AWS Elastic GPUs, now on Frame
Last year, AWS announced a tech preview of their new Elastic GPU technology that promised to lower the cost of graphics in the cloud. This new technology lets you add graphics capability to any instance type, resulting in a much wider variety of GPU accelerated instances.
Today, AWS announced pricing and general availability of Elastic GPUs in their Virginia and Ohio datacenters. The smallest elastic GPU, eg1.medium, starts at only $0.05/hour (for a 1GB GPU frame buffer).
Given the ability to combine virtually any CPU instance with an Elastic GPU, we can now see price points that are dramatically lower than existing AWS GPU options (depending on what your app needs, as much as 80% less). And after many months of testing and working with AWS in their tech preview, we’re announcing Frame support for elastic GPUs as well.
Why GPUs?
We all use GPUs in our PCs, laptops, and mobile devices at home and in our corporate offices. Why settle for less when delivering cloud-based applications via Frame?
For most applications, a GPU is a great way to boost the overall user experience with a richer, more responsive interface. While most think of GPUs as the engines that power 3D applications, many 2D applications make use of a GPU to render the user interface or to provide accelerated application response. The main challenge with cloud delivery for many mainstream apps is that GPUs can be expensive and are often only supported on certain hardware platforms. So this limits their reach to only power-users, specific high-end applications, or niche workflows.
In terms of the cloud, these GPU limitations have historically translated to only a few available GPU options. In fact, for the first couple of years, over 90% of GPU use cases at Frame were served by a single instance type, the g2.2xlarge, which, in fact, never changed in price.
This instance is still great for 3D CAD and other high-end use cases, but it’s sometimes overkill for mainstream applications. As a result, many of our customers opted to go without a GPU to save money, but this was a tradeoff in user experience for certain features.
Since last fall, we’ve seen more GPU options at the high-end from Azure with the N-series and from AWS with the G3 family — all of which we support. But we’ve been looking forward to a lower cost option. With Elastic GPU we get both lower cost and more flexibility.
Naturally, we were thrilled when Amazon announced the Elastic GPU technology at AWS re:Invent in 2016. Soon after, we started working with AWS to put the technology through its paces. And today we’re thrilled to be able to make it available to you on Frame.
So what makes Elastic GPU different and how does it work?
Elastic GPU uses technology that renders OpenGL frames externally on a separate GPU-powered system but displays them in the parent application running on a separate instance. This means that a GPU can be added to literally any system type. And since it comes in 1GB, 2GB, 4GB, and 8GB configurations, it’s a lot more flexible than instances with fixed CPU/RAM/GPU configurations.
Elastic GPUs are not without a few catches — they only work with OpenGL 3.3 and below (e.g., there’s no support for DirectX or any other standard GPU API, like CUDA or OpenCL; there’s also no support for GPU accelerated video encoding). Also, bandwidth and CPU need to be monitored to ensure they don’t cap out and limit performance, and performance optimization may be required for some applications to get the most out of the external rendering engine.
We’ve been testing Elastic GPUs since they were first made available in the preview and have worked closely with AWS to optimize the experience for many apps that our customers use. We’ve seen solid results with apps such as Google Earth, ANSYS AIM, Siemens SolidEdge, Adobe Photoshop, and more.
We tested a variety of Elastic GPU configurations during the preview including a combination that brings together a CPU instance with 16GB of RAM and a 4GB Elastic GPU. This resulted in a profile similar to our Pro 16GB instance (g2.2xlarge) but at about half the cost. As AWS availability expands to more regions and we get more feedback from you, (now that you can use elastic GPUs on Frame, too), we’ll roll-out more combinations.
The right tool for the job
It’s great to have various GPU options delivering the right solution for the right application. At Frame, our comprehensive platform focuses on application delivery and hides all cloud infrastructure complexity delivering applications to you: fast, simple, agile, cost efficient, and secure so you can achieve your goals. So whether you’re running a CAD application like PTC Creo that requires a dedicated GPU, a productivity application like Microsoft Powerpoint where you’d benefit from a partial GPU, or a rendering application that requires multiple GPUs, Frame can accommodate your needs anytime, anywhere and on any device. That’s the power of the cloud!
Want to learn more about Elastic GPUs, give them a try, talk about your use-cases, or something else? Let us know!
