You'll soon be using GPU as a Service

The role and deployment of the GPU is about to change dramatically -- and it will usher in some major new market opportunities.

radeon nano fiji gpu

Most people are familiar with the graphics processor unit (GPU), which is an important part of making your PC or smartphone graphically and video enabled. In fact, the GPU has a long history.

It started out primarily as a peripheral supplementing the central processing unit (CPU), first in high end workstations and then in a PC. It was optimized to support graphical operations (something that a CPU of early days did not do very well). Fast forward to today, and we find that most general purpose CPU chips powering the typical PC have more than enough graphics capability built in from an internal GPU to process normal apps and video operations (although some high end apps like gaming and VR/AR still need an external GPU for maximum performance).

With a troubled market for PC makers in general, and with a dwindling share of device attach rates for standalone GPUs, the primary players (AMD, NVIDIA) have attempted to innovate and stay relevant in a changing market. Much of that has been focused around a small but important segment of the PC market optimized around high performance gaming, an important but niche market (<10% of the PC market).

But despite the limited prospects over the past couple of years, things are now starting to change.

Although gaming will remain a key market for higher end GPUs, the overall share of the GPU market for that segment is about to plummet, not because it is declining in volumes, but because there are major new opportunities emerging for GPU vendors.

One is the expected growth over the next 1-2 years of virtual reality (VR), and not just for gaming. Solving real world problems and providing new ways of interacting with information is emerging as companies find new ways to engage with their workforces supplemented by augmented reality (AR) capabilities. Consumers too will soon see services based on AR/VR, but these will be slower to deploy due to costs and complexity considerations.

As an example of this new wave, AMD and AP are collaborating to bring immersive experience to news and storytelling. This can significantly enhance the ability to get information to content viewers, while also providing a more concise way to impart information, including the ability to see multiple perspectives, exhibit full dimensional accuracy, get a better sense of time, etc. Although still a niche market, this will help accelerate the adoption of VR clients (e.g., Oculus Rift, HTC, Samsung, Microsoft HoloLens).

In addition to VR clients, the need to process immersive information means that there will be a significant need for high performance graphics processors -- not only at the individual server level, but available as an on-demand service based in the cloud.

GPUs as a Service will expand greatly over the next 2-3 years, and will eclipse the PC GPU market in sheer numbers of units. Much like rack-based CPUs powering the Web and the cloud, similar GPU powered blades will be used for all manner of processing, and will be attached to major cloud hubs from Amazon, Google, IBM and Microsoft.

Based on their massively parallel processing capability, GPU as a Service will be used for augmented reality, but will also able to handle massively parallel complex app problems like encryption (or decryption), weather forecasting, business intelligence graphical displays, big data comparisons, etc. far better than standard CPU systems. Indeed, this is already happening with several examples of such massively parallel systems built. But most are immensely expensive, and as such, offering comparable performance as a service makes a lot of sense.

As a result of this trend, the primary GPU providers and major competing entrants (e.g., Intel) will ultimately power the GPU-as-a-Service industry for the highest performance products.

But there will also be a lower end market, much like the difference in CPUs between Intel x86 processors and ARM based chips are beginning to play out in the cloud market. Companies that provide graphics processors for the current mobile market (e.g., ARM Mali, Qualcomm Adreno, Imagination Technologies PowerVR) will be able to produce large numbers of lower end system components powering a blade based GPU environment. We expect that many companies will have such systems installed in their own data centers for complex visualization solutions in data analysis and forecasting, as well as security, especially as costs come down.

But the majority of the opportunity going forward for higher end systems will be in cloud-based services attached to the major cloud provider. This means that most companies will be able to access such services and take advantage of the now emerging services and apps that can benefit from the massively parallel capabilities.

I expect a major new market to emerge for GPU as a Service over the next 2-3 years. This should greatly benefit both the traditional GPU makers as well as the mobile oriented vendors that are ARM focused. As new apps become available -- optimized for these services -- companies will need to upgrade to obtain the benefits that such highly parallel processing power can bring. Enterprises should start planning for these upgrades coming in the next 1-2 years. But most organizations will be able to access needed capabilities as a service, and not have to make direct capital investments.

This should greatly accelerate adoption. Don’t get left behind.

This article is published as part of the IDG Contributor Network. Want to Join?

The march toward exascale computers
View Comments
Join the discussion
Be the first to comment on this article. Our Commenting Policies