Nvidia goes beyond PC gaming

Nvidia CEO Jen-Hsun Huang admits that PCs with integrated graphics have "largely become good enough." So why is the company doing so well?
Written by John Morris, Contributor
Nvidia made its name selling GPUs to PC makers for desktops and laptops. But as Intel and AMD integrated more features into the processor and steadily improved onboard graphics, that business started to shrink. The company endured years of questions about "attach rates," industry jargon for the percentage of PCs with discrete graphics. Then the PC market collapsed altogether as users shifted their attention to smartphones and tablets.
CEO Jen-Hsun Huang admits "PCs have largely become good enough."

In spite of all this, Nvidia just reported record revenues in the fourth quarter and for 2014. So far the company seems to be navigating the rapid shifts in technology, caused by trends such as mobile and the cloud. Gaming remains a key market, but Nvidia is increasingly finding new and interesting uses for its graphics technology.

Nvidia's PC OEM business is now a small part of the company's business (and shrinking) because, as CEO Jen-Hsun Huang admitted on a quarterly call with analysts this week, "PCs have largely become good enough" for most users without discrete GPUs. But PC gaming overall is still thriving. Nvidia did especially we here during the holidays thanks to the launch of its Maxwell GPUs, starting with the high-end GeForce GTX 980 and 970, and more recently with the $200 GTX 960 for the mainstream market.

Nvidia: We can reach 80 percent of enterprises now

In October the company also rolled out mobile versions, the GTX 980M and GTX 970M, which it said were the first notebook GPUs to support 1080p gaming with high-quality settings. Alienware/Dell, Asus, Gigabyte, HP, MSI and Razer are all using the new GPUs in some gaming laptops. The success of the Maxwell architecture is particularly surprising when you consider that it is stuck with the same 28-nanometer process technology that has been used for GPUs for more than four years.

The more novel applications are in the enterprise and the cloud. Nvidia's Quadro workstation business is basically flat but the company has been pushing GPU-enabled servers for desktop virtualization. The GRID virtualization platform now supports both Citrix and VMWare, is available in 100 server designs (compared with about 50 a year ago) and is being tested in more than 1,000 organizations. In theory, GRID lets users work on files anywhere on any device, provides organizations with better security, and enables collaboration on large files since there's no need to download them to local workstations. But the question is whether Nvidia can turn trials, which have been going on for some time, into real production deployments.

The other growth area is in accelerated computing for high-performance computing customers and cloud service providers. Nvidia's Tesla business has been growing rapidly, and at Supercomputing 2014 the company announced the dual-GPU Tesla K80, which provides twice the performance and memory bandwidth of its Tesla K40. China's Baidu is using a Tesla supercomputer for natural language processing, real-time translation and image recognition and earlier this week Microsoft announced that a team of its researchers in Asia had built a computer vision system based on neural nets trained with Nvidia GPUs that for the first recognized and classified images better than humans on a standard test.

Aside from GPUs, Nvidia has it Tegra mobile processor business, which is now focused almost exclusively on automotive and Shield gaming platform. The automotive business is growing (the company claims there are now more than 7.5 million cars on the road using Nvidia chips) and at CES it announced the DRIVE platform focused on the digital cockpit, surround vision and deep learning (natural language processing and computer vision). DRIVE is based on the 20nm Tegra X1, a Maxwell GPU with 256 cores that will be capable of more than one teraflop (one trillion floating point operations per second). The challenge with automotive technology, as I've written previously, is that it can take years for this technology to show up in cars though Nvidia says things are getting better.

The current Shield gaming tablet is based on the 28nm Tegra K1 with Nvidia's custom 64-bit ARMv8 CPU core. But there are rumors that Nvidia will announce an updated tablet, perhaps with the Tegra X1, at an event on March 3.

Not everything the company has tried worked out. The original plan for Tegra was to break into smartphones and tablets. It had some notable wins--the Tegra K1 is currently in Google's Nexus 9 tablet--but overall Tegra processor sales continue to decline. The company's acquisition of Icera and its soft modem did not work out and Nvidia has given up on cellular basebands. Early efforts to push GRID for game streaming services also didn't pan out, though Nvidia continues to develop its own game streaming service for Shield devices. Finally efforts to license its GPU technology have run into a lot of resistance, forcing Nvidia to sue Qualcomm and Samsung.

Finally while Nvidia seems to have the edge for now with its Maxwell architecture, rival AMD is rumored to be working on a Radeon R9 and Radeon R7 300-series GPUs that should bring tougher competition later this year. But overall it's impressive to see how Nvidia has managed to side step the problems with the PC business and develop promising new markets for its graphics technology.

Editorial standards