Recap and point cloud Multi-threading

[Anonymous]

Hello,

I have some questions on the multi-threaded benefits of working with point clouds in ReCap and other Autodesk products. I have a test system with dual 18 core Xeons (36 cores, 72 threads). When loading/manipulating certain point clouds in ReCap, I notice it that all of the cores are being used, although not quite at 100%. Other point cloud data only use a single CPU core.

The point cloud data set that uses 100% of the cores has RGB color data while the data set that seems to be single threaded does not seem to have color information. The color data set is also larger.

Also I notice that when I import the rgb point cloud data into AutoCAD 2015, all 72 threads are being utilized 100%. This shows that manipulating point cloud data in AutoCAD is fully multi-threaded and it takes advantage of hyper threading (ReCap seems to be multi-threaded but does not take advantage of hyperthreading)

Questions:
1) Are point clouds with RGB information the only types that take advantage of multi-threading?
2) AutoCAD seems to be highly multi-threaded with point clouds. What is the actual benefit of using many CPU cores. Is it faster load times? Faster interactivity? Both?
3) I'm curious how other Autodesk apps handle point clouds. I've tried 3ds Max and it doesn't seem to utilize hypertheading. Is Revit, Navisworks and Invenor as efficient as AutoCAD??

Note: I'm not a user, just trying to understand the real compute requirements. I attached a screen grab of AutoCAD using 72 threads.\

Thanks all!
Joe Pizzini

BOXX Technologies

[Anonymous]

Bump.

I too am interested in the above questions as I work with point-clouds and many of the above mentioned apps ... mainly Revit, ReCap, Navisworks.

[ryan.frenz · Accepted solution]

ReCap (and the underlying engine used in the other products) will burst to multiple cores for select compute-intensive operations, including:

• Registration
• Import & Indexing
• Generation of other project artifacts (map images, cache files)
• Point cloud loading and filtering
• Feature extraction (cylinders, planes, snapping)

In most cases, these operations will use as many cores as are available and authorized by the OS.

The reason you're seeing varying behavior in the other Autodesk products is that the ReCap engine allows for a lot of flexibility in how the product (e.g. AutoCAD or 3DS Max) uses the data.  Most products use the same loading, filtering, and rendering pipeline as ReCap.  In this case, you should see multi-core usage during dynamic point cloud loading and interaction.  But it's up to the product how to do it.

Once the points are loaded, the products will vary further in how they query ReCap for things like snapping and measurement.  I would expect that in most case the point cloud part of these features will use a single core (independent from the dynamic loading behavior and any CAD-related activity).

In general, ReCap will perform better with more CPU cores.  However, all of the operations listed above are I/O limited for datasets of any non-trivial size.  That is, CPU cores will only help you once the data is pulled from disk and put into RAM.  So speed of disk and RAM are equally (or more) important to performance.

Hope this helps,

Ryan

[TrevorRitchie73]

Really disappointing that Autodesk is not embracing mutlithreading.

[Jyles.Bowland]