that folks should test their files strictly on the GPU for RAM usage, and do not go by what Max has typically reported in the past for their production files.

    I hope this makes some sense,

    -Alan

    Comment


    • #3
      Originally posted by Alan Iglesias View Post
      So I guess that what I am saying is that folks should test their files strictly on the GPU for RAM usage, and do not go by what Max has typically reported in the past for their production files.
      How can people test if they need more than 8gb when there isnt a card that would let them get close to rendering it?

      Nearly all of our scenes are under 16 by the final render, a couple creep over but they're the exception. A vast number (nearly all, tbh) stay under 8 until the last couple of days when smaller details are added and the settings get ramped up, but being able to use gpu all the way through the actual development would be amazing.
      Last edited by Neilg; 25-10-2012, 04:51 PM.

      Comment


      • #4
        Originally posted by cubiclegangster View Post
        How can people test if they need more than 8gb when there isnt a card that would let them get close to rendering it?

        Nearly all of our scenes are under 16 by the final render, a couple creep over but they're the exception. A vast number (nearly all, tbh) stay under 8 until the last couple of days when smaller details are added and the settings get ramped up, but being able to use gpu all the way through the actual development would be amazing.

        Im with you on that one. We managed crazy production scenes with just 16 gb of ram, but!.. I find that having 24 gb is a lot better, reason why? simple, you will spend more time making a scene fit into 16 gb, then just simply rendering it using 24 or more.

        With that said, I would vote for 24 gb of ram as a standard base, but probably 16 will do.
        Dmitry Vinnik
        Silhouette Images Inc.
        ShowReel:
        https://www.youtube.com/watch?v=qxSJlvSwAhA
        https://www.linkedin.com/in/dmitry-v...-identity-name

        Comment


        • #5
          Originally posted by cubiclegangster View Post
          How can people test if they need more than 8gb when there isnt a card that would let them get close to rendering it?
          I was hoping that Vlado would comment on the RAM differences I am seeing so folks could accurately predict their RAM needs for GPU rendering.

          -Alan
          Last edited by Alan Iglesias; 29-10-2012, 11:19 AM.

          Comment


          • #6
            Originally posted by Alan Iglesias View Post
            I was hoping that Vlado would comment on the RAM differences I am seeing so foilks could accurtely predict their RAM needs for GPU rendering.
            On the GPU, we upload data that we need to do rendering with V-Ray RT for the particular scene. 3ds Max stores a lot more information that allows the scene to be manipulated interactively (i.e. the links between objects; controller animations etc etc). The 3ds Max UI itself takes about 1 GB if I remember correctly.

            Best regards,
            Vlado
            I only act like I know everything, Rogers.

            Comment


            • #7
              im assuming this poll means that chaosgroup is working on their own GPU card with stacks of ram at a super cheap price.

              Comment


              • #8
                Originally posted by super gnu View Post
                im assuming this poll means that chaosgroup is working on their own GPU card with stacks of ram at a super cheap price.
                Hehe nothing of the sort. But the question came up and I didn't really know the answer.

                Best regards,
                Vlado
                I only act like I know everything, Rogers.

                Comment


                • #9
                  im also going to assume that vlado is just denying this to manage expectations, and chaosgroup actually are working on a cheap gpu with tons of ram. let me have my fantasy ok?

                  Comment


                  • #10
                    Is instancing available in the nightlies yet vlado?
                    www.nikand.com
                    3D Freelance

                    Comment


                    • #11
                      Originally posted by super gnu View Post
                      im also going to assume that vlado is just denying this to manage expectations, and chaosgroup actually are working on a cheap gpu with tons of ram. let me have my fantasy ok?
                      Heck yeah...anyone but nVidia!

                      -Alan

                      Comment


                      • #12
                        This is exactly what I asked on the nVidia site, and I was told that it's the software vendor that can tell you that because only they know how their software is coded (that means you, Vlado!). My problem is that a Quadro5000 2.5GB isn't enough with our scenes, but is 4GB? 6? 12? I don't know, so I was hoping there would be a formula for polys x textures x misc = VRAM required. Our scenes are ~ 15million polys (static, no proxys or other plugins) * 2GB textures * misc (lights, glossy materials, whatever may be a factor that I'm not aware of) = VRAM required for VrayRT-GPU. Any way to do that? Thanks.

                        Comment


                        • #13
                          This is probably common knowledge among many here, but if I send a scene to RT, will it load all hidden layers and their textures as well? I assume it will since they can be switched on and off, but is there a way to prevent this?

                          Comment


                          • #14
                            Originally posted by Nicinus View Post
                            This is probably common knowledge among many here, but if I send a scene to RT, will it load all hidden layers and their textures as well?
                            No, they will not be sent to V-Ray RT. If something is made visible later on, it will be transferred additionally.

                            Best regards,
                            Vlado
                            I only act like I know everything, Rogers.

                            Comment


                            • #15
                              Thanks Vlado. What would happen if that then makes the scene to large for my graphics board memory? Will RT crash or give some form of out of memory error?

                              Comment

                              Working...
                              X