Ray Casting & Memory Allocation in V-Ray 1.5

Nov 22, 2020 Update
Tutorial on Ray casting and how to prevent memory allocation errors with virtual memory in V-Ray.


This is the last tutorial we have planned for this series. It explains the main technical problem that any professional 3d visualization artist encounter with at leas once, ie the almost unlimited greed of any 3D renderer to consumption of RAM. As well as ways to solve it by toolset available in V-Ray. This will allow you to complete the rendering of a very complex scene, no matter what, but of course with some trade-off in speed of rendering.

Together with that, all the basic settings of the V-Ray:: System rollout will be explained, especially those related to data and memory management.

Rendering Process Crash Problems

So, anyone who makes very complex scenes, interiors with classic furniture, stucco or exteriors with lots of vegetation, heavy traffic, etc., already knows what will happen if there is not enough RAM during rendering. The entire 3D application can simply crash, losing all computed data. This is a very annoying "thing" that can immediately cause frustration and disappointment, even in very psychologically strong and calm people. IThis can make most V-Ray users nervous, it can even completely disarm someone, killing any enthusiasm to continue any further work on the current project. This "surprise" is especially unexpected when it is a commercial work with a clearly defined and concise term.

Why is this happening? What causes this error? Can’t you do anything simple? Can it really be true that to complete a big project we inevitably need to increase the amount of RAM, installed in computer? Is it true, that the only way out is to go to the computer store and buy a few more sticks of RAM?

Real photo of installing additional RAM stick to the PC motherboard.

To answer these questions, let's look at how V-Ray uses memory and what management tools are available.

In any software, memory management and the way data is organized and structured is a very complex and routine topic that goes deep into programming. Well, at least it is a true for 3d artists. There is absolutely no sense to go deep into describing the data arrays and the ways of addressing to them. But, it's good to have a general idea of ​​how things works to manage the memory settings available to the user. Especially considering that the user is completely limited to the proposed interface options. Fortunately, V-Ray expose to us some of its system settings to manage.

Raycasting Basis of Rendering

Before start the rendering, V-Ray must load the scene geometry into RAM for operative using it during calculation of the final image. This is the initial aspect of the entire rendering process. After realizing such necessity of renderer, the importance of memory management in the rendering immediately becomes clear. But, of course, we wouldn’t just stop on this and will follow the order of rendering step by step.

Once data is loaded into memory, V-Ray rendering can begin. The calculating of each pixel starts with the basic operation - "probing" of scene geometry. This is done by using a special algorithm, the so-called ray casting. The essence of this algorithm is that the camera emits "rays" for each pixel of rendered image, which (rays) fly into the space of the scene until they intersect (hit) the geometry of an object in the scene to determine exactly what scene geometry is actually visible from this camera. Each hit of the ray is recorded, and thus the render engine generates data on the location of the geometry in the scene, its basic properties and other information necessary for further rendering.

Disadvantages of Unorganized Data

The process of calculation of the single raycast, i.e. the ray intersection with geometry and the determining of its properties, is not time significant. However, the number of raycasts during photorealistic rendering can exceed manyfold the number of rendered image pixels. For example, the HD 1920x1080 resolution render is made of more than 2 million separate pixels. This scale changes the situation fundamentally, making the raycasting process very resource-intensive.

Let us take a formal analogy to the technically complicated raycasting algorithm. This analogy will greatly help us understand the specific problems of memory management in V-Ray.

So, let’s imagine the scene geometry as a set of usual green tree leaves, and the raycasting algorithm as a tireless ant, which have to find the most delicious yellowish leave amongst all green ones. In our analogy, this yellow leaf will embody the needed data about the geometry, hit by a raycasting ray.

Initially, the data on the scene geometry, which will be loaded into memory, is arranged randomly in terms of raycasting process. Let us similarly suppose that all the leaves were chaotically thrown into pile.

Our hard working ant knows that he needs to find a yellow leaf among the myriad of unnecessary for him green ones. How can he do this? Of course, he will have to go sequentially through a bunch of green leaves, starting from the closest to one, until he find the one and only yellow leaf.

Allegory showing disorganized data as a pile of leaves, and a raycaster as an ant looking for certain leaf in a pile.

He will have to approach first leaf and see whether it is yellow. Then, come up to the second and look if he needs the second leaf. After this, he has to go to the third one, and again check if that is yellow. Thereby, our ant needs to go through literally every leaf until he finds the desired yellow one.

The situation when the ant reaches the yellowish leaf at the beginning of his search is probable. However, there is also absolutely no guarantee that the yellow leaf not be the latest in pile. In such case, to find a single yellow leaf, our ant would have to go over the whole pile of greens.

A similar situation can occur with unorganized data on the scene geometry. Raycasting algorithm will be forced to sequentially go through all the geometry for determining the properties of a single point in the scene, which was intersected by raycasting ray.

This method of finding the required information is too tedious for our imaginary ant, as well as for the computer’s CPU, which instead of actual rendering has to busy oneself with non-rational searching of data in memory.

Convenience of Data Structured in BSP Tree

To expedite the process of determining the properties of the scene geometry by raycasting, V-Ray organizes all the geometry into a special convenient structure.

V-Ray lays out the geometry on the "shelves" of the hierarchical tree structure into RAM. This structure is called the Binary Space Partitioning tree or simply BSP tree. It is a special method of presenting an array of data as a binary tree structure.

The BSP general idea is that scene with all geometry is subdivided into two parts. This action creates the two branches extending from the origin (the scene), which is called the Root of the tree. After the first subdivision, each of the data branches emanating from the root, are again divided in half, forming new branches. Points from which the new branches emanate called the Nodes of the tree. The process of dividing the geometry of the scene and the record the tree structure branches in memory lasts as long as it reached the limit, which is set in the BSP tree parameters. The result of the last division of branches is called the Leaves of the tree. In that way, V-Ray creates the branched structure of the whole scene geometry, which is great for the quick search of the properties of the required area in the scene.

For the obvious presentation of how this happens, let's go back to our entomological example. The analogy with leaves and leaf-seeking ant perfectly illustrates how the initial organization helps in finding the needed data.

Allegory showing data structured in BSP Tree as a tree, and a raycaster like an ant that purposefully crawls to certain leaf.

If we do not throw the leaves to a disordered heap, but carefully arrange them on the branches of a tree, we can significantly speed up the process of finding a yellow leaf. In such situation, our ant, avoiding all unnecessary branches, can go straight to the needed destination. That happens because he does not need to check all the leaves. Instead, he can almost immediately see the path that could lead him to the place where the yellow leaf grows, and quickly climb there by his strong legs.

According to this principle, V-Ray structures the geometry in the scene, catching the one more opportunity to save our precious time, even on such seemingly insignificant operations.

The basic principle of rendering at this level is quite simple: load, structure, found the necessary properties, and render :)

Now that we are familiar with the principle of data structuring for raycaster, we can move on to the next step in understanding of how to render a heavy scenes with V-Ray.

Static Raycaster or Loading Entire Scene Geometry

As we already know, to start the rendering process the scene geometry must be loaded into memory. In the previous chapter, we looked at the tree structure, the accelerator of raycasting process.

However, as we mentioned at the start of the lesson, there is one rendering issue left. Everything goes smoothly during rendering, as long as the RAM amount, physically installed on your computer, is enough to contain the whole scene geometry.

The fact is the V-Ray has different types of raycasting algorithms. The simplest of them is the so-called Static Raycaster. It loads the geometry of entire scene into memory at once.

Imagine that we are going to start rendering a scene with many millions of polygons, so for its entire loading the available RAM amount is not enough. What happens then? Unfortunately, in this case the one of the most unpleasant V-Ray surprises occurs. It is the program crash.

A Software crash is the unexpected abnormal termination of the program due some critical error that software can not handle. In V-Ray (actually in any software), this happens when an available RAM is already depleted, but V-Ray is still trying to allocate more to load all the geometry into memory. Because this is physically impossible (no more free RAM available), further allocation causes an error and crash. The message about creating in memory the geometry tree structure «Current Task: Building static raycast accelerator...» is usually the last thing you can see before the failure. After that, the rendering is interrupted, 3ds Max spontaneously crashes, and system is showing you an error message box.

For example, by our analogy, it is an attempt to grow too many branches with leaves than a tree root can withstand. Then the crown of the tree becomes too heavy and eventually uproots a whole tree from the ground.

The allegory shows a program failure caused by a memory allocation error like a fallen tree.

The only way out of this situation without increasing the amount of RAM is to abandon the one-time entire scene loading approach. Instead, we need to partially load the geometry into memory, that is, only a certain part of it, which is currently needed to render the current part of the image at the current time. Not the whole geometry of the entire scene in advance.

As might be expected, V-Ray has such a way to load the geometry into the memory. This method is embodied in a special, dynamic type of raycaster, which is designed exactly for such situations.

Dynamic Raycaster or Loading Geometry in Portions

In contrast to the basic raycasting engine called Static, the Dynamic Raycaster allows us to load only a certain part of the geometry, needed for rendering of the current image part.

Dynamic raycaster loads into the memory only a certain portion of the geometry, not exceeding a specified number of megabytes. Once the loaded portion has been used in rendering of the current image area, the occupied part of the RAM is cleared and the new piece of geometry loads to the freed place. Obviously, the new geometry portion will be used to render the following part of the image.

Rendering using dynamic raycaster eliminates the memory overflow problems not only due to the exact memory limit. Dynamic raycaster knows how to use the operating system’s potential in memory management. This means that even if the current image area calculation needs to load objects, the total size of which exceeds the established memory limit, the operating system still is able to allocate a resource for the successful continuation of the rendering. In particular, Windows use so-called paging (or swapping), which is the ability to "virtualize" memory by storing data not into memory, but on a drive.

Despite the aforementioned advantages of dynamic geometry loading, we must realize that if dynamic raycaster was the really universal solution, then an unreliable crashing static raycaster would be simply needless. However, V-Ray uses the static raycasting engine by default.

Not surprisingly, versatility is not without costs. The loading and unloading of the data spends our precious time, as any other operation. Even if we set the ideal (for our situation) RAM limit, the use of dynamic raycaster will significantly increase the overall rendering time. Moreover, if paging will be used, it will slow down rendering even more.

This is the whole basic theory of raycasting that you really need to know to work with V-Ray system settings. Now it's time to consider the actual settings we have at our disposal.

Raycaster Params: Binary Tree Settings

The first three parameters describe the properties of the raycasting acceleration structure, the BSP tree. They are quite abstract, so let’s begin with determining the effect of each of them.

Max tree depth: is a parameter that limits the depth of the branches of a tree. It specifies the number of divisions of the scene space. The numeric value of this parameter reflects the maximum number of levels from the root to the leaf. The higher the Max tree depth: value, the more branches the tree will have.

Min. leaf size: limits the structure of a binary tree by a minimum size of the leaves. This option allows us to stop the division of the geometry in the scene by setting the value of the minimum leaves size in system units. Small ​​Min leaf values size let not to limit the branching of the tree, and thus it grows as much as the previous parameter defines.

Face/level coef. determines the maximum number of triangles (minimal units of geometry) contained in a single leaf node. Smaller values of Face/level coef. can produce more branched tree, with lots of leaves.

As we know, V-Ray uses the organization of the scene geometry in a form of a BSP tree to accelerate the raycasting process. It is logical to assume that the deeper our tree structure, the smaller its leaves, and the smaller the units of the geometry each leaf contains, then the more space in memory we need to hold tree’s branches and the greater is the potential raycasting speed it can provide.

However, there is a limit in such acceleration. After a certain threshold, further division is not reasonable, since the excessively branched structure of a BSP tree makes the work with itself too difficult. The raycasting algorithm may spend the computational resources for walking through the huge BSP tree, ruining all the gain in speed.

So, what is this limit and what is the best BSP tree size for getting the maximum raycasting speed?

The answer for this question is not very encouraging. Only experimentation work can help in finding the optimal BSP tree parameters. There are no fast settings for any scene. For best raycasting speed, the each specific scene must have the individual BSP tree settings.

When rendering still images, the messing with numerous experiments on the above parameters is completely unreasonable. Just because the spending lots of time for saving the same time is nonsense. You may agree that it is an absurd to make 10 times the final rendering in the search for BSP tree optimal settings, to eventually make the same rendering the eleventh time 10 minutes faster.

However, the parameters of a BSP tree are not just a demonstration of technical abilities of the rendering engine. Their purpose becomes clear when it comes to animation. Suppose we need to render sequence of 1000 frames. Then we can safely perform an experimental selection of optimal BSP tree parameters. Even if the choice of the best values will take up to 10 or 20 final renders, and the rendering speed of each frame will increase by 20 seconds, such effort is very useful. After all, this optimization can easily save about 5 hours of computing time.

In practice of static rendering, the need to change the values ​​of these parameters exists only in theory. It can only make sense when we want to free memory to suit some other needs, so we reduce the size of the BSP tree at the price of slowing the rendering time.

In the best V-Ray settings, the parameters responsible for the BSP tree structure should be set to the default V-Ray values, without making any changes without well-considered necessity.

Raycaster Params: Memory Limit & Geometry Type

Dynamic memory limit: is parameter that determines the limit of memory available for dynamic raycaster, in megabytes. It defines the maximum amount of RAM to be used for every geometry portion during the work of the dynamic raycaster. Literally, established in this parameter number of megabytes is divided between computational threads of the current computer. Low values ​​of Dynamic memory limit: can significantly reduce the raycaster performance; they may lead to the frequency increase of loading-unloading the geometry portions. Additionally, small values can cause the regular use of slow paging. From the other hand, the very high values ​​may slightly slow down the renderer and greatly affect the comfort of work in operating system. The value of 60-80% of the total amount of RAM, installed on your computer, gives the best performance. That value we should use in universal V-Ray settings.

Default geometry: parameter determines the method of loading the geometry into memory. Here we can select the type of raycaster, which will calculate the geometry of the current scene. The Default geometry: parameter is a drop-down list from which you can choose the type of raycaster.

Before describing the Default geometry: options, we should mention the VRayProxy and VRayFur objects. The feature of rendering such objects is that no matter what is selected in Default geometry:; these exact objects are calculated by dynamic raycaster. Such a feature is implemented because these like objects are extra high polygonal, and it is better to load them by portions for avoiding the program crash.

Static is the most trivial raycasting mode. When we select Static, all the data about the scene geometry is entirely loaded into memory before rendering.

Dynamic is the compromise mode of Default geometry:. Dynamic raycaster does not load all the scene geometry into memory at once. It loads the geometry data by portions. The number of megabytes set in Dynamic memory limit: determines the maximum portion size.

Auto is the initial Default geometry: option. In this mode, V-Ray decides which objects in the scene are better to calculate with static raycaster, and which are with dynamic. This decision is based on the properties of the particular object. Whether this object should be calculated with Dynamic or Static raycaster depends on the polygons and instances number of this exact object. This is a hybrid method of loading geometry, which combines the work of Static and Dynamic raycasters.

If we compare the types of raycasters, then we can conclude that dynamic is the most unpretentious, but also the slowest raycaster. Using it, we are free from the 3ds Max memory overflow issues, however, due to the need to fill and clean the limited amount of memory continually, rendering speed substantially slows down. When rendering with static raycaster, it on its one manages all RAM resource, available to the system. Static is the most productive, but also the most unstable mode. If all the geometry can completely fit in memory, then the rendering is as fast as possible. After all, we do not need to dose the geometry, wait for clearing the memory to load the next portion of data, and use the sluggish paging. However, if we have more data than can fit in available RAM, it will immediately cause the downfall of the program. Just recall the picture with a broken tree.

Nominally, the Default geometry: set to Auto. The Auto is the most universal mode, so it is reasonable to use it in the Best V-Ray Settings.

Buckets Options

The bucket is a rectangular section or portion of ​​a certain size into which the rendered image is divided. This is a special technique by which the work of calculating the whole image is divided into parts. This allows a 3d renderer to "submit" individual portions of the task to a separate computing process (processor cores) and calculate several portions of the same image in parallel. This is the main "trick" of how render engines implement multi-threaded visualization. The idea of such a partitioning is similar to the idea of how Number of passes function work in Light Cache. Buckets are those "little squares that quickly (and sometimes not...) ​​running around" the frame buffer, gradually opening the final result of rendering to our view.

It may sound banal, but the actual rendering process, which is considered a model of multitasking, is not multi-tasked actually. The exact computation cannot be performed by more than one core, even with the multi-core CPU. You can imagine how long will the rendering last, if we use only a single processing core, while other are idle. This situation is completely unacceptable and, of course, we need a way out it. The solution for this is the equal division of the rendering process between different computing threads. The rendering is divided into many independent parts, in other words, into many small renderers, and the each of them can be computed by a separate processing thread. Thus, uniform computation is divided into several parts, allowing the full load of all computing pipelines of the system. This is how the little images, called buckets, render. When the bucket is rendered, it is displayed in a frame buffer. The rendering result is like a puzzle, assembled from small pieces into one final image :) In order for the final render does not look like a jigsaw puzzle or a mosaic, with clearly distinguished borders of neighboring cells, buckets are the subject for mutual interpolation.

Let us take a look at the exact buckets parameters.

X and Y values ​​determine the size of the bucket in pixels horizontally and vertically. By default, these two numeric fields are dependent and equal. Although, it is enough to press the button L (lock) and the independent Y value becomes available. The X and Y determine the size of bucket in pixels until the drop-down list next to X is set to Region W/H. If it set to Region Count, then the numerical values ​​in the X and Y determine the number of buckets, to which the final render will be divided horizontally and vertically.

Obviously, if the final image is divided into many portions, then the more computing resources V-Ray have to spend on interpolating their neighboring boundaries. At the same time, the small bucket is a guarantee that one rendering portion catches the minimum amount of geometry. That is good factor, because bucket is exactly that portion, which the dynamic raycaster loads into RAM. Reducing the size of the bucket can be useful, when filling the bucket objects exceed the limit of memory, available to a single thread. A smaller bucket could theoretically take less amount of geometry, so it can possibly save rendering from the use of paging. The long and short of it, this situation is difficult to predict, so the default values ​​64x64 and Region W/H are the best for starting V-Ray configuration.

Region sequence is the option that specifies the type of portions rendering sequence or, in other words, the pattern of buckets rendering. We can choose the one of the six rendering patterns - Top/bottom, Left/right, Checker, Spiral, Hilbert curve, and Triangulation.

From the name of the first four of them it is quite to guess a way they render. With the Hilbert curve mode, the buckets follow each other in a form of a snake out of blocks, just like in game on first mobile phones.

The last and the most interesting mode is Triangulation. With it, buckets plot the triangles of different sizes, consistently covering all areas of the frame buffer. The main feature of Triangulation is that its pattern gradually covers very dense image areas, what is very useful for dynamic raycaster work. With such a pattern, there is a high chance that the part of geometry used for rendering previous bucket, can be used for rendering the current one. This certainly saves computational resources, avoiding unnecessary operations of loading and unloading of data to limited memory amount. It is for this reason, even if you like a snake or a checker pattern, in starting V-Ray settings always use a Triangulation as a pattern for rendering buckets.

Reverse sequence can reverse the direction of buckets calculation. For example, if the rendering pattern is set to Left/right, the activation of Reverse sequence option will make the buckets computed from right to left, i.e. in the reverse direction. This feature does not really affect the rendering process. However, it can be very useful in cases when rendering of the scene stopped before completion for some reason. Then the rendering can be resumed by activating the reverse direction so we can continue to render the scene from the not-rendered part of the image. In the Best V-Ray settings, the changing of this option has no practical sense.

Previous render is an parameter to specify an action performed with the previous image into a frame buffer during rendering of the new one. The Unchanged mode allows us leave the previous rendering unchanged. Cross and Fields modes let us cover the previous render with the corresponding pattern. Darken and Blue modes allow us to darken or give a blue tint to the previous renderer for highlighting the contrast. The most useful of these modes is the default Unchanged, which allows us easily watch the changes in the new render, comparing to the previous. That is the mode we should use in the optimal V-Ray configuration.

On this, all the important system configurations of V-Ray end. Following zones and setups of V-Ray:: System rollout are secondary and do not significantly affect the actual rendering. However, we will surely learn how to use them also.

Frame Stamp Statistics

Frame stamp enables adding the stamp with statistical information to the image of final rendering. With this feature on, we can add to a render a stripe with different technical information. This info can be the current frame rendering time, the name of the CPU, the current version of V-Ray, and even the size of the final image in pixels. This particular feature is designed for special occasions. For example, it may be useful when testing the scene for determining the optimal size of a BSP tree. Enabling Frame Stamp also may help you to boast before colleagues the new computer and the speed it renders the most difficult scene. Of course, enabling this function is useless as for the starting V-Ray settings.

Distributed Rendering Setups

A distributed rendering of an images is one of the most revolutionary features of modern rendering engines. Distributed rendering allows us to render a single image by using a computational power of multiple computers. To do this, the computers must be properly configured and be in a common network. V-Ray can distribute and send separate buckets to different available computers, and then collect the results of their calculation in a single frame buffer.

For now, the rendering of an image by multiple computers is far from the idea of starting V-Ray settings, so we do not need to activate the Distributed rendering function.

Rendering Log

The VRay log area is fully responsible for a single function - to write and display the contents of the log file.

Unlike Frame stamp, the usefulness of this function is obvious. Enabling the Show window checkbox in the VRay log settings, opens the information window called V-Ray messages during rendering, which will show logs of V-Ray. It is very convenient to see information about the rendering process. Such information can be different warnings, errors, and just the statistics. For example, VRay log is indispensable in the process of configuring the aforementioned distributive rendering. V-Ray messages can clearly display which computers took part in the current rendering, and which are idle. Such awareness allows us to save a lot of time and prevent many problems on the initial stages of rendering.

VRay log additionally makes the records to a text file, the path to which you can specify in the box below the Show window checkbox. It can be useful if 3ds Max has suddenly crashed. In this case, we can always open the written file using any text editor and look for error messages, among which is the one that caused the crash.

A numeric value Level indicates the depth of the ongoing statistics. The one (1) in this parameter means that the log will contain only the error messages. Two (2) is the errors and warnings. Three (3) specifies that logs must display the errors, warnings and general informational messages. Four (4) in this parameter tells V-Ray to log all the events that happen during rendering.

In order to be well informed about the most important events during rendering, in the nominal starting V-Ray settings, the Show window option must be enabled and statistics Level should be set to three (3).

Miscellaneous Options

The Miscellaneous options is the last area of system settings and contains all the remaining secondary system parameters of V-Ray.

MAX-compatible ShadeContext (work in camera space) is a feature that allows V-Ray to work with so-called camera space.

The case is that V-Ray nominally performs all calculations in the scene space. In other words, the V-Ray itself uses geometry coordinates relative to the whole scene space. However, some plug-ins, especially the older ones, work with coordinates relative to the current camera. Activation of work in the camera space allows V-Ray to work with these plug-ins, bypassing the problem of incompatibility. That is why, in the starting V-Ray configuration, we must always keep the MAX-compatible ShadeContext (work in camera space) option active.

The option Check for missing files enables check for missing files, such as not found textures or proxies in the current scene. The result of this check is immediately displayed as a warning in the log. Nevertheless, we usually know about the missing files in the scene. Just remember the popping out list of missing files when the scene is loaded into 3ds Max. In practice, the missing files check before actual rendering is more annoying than useful, and therefore, we are better turn off this function when setting the starting configuration of V-Ray.

Optimized atmospheric evaluation is an option that disables shading on the scene objects that are inside or outside the zone of atmospheric effects. When the option enabled, renderer does not shade the objects, which are slightly visible through the opacity of atmospheric effects. The enabling of optimized atmospheric evaluation can save significant computational resources on skipping materials calculation for barely visible objects. In such case, V-Ray first calculates the atmospheric effect and estimates the visibility of objects in advance. Since that evaluation, the non-visible objects are treated like those that do not need the shading. Of course, the automatic determination of the objects visibility through atmospheric effect is not always perfect. There may be a situation, in which the object actually is visible enough, but still not shaded. In the Best V-Ray Settings, this function should nominally be disabled.

Low thread priority is an option that automatically assigns the low priority to the rendering process in the operating system. The same can be manually done using the Windows Task Manager on the Processes tab. Low thread priority option use can relieve the computer a bit, freeing up computing resources for other applications. It allows us to work with those comfortably, even during rendering. However, resource-intensive applications can still lag despite the activation of the Low thread priority.

Additionally, assigning the low priority to any process reduces its effectiveness and, as a result, slows down the computation speed, which is very unacceptable in most cases. Therefore, in starting V-Ray settings, this option must be turned off.

Last part of Miscellaneous options section are the Objects settings, Light settings, and Presets buttons.

Objects settings and Light settings buttons call the Options windows of objects and light sources settings, similar to those that can be accessed by clicking the right mouse button on the selected objects in a viewport. These features are not actually the V-Ray settings, so they have no relation to the optimal V-Ray settings.

Presets is a function that allows us to save the settings of the current scene to the V-Ray profile and its further loading into any other scene. Using this feature can leave us from routine of setting the same checkboxes and values from project to project. This function is directly related to the nominal starting V-Ray setup. If we configure V-Ray and save its setting to profile, then we can easily use them in any new scene.

This function is obvious even without further explanation, but, in any case, its practical application and its alternatives will be described next tutorial in detail.

GI Maps Out of Control

It should be noted that V-Ray:: System rollout settings do not affect a memory allocation for GI data in RAM. This means that regardless of what the BSP tree size is, what kind of raycaster we have selected, and what the memory limit is, the GI maps will be fully loaded into memory and will occupy there as much space as they need.

It is a real situation, when the render passed all preparatory stages, associated with loading geometry into memory, and even started GI maps rendering process, Light Cache for instance. Then, after a certain number of LC samples have been rendered, you have the same hateful memory error, because the memory that left after loading the geometry is simply not enough to meet all the needs of LC computing.

Unfortunately, there are no system settings to solve the problem of memory overflow during GI computing. However, some way out exists. The most obvious is to use a dynamic raycaster to limit memory usage by reducing the value of Dynamic memory limit:. There is also another compromise. We can go the way of subdivision, simply breaking the rendering of one image into several parts. It is possible to render entire image using region render option. For example, we can render first half of the image using region render, and then render the remaining half. Of course, if we render Light Cache only for half of the current frame, then it will take twice as less memory than when rendering the whole image at once.


We hope that this tutorial will save you if you face a difficult situation when the project deadline is close, but the visualization process suddenly fails. Now that you understand the basics of working with V-Ray system settings, you can handle very complex scenes rendering, almost regardless of the amount of RAM in your computer.

And that's all. If you were strong enough to go through all seven tutorials with us, whether you are a beginner or a professional, you could find something useful in them, and it will help you in your current and future work.

We wish you simple settings, as well as fast and beautiful 3D visualizations.

See you in the next tutorials, but this time, probably, from another series!

Sep 19, 2011
As always excellent material. Big-ups! I wonder what the next tutorial will be about?
Sep 19, 2011
Thank you for another useful tutor. Guys, you are, as always, at the highest level! Well done!
Sep 19, 2011
Thumbs up!
Maks (RenderStuff)
Sep 19, 2011
Thank you all for all the good feedback! There are several topics for tutorials on our roadmap, and we have not yet decided which one we will write next. But I'm pretty sure it will be something rendering-related.
Sep 25, 2011
Why do you suggest using only 60-80% as a 'Dynamic memory limit' instead of 100%?
Anton (RenderStuff)
Sep 27, 2011
Hello Dan! The operating system itself and other programs that run simultaneously with the renderer (3ds Max, web browser, audio player, etc.) on your PC, in fact, also require a certain amount of RAM. If they or the renderer lack RAM, a paging will be performed that slows down everything, including the rendering process itself. Therefore, to avoid this situation, it is better to do dynamic loading / unloading of portions of gneometry more often than to do paging. Since we do not know which programs run on your PC, and we do not consider it rational to 'measure' how much RAM they consume, the range of values ​​from 60% to 80% is only an approximation. In addition, the DML value is also depends on the amount of RAM. The more RAM you have in you computer, the higher the percentage you can set. So finding a specific optimal value is up to you.
Sep 27, 2011
Please explain, does the dynamic memory limit is determine the memory allocation only for dynamic geometry (ie proxy, displacement and VRayFur)?
Sep 27, 2011
Thanks for the clarification, Anton! Now everything is in place.
Sep 27, 2011
I have a Windows 7 operating system and 8 GB of memory. I set Dynamic Memory Limit to 6200, but Task Manager does not display 3ds Max memory consumption greater than 3000. What might it be?
Maks (RenderStuff)
Sep 27, 2011

Honestly, only V-Ray developers know exactly how their software manages memory, but as we understand that dynamic memory limit only affect objects with a dynamic geometry type.

It looks like you're using a 32-bit version of 3ds Max and V-Ray, which can't use more than 2800-3000 MB of RAM due to technical limitations (maximum address in a 32-bit bus). In other words, your 5 gigabytes of RAM are not working right now. Make sure you have 64-bit Windows, and install 64-bit 3ds Max and the appropriate V-Ray, then you can use the full potential of your system.

Victor Skea
Oct 12, 2011
Thanks for the helpful info. I've been looking for V-Ray tutorials for a long time, which are written in simple language with explanations of tips and hints. I'm sick of the official docks because of the very short and formal explanations. But I'm here now, and I've set a memory limit of 14400 MB. But I do not see that it somehow affected memory consumption. Visualization uses only 25% of available memory. Although the scene contains 1.5 million of polygons and 150 types of materials, most of them use different textures.
Anton (RenderStuff)
Oct 12, 2011

First, as we wrote in this tutor, do not use dynamic memory mode without need. Use static mode by default unless 3ds Max and V-Ray do crashs.

Second, if the scene with all the polygons and materials 'weighs' less than you have RAM, the renderer will still not use more than necessary to accommodate all the data, regardless of how much free memory is left in your system.

It is also worth mentioning that Windows Task Manager does not display real or accurate RAM usage. Windows is a very complex software with a lot of optimization and complex caching, such as SuperFetch, so just use static until you have problems and do not torment yourself with memory consumption questions untill than.

Victor Skea
Oct 12, 2011
Hello Anton. The problem is that max (2010 x64) began to fall shamelessly. It may fall while I press CTRL + S instead of saving my last work. Lack of RAM may not be a source of problems, but interestingly, when I use V-Ray RT, crashes happen much less often.
Anton (RenderStuff)
Oct 12, 2011
Saving failures are not related to the settings described in this tutorial. Navigate to the disc and measure the scene size. If the size of the .max file is much smaller than your RAM, this is definitely not a memory problem. This is more like a video driver problem. I say this not because you mentioned RT, but because I personally had similar problems when my scene had a lot of textures and just rotating it in the viewport caused 3ds Max to crash. Try installing the latest video driver, and try to merge your current scene into a new empty scene as an option. It's just a 'guessing' but maybe these manipulations will help somehow.
Victor Skea
Oct 12, 2011
Thank you very much for the help!
Oct 18, 2011
Oct 19, 2011
Thanks for this article, very interesting. This solved many questions that I always wanted to know. Regards.
Oct 19, 2011
Hello! Is this relevant for sketchup vray?
Oct 19, 2011
Hello Pedro,

Despite there are little distinctions in some settings, this probably true for all the V-Rays. The V-Ray for 3ds Max, V-Ray for Maya, V-Ray for Rhino, V-Ray for SketchUp, V-Ray for Softimage, V-Ray for Cinema 4d, and V-Ray for Blender an so on.

George M.
Nov 16, 2011
Hello RenderStuff,

I have a problem for quite a while and I hope you can aid me. When I try to render for the first time after opening a scene, the render, computer, everything goes unresponsive for a while (like 5-8 minutes) when it calculates the static raycast in V-ray. After that, everyting goes to normal and continues rendering. This happens only once after I open the scene in 3ds Max. If I close 3ds max and open the scene again it happens again. The Raycast is set by default values. If I merge the objects into another blank scene, the problem goes away but only for the first 2 or 3 saves. Can you give me perhaps some advices?

Thank you!

Nov 16, 2011
Hello George,

This is definitely a lack of RAM for the scene you are trying to render. You should either buy a few RAM sticks, or bring the optimization into your scenes creation approach. Make sure you have adequate polygon count for your 3d models, use the Instances for objects in scene wherever possible, also check the textures for excessive high resolution.

As for why it happens only the first time, it's simple. The some data filled in memory only once; all the following time it stays in memory ready to use. In addition, 3ds Max has undo history, ie it saves the state of the scene with each change, which allows you to cancel changes, to revert back to anu of saved state. All scene states are stored in RAM, so longer you use the scene, the more memory it takes, of course, no more than the s Scene Undo Levels set in the 3ds Max preferences.

George M.
Nov 16, 2011
Thank you for your prompt and efficient answer.

Well..things make more sense now. My scene is very heavy indeed. I excluded this possibility for a while because I thought 8GB RAM was enough. Now that you told me, I rechecked the RAM usage and it goes suddenly to 99% when is raycasting. I will try to do some optimization maybe some vray proxys and instances as you suggested.

I also read your other tutorials, and I have a small question regarding gamma 2.2 workflow which I use it for some time but I still have some gaps. I will post it there. I hope I'm not very annoying posting a lot of questions.

Nov 16, 2011
We recently decided to make some hardware updates and found that DDR3 memory is very available today. Only $ 50 for 8 GB of RAM.

George M., of course, ask your questions, but under corresponding tutorial.

Dec 28, 2011
Thank you, Anton. Today I bought 8 GB of RAM and am incredibly pleased. Rendering the same scene took only 15 minutes instead of 2 hours.
Dec 28, 2011
Carefully read your say about the memory access on slow hard drive. There are now SSDs, and they are much faster than conventional hard drives. So, if you have SSD and don't have enough RAM, will the render with paging be faster due to the speed of SSD?
Maks (RenderStuff)
Dec 29, 2011
Of course, all I/O operations will be faster with the SSD, including paging. But the speed of SSDs (the most expensive solid state drives now run up to 150-255 MB/s) can not even be compared with the speed of RAM (the cheapest DDR3 will gives you up to 10-30 GB/s). So, even with a SSD, swapping still slows down rendering a lot.
Dec 29, 2011
Thank you, now everything is clear, it was interesting to learn how V-Ray works. Visualization with low RAM will affect performance, of course, but the ability to render a scene under that condition, if you have no choise, is invaluable.
Jan 12, 2012
Thank you so much for the stuff you shared! Read all the tutorials in one breath. Everything is very detailed and easy to understand !!! Respect to the authors.
Rekha Maurya
Jan 25, 2012
outstanding explanation... good job
Feb 7, 2012
Starting to read the first tutorial, I couldn't stop until I read all the tutorials in this series. They are informative, greately structured! After reading them, I finally had the feeling that I know V-Ray and understand the essence of how it work! I adjusted everything as suggested in the tutorials, and rendered the old scene in just 20 minutes in decent quality, although not so long ago its rendering took me as long as 9 hours, and the quality was at the same level. In terms of gamma and contrast, the picture become much better than before, even without postwork. Also, with the Antialiasing filter turned off, I get a lot more room for further editing in Photoshop! Thank you so much for these tutorials! These are the best things I've seen while learning 3D.
Feb 7, 2012
Thanks for these Vray tutorial, help me a lot!!!
Feb 20, 2012
you're a genius!!!!............thanks.
Jun 27, 2012
Been in the game for 8 years and I have found your tutorials to be BRILLANT! Answered questions that i have been mulling over for years. The best write ups on vray i have read - period.
Jun 28, 2012
Great article, thanks !! Questions; I use SU8 + VRay Is it possible in VRay for SU to choose between static or dynamic raycasting? If so, where can I find it? It doesn't appear in the system-rollout. Regards, PJ
Antonio Neto
Jul 19, 2012
Congrats for this nice informative website. I have one question: Do you know what's the diference between the types of build tree low, avarege, high? What's the pros and cons? If the render will be faster or slow?
Aug 4, 2012
Thanks for all your tutorials, really great stuff. Could you do a tutorial showing of how to compose an image with specular, vraylighting, vray reflections and so on? Sorry my english, hope you can understand me.
Jan 10, 2013
I almost cried reading this!!! You are the best!!
Ioan Alex
Jan 23, 2013
THX !!!
Jun 16, 2013
Thanks you soooo much for this you are the best
Jan 28, 2015
Thanks for the best teaching ever
This discussion has been closed.