Qube 6.1 + AE CS5 + OS X 10.6.5

[RESIDENT_CREATIVE]

Howdy,

When rendering a job via Qube, no matter how I configure the subprocess (from the number of hosts to number of hosts x 2) and reservations (1+, 8, 16) to the type of rendering (single frames, chunks, chunks with n frames), I can't seem to get AE to use all the threads on any of the workers.

When rendering the same AE comp on a local machine via render queue, It immediately uses all threads.  When rendering the comp with just two hosts via qube (submit from ae rq->qube gui), it only appears to use one thread on each worker.  The render times reflect this, as well.  I tried adding "mp" to the command line window in the cube GUI as well ... no change. 

Anyone have any ideas?

Thanks,

Matt

[jburk]

What platform is this?  Adobe's aerender executable behaves differently than rendering from inside the GUI.  Aerender decides internally when and whether to spawn additional threads.

The use of the 'mp' flag (despite it's name) is not recommended when running renders strictly through aerender.  This is to allow multiple instances of aerender to work on different frames, which is not how Qube works with single instances of aerender.

Qube's aerender jobs often have the opposite behavior you're seeing; aerender spawns as many distinct threads as it deems necessary, and there's no way to throttle it back.  This makes it difficult to use concurrently with other jobs running on the same worker, so for this reason we recommend that an aerender job have the reservations set to "host.processors=1+" so that no other jobs will run on the worker at the same time.
 

[RESIDENT_CREATIVE]

Hey,

Thanks for the reply.

I've tried submitting it without the "mp" flag, and I've used "1+" as well.

I've pretty much tried every possible permutation of job submission options, and can't seem to get any of the workers to use more than one thread.

As I said, rendering the project locally always results in all 16 processors ramping up.

Thanks in advance for any advice.

Best,

Matt

[jburk]

This is a tough one; aerender and the AE GUI seem to make use different criteria about how many threads to launch. 

The job reservations won't affect the number of threads or cores that the aerender executable will utilize, it's to instruct the supervisor how much of the worker's resources to reserve for the job, which in turn affects how many (if any) jobs Qube will send to that worker at the same time that the aerender job is running.  As I mentioned earlier, most of the time user have the reverse problem; they send a job that only reserves 1 processor, while aerender renders across all the cores.

Are your individual frames in the render very lightweight or quick to render, or do they load in quite a few different assets to render a frame (are they i/o-bound?)

And when you render in the GUI and AE uses all the cores, is it rendering 1 frame at a time, or is it working on several frames at once, and may still be allocating only 1 core per frame?