I’m less than a week into using Deadline, so I’m very new to it.
I have just two PCs (i.e 2 nodes) - an i9 (my main PC and what I use everyday) and a 3 year old i7 Slave that is switched on and used only to give a helping hand with rendering (3ds Max, gave up trying to use the included Backburner).
As I’m new to using Deadline, I tested things by submitting a simple animation of just 10 frames, and assigned both machines to help out.
What I noticed regarding render times was a bit disappointing - the 3 year old, i7 Slave was getting through renders slightly faster than the much higher spec’d i9 Server
How could that possibly be ?
Furthermore, the i9 was not being very stressed…
What must I do to rectify this (control it even) ?
I seem to remember reading in the documentation that it is not recommended to include the machine which submits the job in the network of available nodes, but I can’t find it again.
I can’t be the only one who has just a couple of machines ?
How complex is the “simple animation of 10 frames”?
If a frame takes half a minute or just a few seconds to render, you don’t have enough time to load all CPUs with meaningful work, and a lot of the Task time ends up being management overhead. The first Tasks would also have to load 3ds Max (which can take several minutes), and load the scene and any assets (which is an I/O bound task, and usually not very CPU intensive). When rendering very fast frames, we generally recommend submitting with a larger Task Chunk Size, in your case a value of 10 would mean all 10 frames would be rendered in one go by one machine without any Deadline job/task management overhead in between.
On the other hand, if the frames you are rendering take around an hour each, then 10 frames over 10 hours would give you a much clearer picture of how fast each machine really is, as the influence of non-rendering overhead would be diminished.
Also note that some parts of 3ds Max are still single-threaded, so depending on the content of the scene, a large portion of the scene preparation can be single-threaded. Also you did not mention which renderer you were using.
In the Deadline monitor, you should be able to see statistics about the peak and average CPU load. Compare the results of the two machines and see if they were tasked in a meaningful way.
I would suggest downloading and running the V-Ray Benchmark application to get a better idea what the expected performance difference between the two machines really is. It is amazingly precise, and does not take the overhead of scene loading/preparation into account, so you get the actual rendering performance of your CPUs.
Thanks Borislav for your answer, all makes perfect sense, and yes, the frames are quick, around 2-3 minutes.
I’ll retry stressing both with a more complex, trying scene.
Yes, I use VRay for rendering and am aware of their Benchmark software - I’ve used it a few times and know that the new machine is 3 times faster than the 3 year old i7.
My situation has now changed in that my small network of 2 PCs consists of two identical i9 7980xe machines.
And this issue of the 2nd Slave machine renders the frames of my animations at a very normal-to-very-fast render time, whilst my Main PC now takes well over twice the time it should be taking to render a frame.
It’s as though there is something in Deadline telling the main PC to not run at full speed !
Each from should be taking around 20-25 seconds.
The slave does them in that time, the main pc does them in 40+ seconds…
Both machines are absolutely identical spec, components and software.
When I render the frames just on the main pc, with NO deadline involved, it will render them at the normal 20-25 second render time.
Where are the frames being saved? Any chance that networking between the two machines and waiting for the files to be saved from machine1 to machine2 accounts for the time difference?
