linuxwebcluster.com

EC2 render speed comparison

Just an update on how the speed tests are going with the Amazon EC2 render farm. There does not seem to be any speed up by going 64 bit for my test renders. Possibly this is due to the particulars of the job I am running... But the "extra large" 64 bit EC2 instance I ran that had 4 dual cores and 8 GB of memory finished the render exactly 4 times faster than the 32 bit instance with 1 dual core. 

So the price and time to completion was identical either way. Now I am moving on to testing the EC2 cloud versus my own render nodes, mano a mano. And the results are in!

Several tests showed the same result, so I will not bore everyone with every last detail. The bottom line is that local renders were performed on a DL380 with 2 GB of memory and dual Xeon 2.8 Ghz CPU's. They are hyperthreaded so they show up in Linux as 4 CPU's. On average, frames were rendered fastest when processed 4 at a time and they finished in about 40 seconds per frame. 

In EC2 I used a single "high-cpu medium" instance for comparison. It claims to provide "virtual" dual Xeon 2.3 Ghz CPU's and 1.7 GB of memory.  It rendered the same images in about 65 seconds each, on average. So the remote renders were about 1.6 times slower than local, about 90% of which I attribute to CPU power. To estimate actual CPU power based on the experimental results, calling it a virtual dual Xeon 2.0 Ghz would be more accurate. Of course, the wonderful thing about EC2 is that I can provision as many of them as I want! The only thing you really need to know is how to baseline them... And now you have something for comparison. At 20 cents per hour per instance, it's a bargain if you want to burst your render farm capacity with Amazon's EC2. 

 
Comments (1)
1 Thursday, 30 April 2009 10:15
I'm not too surprised, I've never seen a real world scenario where moving to 64bit improved performance unless your bottleneck was the fact you were limited to 4GB of ram. Although at least in Windows, as far as I know Windows in 32bit mode can only address ram in 2GB chunks, so that alone could be a silent limitation in large datasets.

Despite, 4GB of ram and more is here to stay and we all need to move onto 64bit. I'm more curious what kind of performance boosts you'll see when you get ahold of those Core i7 based Xens.

Add your comment

Your name:
Subject:
Comment:

Subscribe to the Linux Admins Blog and get new posts delivered by email!
Enter your email address:

Delivered by FeedBurner

Linux News


Tell the developers:

The type of clustering you are most likely to deploy is:
 
What Linux distro do you use for clusters?
 

Copyright 2010    RapidScale Clusters, LLC