NVIDIA Wants to Block Use of Cheaper Consumer GPUs in Data Center update from January 2018

NVIDIA ruffled some feathers with a recent change to the end user license agreement for its GeForce GPU software, used with its consumer-oriented GeForce and Titan GPU brands.

The agreement now includes language that with some exceptions prohibits use of the software in data centers, prompting angry comments online saying the company was trying to force data center users to buy its much more expensive Tesla GPUs, designed specifically for data center applications.

The language doesn’t prohibit the use of GeForce and Titan cards in data centers. Users who don’t update to the latest version of the software or write their own software are free to use them however they please. Researchers using the lower-cost chips in their systems are also allowed to continue using the software, and so are data center systems used for blockchain processing.

In a statement emailed to CNBC, an NVIDIA spokesperson said the new language was meant to steer data center users toward its products designed specifically for critical enterprise applications, with all the guarantees and services that come with that.

Data center GPUs have been a goldmine for the chipmaker. Its stock price skyrocketed over the course of 2017, driven largely by the quickly growing use of deep neural networks – a type of computer system used for artificial intelligence applications, which GPUs are particularly well suited to execute.

Steering users away from buying consumer PC GPUs for deep learning (the AI technique that uses deep neural networks) and toward Tesla can be viewed as a way to protect NVIDIA’s revenue growth driven by the deep learning trend. The highest-end GeForce GTX 1080 costs about $550, and the price of Titan V is $3,000, while the latest Tesla V100 comes at about $8,000.

The NVIDIA spokesperson advised customers who want to use GeForce software in data centers to contact its salespeople, and said the company would try to address the concerns on a case-by-case basis.