Tuesday, June 19, 2012

Rendering in AutoCAD: Ralph Got It Wrong

Judging from Ralph Grabowski's response to an email I sent him yesterday, I don't think he has any intention of publishing my response to an article in this week's edition of his Upfront Ezine. So I'll share my response here.

I have a great deal of respect for Ralph. He is usually quite thorough and calls it as he sees it. But this week, he simply got it wrong.

In this week's lead article, Ralph claimed to have done a head-to-head comparison of AutoCAD's built-in rendering command with Autodesk's new cloud-based rendering service and concluded that the Autodesk 360 rendering was much slower than built-in local rendering. (You can read Ralph's article at http://www.upfrontezine.com/2012/upf-740.htm.)

After reading Ralph's article, I wrote the following email:

I really must take issue with the methodology you used in comparing AutoCAD's built-in rendering command with the new cloud-based rendering. A rendering that takes only 4 seconds on your local machine is hardly a valid test.

As you know, I have been testing computers and graphics cards and writing about AutoCAD about as long as you have. My reviews are regularly published in Desktop Engineering magazine, and I have taught numerous classes about rendering at Autodesk University. I also recently added more that 3 dozen lessons on rendering, both local and cloud-based, to the CADLearning series ofvideo-based tutorials for 4D Technologies' AutoCAD 2013 course.

As part of my reviews for Desktop Engineering, I perform a rendering test using a drawing I obtained from Heidi Hewett at Autodesk. Although the drawing itself is relatively simple, it includes numerous materials and multiple light sources, all of which result in much longer rendering times. On a relatively fast dual-core workstation, it is not unreasonable for the rendering to take up to 5 minutes, whereas on a system equipped with two 8-core CPUs with hyper threading enabled, the rendering might be completed in less than a minute. In addition to the number of materials and light sources, the resolution of the rendered image as well as the render quality will also impact the overall rendering time.

Again, this is for a relatively simple model. It would not be unreasonable for a highly detailed model with multiple light sources, all casting shadows, to take 30 minutes or more even on a very fast system with multiple CPUs. While I have not performed any benchmark tests yet comparing local rendering of such a model to Autodesk's online rendering service, I would expect the online service to be considerably faster.

While you do make some mention of the fact that it does take some time to initially transmit the model to the Autodesk 360 service (and you experienced some problems in doing so), it is also worth noting that the online service can render multiple scenes from the same model, whereas to do this on a local system, you would need to render each scene individually. Furthermore, when you render locally, you cannot do anything else in AutoCAD while the rendering is being computed, whereas when you use the online rendering service, you can go back to work in AutoCAD as soon as the online render request has been processed.

I also suspect that most people who use the Autodesk 360 service log in when they first start AutoCAD. They would therefore already be connected when they clicked the Render Online button and therefore would not experience the same delays as you. 

From the image you supplied, I can see that you rendered the model at a resolution of 640x480 using the Medium render preset, settings that are hardly representative of real-world situations. In my tests, I typically render at a resolution of 1280x1024 using the Presentation Quality preset. It's also apparent from the lack of shadows and materials, that you likely rendered using the default lighting mode and that the Oil Module drawing contained no materials.

I suggest that you come up with a more realistic test before you start publishing results that are purported to be representative of real-world results or typical user experiences.
And what was Ralph's reply to my email?

"I clearly laid out the specifications for the benchmarking so that readers could take it at face value. Thanks for your input!"

That's hardly a useful response and I think it does his readers a disservice.

Come on, Ralph. You got it wrong.