Samsung sues Nvidia, alleges benchmark fudging

November 12, 2014 // 12 p.m.

Tags: #benchmark #exynos #exynos-5433 #lawsuit #legal #nvidia #patent #samsung #tegra #tegra-k1

Samsung has responded to Nvidia's lawsuit claiming patent infringement with a suit of its own, accusing its rival of infringing Samsung patents as well as fiddling benchmark results to give its Tegra chips an artificial boost.

Nvidia filed suit against fellow ARM licensees Samsung and Qualcomm back in September, accusing the companies of infringing seven patents held on graphics processing unit (GPU) technologies including mobile GPU design, programmable shaders, unified shader hardware and multithreaded parallel processing. It marked the first time Nvidia had initiated legal proceedings using its 7,000-strong patent portfolio, with previous negotiations with companies including Intel leading to licensing agreements rather than lawsuits.

Samsung's response to the case is predictable: a counter-suit, claiming that Nvidia is infringing six Samsung patents while naming Velocity Micro as an alleged infringer of a further two patents. According to a write-up of the case on the registration-required Law360 site first reported on by Engadget, the suit also accuses Nvidia of attempting to 'confuse customers' with misleading benchmarks designed to show the Tegra K1 processor as outperforming Samsung's own Exynos 5433 when, Samsung argues, the reverse is true.

Nvidia has stated it is actively examining the case, but has already issued a strong rejection of Samsung's benchmark-fudging claims. The benchmarks used, Nvidia has told the site, used unmodified software on a stock-standard installation on both devices - allowing no room for customisation or other tricks to fudge the benchmark results.

Samsung's claim of benchmark-fiddling comes after the company was found to be placing a secret turbo mode into its own Exynos mobile processors which would boost their performance above normal levels when benchmark programs were detected, but not allow the same high-performance mode to be accessed by any other program. Further, renaming the benchmark's executable would result in the company's benchmark-detection system failing to spot the package - resulting in a true measure of real-world performance some 20 per cent lower than its 'official' results.
Discuss this in the forums

QUICK COMMENT

SUBSCRIBE TO OUR NEWSLETTER

WEEK IN REVIEW

TOP STORIES

SUGGESTED FOR YOU