More on AWS Elastic GPUs

First, as this is my first post here, I’d like to say thanks to Adam for allowing me to join Virtual Bonzo as an author.  For a few years now I’ve had the pleasure to work along side Adam and now we’re blogging….I look forward to working with him on this medium, as I think it will be fun, and hope that our posts will be beneficial to IT community.

Now for the post on AWS Elastic GPUs as Adam promised on Twitter…….

AWS has released Elastic GPUs for their EC2 instances….you can read more about it using the links found at the end of this post but I’m very excited to test this feature for both professional and personal reasons.

Prior to elastic GPUs, AWS users desiring increased graphics performance would need to deploy a G2/G3 EC2 instance that includes a “hardware” GPU.  Running Windows, the cheapest G2 instance, the G2.2xlarge, costs about $.77 per running hour to operate. Depending on your EC2 instance and graphics needs, the G2.2xlarge instance could be overkill and the over allocation of resources leads to waste as you’re not using the full capacity of the G2/3 instances, which means you’re paying for resources you’re not maximizing, which means you’re paying more than you need to or should.  But that has changed with the introduction of Elastic GPUs for AWS which provide easily attachable and lower cost graphics acceleration for many EC2 instance types.

Elastic GPUs come in 4 “sizes”:

  • Medium – 1GB | $.05 per running hour
  • Large – 2GB | $.10 per running hour
  • XLarge – 4GB | $.20 per running hour
  • 2XLarge – 8GB | $.40 per running hour

Additionally, Elastic GPUs support the following instance types:

  • C3 | C4
  • M3 | M4
  • R3 | R4
  • T2.Medium (or greater)
  • X1
  • D2
  • I3

According to Amazon, if your instance and graphics needs can be met with a T2 Medium Elastic GPU instances, you could obtain significant cost savings (up to 80%) as opposed to using the G2/3 instances.  That’s pretty cool!

Professionally, I am looking forward to testing (at least initially) elastic GPUs on Citrix XenApp instances to see what benefit, if any, an elastic GPU will provide in regards to end user published application performance.  Seems like it has to help to some degree though elastic GPUs may be more beneficial on single-user instances.

Personally, I like to take pictures and so I run Adobe Lightroom and Photoshop on occasion.  Not too long ago, my photo-editing laptop was stolen.  Granted, it was not an ideal platform on which to edit photos but it was all I had.  While considering the purchase of a better system on which to run the Adobe applications, I wondered if an EC2 instance would be capable of performing this task.  Do I really want to spend $1200 (or whatever) on a laptop to edit photos or simply pay for an on-demand G2 instance in the neighborhood of $20/month for 10-15 hours of usage and storage?  The thought of spending $1200 made my stomach crawl so I opted to try the G2 AWS instance.  It’s worked well but with the introduction of elastic GPUs, I began to wonder if a lower cost EC2 instance with an elastic GPU could run the Adobe applications as well as the G2.

To that end, I deployed a T2.Large EC2 instance with a 2GB elastic GPU and have begun testing….actually, I’m further along than that now.  On my next post,  I’ll walk through the setup and the issues I had when trying to “migrate” my existing G2 instance to an elastic GPU.

AWS Elastic GPU Links:

One thought on “More on AWS Elastic GPUs

Leave a Reply

Your email address will not be published. Required fields are marked *