6 August 2018
GraphicsFuzz is acquired by Google
We are excited to announce that GraphicsFuzz Ltd. has been acquired by Google, and that the GraphicsFuzz team – Alastair Donaldson, Hugues Evrard and Paul Thomson – will be joining Google’s Android Graphics Team to integrate their specialist graphics driver testing technology within the Android ecosystem.
GraphicsFuzz has pioneered the combination of fuzzing and metamorphic testing to yield a highly automatic method for testing graphics drivers that quickly finds and fixes bugs that could undermine reliability and security before they affect end users. As reported in a series of blog posts on Medium, the approach has been successful at exposing a large number of graphics driver defects across a wide range of mobile and desktop platforms.
The GraphicsFuzz technology was developed by Alastair, Hugues, Paul and Andrei Lascu from the Department of Computing at Imperial College London, with funding support from the UK Engineering and Physical Sciences Research Council and the TETRACOM EU project. The team entered and reached the finals in the first edition of Programm/able, a combined business and software competition co-organised by Imperial Innovations, the College’s technology commercialisation partner, and Imperial Enterprise Lab.
The inventors continued to work with Imperial Innovations after the competition to set up the company, and GraphicsFuzz was one of the first companies to found a startup through the College's Founders Choice™ Programme, which offers academics the option to keep a much greater share of founding equity – up to 95% – in exchange for a more basic level of support.
The team was also supported by the Imperial Venture Mentoring Service, the ICURe Programme, and the Cybersecurity Academic Startups Programme funded by Innovate UK and the Department for Digital, Culture, Media and Sport (DCMS).
Alastair Donaldson, Director of GraphicsFuzz, said: "The acquisition by Google is a fantastic opportunity to maximise the worldwide impact of our graphics driver testing technology."
Lamia Baker, Senior Licensing Executive at Imperial Innovations, said: "The technology developed by GraphicsFuzz helps graphics technology vendors to build more reliable products, which is in high demand in today's technology-driven world. This acquisition is an endorsement of the company's progress, as well as a unique opportunity for GraphicsFuzz to have a positive impact in one of the world's leading mobile operating systems."
Simon Hepworth, Director of Enterprise at Imperial College London, said: "At Imperial, we encourage innovation at all levels. We have a unique entrepreneurial ecosystem, where our world-leading academics and students are encouraged to act boldly, pursue new opportunities, and translate their work into real-world impact. GraphicsFuzz embodies this culture, and we commend them on their tremendous success."
Try the GraphicsFuzz Demo right from your browser!
GraphicsFuzz in the News
- [23 Apr. 2018] XDA: The Snapdragon Samsung Galaxy S9 has a GPU Stability Bug that can be Exploited to Trigger Remote Reboots
- [23 Apr. 2018] wccftech: Critical GPU Bug Capable of Triggering Reboots Found in Snapdragon Variants of the Samsung Galaxy S9/S9 Plus
- [28 Feb. 2018] Phoronix: GraphicsFuzz Demo Works On Fuzzing Your GPU Drivers Through WebGL In The Browser
- [17 Jan. 2018] Hackernoon (re-publication of our blog post): A tale of two Samsungs: ARM vs. Qualcomm in Android graphics
Comparative reliability of graphics drivers
Click on header to sort the table.
|Huawei Honor 10 (ARM)||0||10||0||10|
|Huawei Honor 9 Lite (ARM)||1||4||6||11|
|Samsung Galaxy S9 (ARM)||0||13||0||13|
|Apple iPhone 7 (Apple)||6||9||0||15|
|Apple iPhone 6 (Apple)||6||9||0||15|
|Samsung Galaxy S8 (ARM)||1||16||1||18|
|Samsung Galaxy S6 (ARM)||8||13||3||24|
|Apple iPhone X (Apple)||17||11||0||28|
|NVIDIA Shield TV (NVIDIA)||3||11||14||28|
|Apple iPhone 8 (Apple)||17||11||0||28|
|NVIDIA Shield Tablet (NVIDIA)||2||16||11||29|
|Samsung Galaxy S7 (ARM)||13||17||0||30|
|Huawei Honor 9 (ARM)||12||16||2||30|
|Samsung Galaxy S9 (Qualcomm)||4||27||0||31|
|Google Nexus Player (Imagination Technologies)||17||25||1||43|
|Samsung Galaxy S8 (Qualcomm)||6||37||0||43|
|Google Pixel 2 (Qualcomm)||4||45||0||49|
|Google Pixel 1 XL (Qualcomm)||6||43||1||50|
|Samsung Galaxy S7 (Qualcomm)||31||33||0||64|
Latest test results
The GraphicsFuzz Demo
Try our online Demo
Run our 60 second Demo right from your browser!
This web app showcases a range of graphics driver issues in phones, tablets, smart TVs. The tests were selected to highlight graphics driver issues especially on Android devices, yet it runs on any recent web-browser. How does your device score?
See the results that users are posting on Twitter by searching for #GraphicsFuzz.
ShaderTest GLES contains 3000 shaders crafted to stress-test OpenGL ES drivers.
- Find drivers bugs: crashes, errors, miscompilations
- Reduce bugs: our intelligent reducer creates minimal reproducion test cases
- dEQP integration: run ShaderTest within the official Khronos CTS
WebGL, OpenGL, Vulkan...
We can provide bespoke products to test drivers for various graphics APIs, including WebGL, OpenGL, and Vulkan.
The GraphicsFuzz technology has been publicly credited for detecting several vulnerabilities:
- CVE-2017-2424: Apple (iOS-Webkit), information disclosure
- CVE-2017-6259: Nvidia, denial of service
- Information stealing between Chrome tabs on Samsung S6 (received Google Chrome bug bounty: $2,000)
Blog and articles
Our series of blog posts describes our testing approach and illustrates results on all major GPU vendors. Start the series with the intro, or jump directly to a post:
GraphicsFuzz started as an academic research project by the Multicore Programming Group at Imperial College London. Check out our article published at the OOPSLA 2017 conference.
GraphicsFuzz testing framework
The GraphicsFuzz testing framework automatically exposes bugs in graphics drivers in a simple form that supports debugging.
Our key technology enables detection of bugs that lead to an incorrect image being rendered, often the symptom of a shader miscompilation. Rigorously testing for such bugs without our technology is extremely challenging, and so is typically not done. Thus, these bugs are released into production where they will eventually be found by end users and developers, leading to damaged reputation and, in some cases, severe security bugs.
Even after finding bugs using our technology, the process of fixing these bugs would normally be extremely time-consuming, requiring a huge number of developer-hours. Our intelligent reducer tool analyses the bugs and creates simplified bug reports that dramatically decrease the time needed to understand and fix the underlying bugs.
Our reducer is intelligent in two different ways:
- Our reducer is able to simplify inputs that trigger miscompilation by utilising information embedded in the input by our GraphicsFuzz framework.
- Our reducer is able to understand the semantics of the shader language, allowing for faster and more precise simplification than that which could be offered by any general-purpose reducer.