The iray FAQ

Image courtesy Jeff Patton . Vehicle model from Chrysler The Autodesk® Subscription Advantage Pack for Autodesk® 3ds Max® 2011 & Autode...

Image courtesy Jeff Patton. Vehicle model from Chrysler

The Autodesk® Subscription Advantage Pack for Autodesk® 3ds Max® 2011 & Autodesk® 3ds Max® Design 2011 software integrates the latest state-of-the-art technology from NVIDIA and Mental Images who are working with Autodesk to revolutionize rendering and simulation workflows. This exclusive Autodesk Subscription only release features one of the world’s first physically accurate “point-and-shoot” renderers, the iray® rendering technology from mental images. Creating realistic images has never been easier with 3ds Max, using iray eliminates the need to spend countless hours tweaking sample settings and other tedious render settings

Video Introduction to Iray

From Shane Griffith's blog post Here is Iray FAQ

* Does 3ds Max use both CPU & GPU hardware while rendering with iray?

Yes – iray uses all the CPUs and CUDA-capable GPUs it finds within your system simultaneously.

* What graphics hardware is required to use iray’s GPU acceleration?

The iray renderer within the Subscription Advantage Pack for 3ds Max 2011 takes advantage of the NVIDIA CUDA architecture and gains considerable acceleration from any CUDA-capable NVIDIA GPU. This allows iray to gain acceleration from over 250million GPUs shipped over the past 4 years.





* What are the performance differences for iray between NVIDIA’s GeForce, Quadro, and Tesla products in 3ds Max?

NVIDIA GPUs vary in their compute performance according to the number of CUDA cores they have, the GPU’s clock speed, and the GPU’s architecture. The GPU brand will not impact iray performance beyond these characteristics. Put into formulae, GPU ray tracing performance looks like this:

(CUDA Cores#) x (Clock Speed) x (GPU Architecture) = Relative Performance.

This means that for a given GPU architecture and clock speed, iray performance will scale linearly with the number of CUDA cores present on the GPU, and for a given architecture and core count iray will scale linearly with clock speed. The GPU family (Quadro, Tesla or GeForce) does not have additional influence on how iray uses it.

Note, an exhaustive list of GPU characteristics can be found here


* What are the advantages of Quadro and Telsa boards over GeForce cards?

At a high level, GeForce products are designed for the ultimate gaming performance, Tesla is designed for the best compute performance and capabilities, and Quadro is a superset of both the best in graphics and compute. While each GPU family has different features, for GPU ray tracing the primary difference is in reliability and maximum scene size. NVIDIA oversees the manufacturing of its professional products (Quadro and Tesla), ensuring each product is available for three years and warranted for three years. Consumer products (GeForce) are manufactured by numerous companies and have much less availability and warranty. NVIDIA’s professional products may have slower clock rates than their comparable consumer products to fit within the power constraints and reliability needs of workstations and server rooms. Quadro and Tesla products also offer much more memory than GeForce products, allowing far larger 3ds Max scenes to be accelerated by the GPU. NVIDIA only warranties Tesla and Quadro products for running in a server room environment, where non-stop computing is typical – as with a dedicated render farm.

While it’s much more affordable to outfit your workstation with a GeForce card, in the long term you may have to replace that hardware more frequently and there might be risk of a major meltdown if you don't take extra precautions. With iray renderings you will be putting the hardware under unusually long strains that the card is not equipped to handle.


* Do I need an SLI certified motherboard to use multiple GPUs for iray rendering?

No – SLI is not needed to use multiple GPUs for iray. Iray will use whatever GPUs it sees within the system, regardless of whether the motherboard is SLI certified or the boards are placed in SLI mode.


* What NVIDIA Driver settings should I make to ensure iray runs as well as possible?

For best iray performance, please ensure the following are set within the NVIDIA Control Panel:
- SLI should be OFF (for multi-GPU systems)
- ECC should be OFF (for Quadro and Tesla GPUs based on “Fermi”)
- PhysX should be set to CPU


* Do NVIDIA GPUs support Direct3D and OpenGL in 3ds Max?

Both GeForce and Quadro boards support Direct3D and OpenGL, although the drivers may behave somewhat differently as GeForce drivers are tuned for Games performance while Quadro drivers are tuned for Tools performance. Direct3D is the most feature rich graphics API for 3ds Max, and Quadro products include a 3ds Max Performance Driver for additional performance when using Direct3D. In contrast, Tesla boards have minimal graphics API support and are not recommended for driving graphics applications. Tesla products based on the Fermi architecture have DVI ports, but this is solely to make it easy to load the NVIDIA driver (earlier Tesla products had no video out).


* What NVIDIA GPU rendering options are available for data centers?

NVIDIA’s Tesla products are designed for server rooms, with the “M” (Module) series being add-in boards for 1u systems which provide their own active cooling. Solutions vendors can be found here. NVIDIA also provides some “S” (Server) solutions that are 1u systems that connect to a CPU unit via a PIC Express cable. The S2050 (Fermi class) is the latest product here, but it’s also likely the last, as NVIDIA has found third parties are better equipped to provide such solutions from Tesla M class products.

* Can I setup data center GPUs for use with workstations.

No. Tesla M products are passively cooled and do not include fans, so they will not operate in a PC or workstation. Conversely, there’s no issue using Tesla C products (which are actively cooled with their own fans) in a dedicated server.


* What is RealityServer and is it required for data center rendering?

RealityServer is a platform product for Web developers to create server-based applications that employ 3D Web services that enable Web clients to interact with 3D data remotely. RealityServer provides numerous rendering modes for processing server-side data, with its iray mode being compatible with the iray renderer within 3ds Max. This commonality allows a RealityServer application to accept 3ds Max scenes (prepared for iray) and deliver a result that is identical to what a 3ds Max session renders. While RealityServer provides a unique and powerful option for processing 3ds Max content remotely, it is in no way required for 3ds Max users to render with iray across their LAN using Autodesk Backburner.



* I’m using just one NVIDIA graphics board – can it be used for iray as well as my viewport graphics?

Yes. The same GPU powering 3ds Max can also be used to accelerate iray. When this is done, you may find your Windows GUI interaction is sluggish while iray is rendering.


* What options are available for NVIDIA multi-GPUs and 3ds Max?

Iray will automatically take advantage of all CUDA-capable GPUs within the system – as it does with CPUs. Iray will run well on all classes of NVIDIA GPUs – GeForce, Quadro, and Tesla.


* How does the performance scale with NVIDIA multi-GPU cards (are two cards twice as fast as one)?

Performance for iray should scale nearly linearly between the same GPUs – as it is across CPUs.


* Does NVIDIA GPU cards accelerate Distributed Bucket Rendering (DBR) or support parallel rendering in 3ds Max?

No. iray is not currently compatible with DBR within 3ds Max.


* Will iray work with 3ds Max BackBurner network rendering?

Yes - GPUs on networked machines will be used in the same way that CPUs are when called upon to render individual frames.


* How do I enable GPU rendering in 3ds Max?

Iray automatically leverages any CUDA-capable GPU it finds within the system, along with any CPU. There is no user interface for enabling this (or tuning it off) – it just happens.


* What are the performance differences with 3ds Max CPU rendering verses iray GPU rendering?

This will vary between the GPUs and CPUs being compared, but between newly purchased, top-end processors, a Tesla GPU is rougly +6X faster than a single quad core CPU.


* How can I tell how much performance I’m getting from either the CPU or GPU if iray is always using both?

3ds Max 2011 doesn’t provide a user interface for controlling how iray uses CPUs and/or GPUs. There are MaxScript commands that can be used to control which processers are used.

- To disable GPU mode you need to run the following string option in the listener:
mental_ray_string_options.addoption “iray devices” “”

- To turn it back on:
mental_ray_string_options.addoption “iray devices” 0

- To turn off CPU and use GPU exclusively:
mental_ray_string_options.addoption “iray threads” 0


* How much GPU memory is needed for 3ds Max iray rendering?

The entire scene must fit into GPU graphics memory in order for iray to take advantage of GPU acceleration. If the scene does not fit unto into a GPU’s memory, that GPU is ignored and all processing takes place on the CPU and in RAM and run much more slowly. If the scene exceeds RAM, then iray will use virtual memory and run more slowly yet. For estimating memory usage, budget about 1 GB per 8 million triangles, to which you must also add 3 bytes/pixel for any referenced bitmaps.


* Do NVIDIA GPU cards also use system memory for iray rendering?

3ds Max uses system memory for processing the scene, and iray uses it for utilizing the CPU. When iray is rendering, the CPU and GPU will be consuming a similar amount of memory in graphics and system memory (respectively).


* If I have three NVIDIA cards in my workstation that each have 1GB of video memory can I render scenes that use up to 3GB of memory?

The entire scene must fit entirely into GPU memory on each GPU in order to use that GPU for processing. In this example, the maximum scene size is 1GB, and scenes over 1GB would be rendering exclusively on the CPU.


* What if I render a 2GB scene and have one GPU that has 1GB and another that has 3GB?

In this case, iray will ignore the 1GB GPU and only utilize the 3GB GPU (along with the CPUs)


* Do NVIDIA GPU cards support ECC memory?

Yes, the latest Quadro and Tesla “Fermi” generation boards support ECC, but using it will not increase accuracy or performance for iray. It is recommended that ECC is always turned OFF when rendering with iray.


* How can I view my NVIDIA GPU load and GPU memory usage?

There are third party utilities for this such as: http://www.evga.com/precision/


* What companies offer NVIDIA Tesla or Quadro configured workstations?

Most vendors have preconfigured options, including HP, Dell, Lenovo, PNY, BOXX, etc.
For Quadro solutions see: http://www.nvidia.com/object/workstation_wheretobuy.html
For Tesla solutions see: http://www.nvidia.com/object/tesla_wtb.html


* Can I build my own multi-Tesla computer?

Yes. Telsa “C” (card) products are similar to standard graphics boards and are easy to add to standard PC’s. Tesla “M” (module) products are designed for 1u heights in rack mounted systems that provide their own cooling. A motherboard layout for accommodating the number of full length boards you wish to include, and an adequate power supply for powering them, will be two of your primary design choices.


* Can I mix Quadro, Telsa and GeForce boards? If so, what driver do I use?

Yes. For drivers, you should load the one driving your Windows display.


* Can I mix GPUs of different generations and memory size?

Yes, iray supports the use of different generation GPUs.


* What do you recommend for optimal iray performance?

For workstations, a Quadro graphics board and as many Tesla “C” boards as can fit in the box will give you the most reliable, high performance blend of interactive graphics and GPU ray tracing performance while also accommodating the largest possible scenes.

For dedicated render servers, the Tesla “C” or “M” products deliver the most reliable, high performance for GPU ray tracing. Quadro products can be used here too, but will not provide added benefit over Tesla unless they are also serving Direct3D or OpenGL graphics in applications other than iray rendering. GeForce products are not warranted for data center use.

For portable rendering, there will soon be laptops with mobile GPUs based on the Fermi architecture that will provide very reasonable performance (the equivalent of several quad-core CPUs). Because NVIDIA never made a GT200 mobile part, these new laptops will be many times faster than any laptop from earlier this summer.


* When building a system for iray performance, is it best to add an additional CPU or GPU?

Given that a new GPU is worth numerous quad-core CPUs, it’s far more cost effective to add an additional GPU rather than a CPU for increasing iray rendering speed.


* How much slower does iray run when my scene exceeds graphics memory?

When your processed scene size exceeds the memory on your GPU, iray automatically switches to only using the CPU, and will run at the speed your CPU(s) can process.


* How much slower is a GPU placed in Gen1 x8 PCI Express slots versus in a Gen2 x16 PCI Express slot?

For operations that are constantly updated, like viewport interactive graphics, Gen1 is much slower than Gen 2. For compute-intensive operations that keep the GPU very busy before sending its results across the bus there is nearly no difference between the two PCIX types. In the case of iray and GPU ray tracing, you might see a GPU render faster on a Gen2 slot when the scene is very simple, but nearly no difference on a Gen1 slot when the scene is reasonably complex.


* Can a render farm mix the use of GPUs and CPUs – some using only CPUs, some with GPUs?

Yes. Iray delivers the exact same result on GPUs and CPUs and so can be used in highly heterogeneous GPU & CPU environments.


* Can a render farm mix GPU generations? e.g., some using Quadro FX 5800 (GT200 generation) and others using Quadro 6000’s (GF100 generation)

Yes. Iray delivers the exact same result across GPU generations, and can be used in farms having different GPU generations.


* Does iray render depth of field and motion blur effects?

Yes - iray can render depth of field effects
No - iray does not render motion blur effects at this time


* Can I render animation sequences with iray?

Yes. you can set iray to render the same number of iterations per frame that will create predictive results frame to frame. You can also specify time spent rendering per frame if you need to get results within a fixed time period. For example: if you have a 2 minute animation sequence to render overnight (12 hours), then you can set iray to render 12 seconds per frame. If you use BackBurner to distribute the frames across several machines you can then increase the time per frame to get better results.


More infomations


Related links:
autodesk.com