Discrete GPU shipments have been in decline for some time, as software to take advantage of their power lags and Moore’s Law let free- GPUs, those integrated GPUs that come with the processor and aren’t charged for, catch up and run three- and four-year-old apps pretty well. Seems most folks are happy with three- or four-year-old apps, and operating systems, too—if it ain’t broke, don’t fix it?
Today’s high-end GPUs are acceptable if they can run a game at 30 frames a second (fps). Sixty fps would be noticeably better, and technologies like FreeSync and G-Sync are examples of the things done to compensate for lower frame rates. For some examples of this, take a look at the tests we just did on the EVGA GTX 980ti and AMD Fury X high-end AIBs in this issue (page 17). 
Those games and tests were run on one monitor. VR with two high-resolution screens, i.e., 1080-line HD or better, needs to run at 90 fps to stay up to speed when the user is moving her head. And if you don’t like looking at little black dots, then you need a screen resolution of 1440 lines or more per eye. So for the head-mounted displays (HMDs) with dedicated displays (as opposed to the HMDs that use a smartphone for the display), in order to create differentiation, and justify the significant cost difference, you need a high-resolution screen per eye and a damn powerful AIB or GPU driving each screen. That’s right: two high-end GPUs to get VR right. And “right” means fast response screens to minimize the VR puke effect. You also need great graphics to deliver full color, process all the magic graphics functions to make realistic images and scenes, and process the real-time effects using physics engines. That’s a lot of work for a GPU, and in VR you need two, and two big ones—you can only do that with discretes and a lot of dedicated high-speed memory.
But how big can the VR market get, and what portion of it will be high-res dedicated screen HMDs?
Discrete GPUs are also finding homes in servers for virtualized and remote graphics, but relative to overall PCs, that’s a small, but growing market.
So virtual reality head-mounted displays look like a good opportunity for the discrete GPU suppliers. And if you believe some of the forecasts that have been offered, it’s going to be a huge opportunity. I’m a bit more cautious about some of the forecasts; there are still a lot of problems to solve before VR HMDs will become a general consumer item such as smartphones or TV, but there is no question it’s gotten a lot of people excited and interested, and new ideas for applications are popping up every day.
There’s still the issue of wearing an HMD for an extended period. Of people I’ve asked the first question, Have you tried it?, 5% to 10% say yes. Second question, How long did you try it?, most people say 5 to 10 minutes. Third question, Did you get uncomfortable? 50% to 60% say yes. Did you enjoy it? 70% to 80% say yes. Some got tired of the HDM, and some said it felt claustro-phobic, and there is the price. I think an HMD with controller(s) will have to be in the $200 to $250 price range to get any significant market penetration. The non-commercial, scientific, military, and industrial applications will continue to be steady and grow, and those segments will grow a bit faster due to the cost reductions in the components. All of that will create demand for powerful GPUs.
So virtual reality is a great opportunity for the discrete GPU suppliers, and if no serious roadblocks are hit in the next six to twelve months, it could indeed stem the decline in sales.