Test performance with Evaluation Modes

Test your scene for performance improvements with the various evaluation modes. Take note of the frame rate in each mode.

To test evaluation modes:

  1. Load your scene.
  2. In the Preferences window (Windows > Settings/Preferences > Preferences), select the Animation category (under Settings).
  3. Set the Evaluation Mode menu to DG to ensure you are in Dependency Graph (DG) mode.

    DG mode was the previous default evaluation mode.

  4. Play the scene back in DG mode and take note of the frame rate.
  5. Switch to Serial mode to ensure the scene evaluates correctly; take note of the frame rate.
    Note: Sometimes scenes run more slowly in Serial mode than in DG mode. This is because Serial mode evaluates more nodes than DG mode.
    Note: Animation using FBIK reverts to Serial Evaluation as FBIK creates missing dependencies in the Evaluation Manager. As FBIK has been deprecated, use HIK for whole-body inverse kinematics and retargeting.
  6. If the scene looks correct, switch to Parallel mode.
    Note: If the scene looked wrong in Serial mode, it is unlikely to work properly in Parallel mode, as it uses the Evaluation schedule, but distributes computation across all available cores. Occasionally, improper evaluation in Serial mode is caused by custom plug-ins that use setDependentsDirty to manage attribute dirtying.)
  7. Next, while still in Parallel mode, activate GPU override . If your scene has standard Maya deformers or your mesh's geometry is dense there may be a performance gain. Results vary based on deformers and density of meshes in your scenes.
    Note: You can use the GPU override evaluator with either Serial or Parallel Evaluation modes, but GPU override does not work in VP1.

When analyzing speed:

Troubleshoot Parallel Mode

Parallel scene evaluation takes advantage of all available computational resources to evaluate scenes more quickly. This is done by building an evaluation graph from the scene description using the dirty propagation mechanism.

For the evaluation graph to generate correct results, dependencies must be accurately expressed in the DG. If node dependencies are incorrect for your scene, parallel evaluation will produce wrong results, or Maya may crash.

For example, in a scenario where NodeA depends on NodeB and NodeC, the evaluation manager must compute the state of NodeB and NodeC prior to calculating the state of A.

A Safe Mode performs sanity tests when parallel evaluation is activated. This Safe Mode prevents the most common types of errors from invalid dependencies. If the Safe Mode detects errors, the Evaluation Manager reverts to Serial evaluation mode, as indicated in the HUD. One reason Safe Mode may revert to serial evaluation is if a node instance is evaluated by multiple threads simultaneously.

While Safe Mode catches many problems, it does not catch them all. Therefore, there is a new analysis mode to perform more thorough (and costly) checks of your scene to detect problems.

Note: Performance issues can occur when executing large amounts of Python commands. If this occurs, define MAYA_RETAIN_PYTHON_GIL in your environment. (MAYA_RELEASE_PYTHON_GIL is deprecated. It is true by default and can be overwritten by MAYA_RETAIN_PYTHON_GIL.)

Also, Python performance issues can occur when evaluating other threads with custom Python nodes.

Analysis mode

Analysis mode is not meant for use by animators when playing back or manipulating the rig. Rather, it is designed for riggers and Technical Directors to troubleshoot evaluation problems when creating new rigs.

Analysis mode:

Activate Analysis mode with the following command while in Serial mode:

dbtrace -k evalMgrGraphValid;

To disable Analysis mode, type:

dbtrace -k evalMgrGraphValid -off;

Once activated, error detection occurs each time an evaluation is performed. Detected missing dependencies are then saved to a file in your machine's temporary folder. For example, on Windows, you would find results in %TEMP%\_MayaEvaluationGraphValidation.txt.

The following is an example report that shows missing dependencies between two nodes:

Detected missing dependencies on frame 56

{

NodeA.output <-x- NodeB

NodeA.output <-x- NodeC [cluster]

}

Detected missing dependencies on frame 57

{

NodeA.output <-x- NodeB

NodeA.output <-x- NodeC [cluster]

}

The "<-x-" symbol indicates the direction of the dependency. The "[cluster]" term indicates that the node is inside of a cycle cluster, which means that any of nodes from the cycles could be responsible for attribute access outside of evaluation order

In the above example, NodeB accesses the "output" attribute of NodeA, which is incorrect. These types of dependency do not appear in the Evaluation graph and could cause a crash when running an evaluation in Parallel mode.

Missing dependencies in Analysis mode

There are many reasons that missing dependencies occur, and how you handle them depends on the cause of the problem. If Analysis mode discovers errors in your scene from bad dependencies due to user plug-ins, revisit your strategy for managing dirty propagation in your node. Make sure that any attempts to use "clever" dirty propagation dirty the same attributes every time. Avoid using different notification messages to trigger pulling on attributes for recomputation.

See Locate animation bottlenecks with the Profiler for information on identifying problematic areas in animated scenes.

Parent topic: Increase performance with the Evaluation Manager