Nvidia has developed location tracking for its GPUs in an effort to prevent smugglers from reselling its devices in territories where they’re restricted.
Citing sources familiar with the matter, Reuters reports that the tracking would be optional for customers and taps into technology that otherwise monitors a chip’s performance. It would roll out first to Nvidia’s Blackwell chips.
“We’re in the process of implementing a new software service that empowers data center operators to monitor the health and inventory of their entire AI GPU fleet,” Nvidia tells Reuters. “This customer-installed software agent leverages GPU telemetry to monitor fleet health, integrity, and inventory.”
As PCMag reports, this comes amid concerns that Nvidia would add remote kill switches to its GPUs. In August, however, Nvidia published a blog post titled “No Backdoors. No Kill Switches. No Spyware,” which said “hardwiring a kill switch into a chip is…an open invitation for disaster.”
However, while the company will not brick GPUs at will, it turns out that it can use GPU telemetry to estimate the location of a graphics card. Nvidia describes the system as using the communication latency between the GPU and Nvidia’s own servers. If they appear far from their intended destination, there’s a real chance they’ve been smuggled.
Reportedly, the feature is “Read only,” meaning Nvidia can’t use this as a two-way communication channel to send any commands, like a kill switch. It aims to make this software open source, allowing security researchers and customers to verify that nothing untoward is happening.
In the US, lawmakers have introduced legislation that would require chip tracking; however, the bill hasn’t garnered much traction. More recently, the Trump administration reopened H200 sales to China this week, this time demanding a 25% cut of the proceeds. Blackwell and Rubin chips are not part of the deal.
—
Photo Credit: DANIEL CONSTANTE / Shutterstock.com