Advanced Micro Devices, Inc. (AMD) is betting a history-making advance on a bold initiative. Their mission is to change the way the world develops software through AI. AMD is developing an advanced combination of discriminative and generative AI. This project is intended to increase its capacity to identify vulnerabilities in firmware code. All of these steps are focused not just on greater efficiency but on raising the strategic caliber of the DOT’s end product — software.
As a result, the AI initiative was rooted in lessons learned from a detailed survey distributed to all of AMD’s development and quality assurance teams. The survey found that developers spend less than 40 percent of their time actually writing code. Their time is then wasted on other activities, such as testing new software. The reality is that sixty percent of your time is already spent reading reports, attending meetings and other activities. These tasks range from learning new technologies, debugging problems, optimizing existing code, and documenting your work.
This bottom-up, strategic approach to project development shows AMD’s dedication to getting a beat on the daily struggles of their rank-and-file engineers. Armed with this knowledge, AMD is now committed to producing targeted tools. These tools are going to help developers do everything from improving code quality to automating repetitive tasks and ultimately even increasing developer productivity.
AI Training and Development
AMD is opening its AI program with the training of deep convolutional neural networks (CNNs). They achieved these results through a novel deep learning architecture, utilizing an earlier, proprietary dataset with over 20,000 “golden” images and 2,000 distorted images. Using these neural networks to play video games in order to identify artifacts in image rendering. More importantly, they are specifically focused on AMD hardware.
All AI technologies become truly revolutionary during the functional testing stage of software development. Bringing AI into this process is helping AMD avoid 68 percent of manual testing and counting. Well, turns out they’ve done even better—an amazing 15 percent reduction! With that, you can now test 20 percent more scenarios. This uptick reflects the ability for AI to improve both the speed and depth of testing processes.
AMD’s efforts don’t stop there, going well beyond functional testing. The company is hard at work creating complementary individual tools to help developers conquer their everyday challenges. These tools will be indispensable for code generation, code review, code triage, debug assistance, and learning new technologies. These tools are positioned to be an instrumental part of rethinking software development workflows to maximize productivity.
Productivity Gains and Usage
AMD has a thoughtful, forward-looking strategy under the hood. They plan to improve productivity by a minimum of 25 percent over the next few years by rolling out AI tools. Today, nearly half of AMD’s active userbase are using these modern, creative solutions to enhance the efficiency of their work processes. The company is convinced that complete, deep integration of these AI utilities into the developers’ workspace will maximize productivity with teams of developers. They already anticipate this increase to be well above what was originally projected.
Beyond measurable productivity improvements, AMD is deeply concerned about the ethics of AI. Further, the company is concerned over possible infringement on intellectual property when adopting AI-driven recommendations. In addition, there is increasing concern about unintentionally releasing AMD’s own intellectual property. This is the problem with any broadly available AI tools used in public. This vigilant effort is indicative of AMD’s continued dedication to protecting the integrity and security of its proprietary technology.
Team Structure and Development Phases
AMD’s development team comprises a diverse group of professionals: 60 engineers are involved in the development and testing phases, while an additional 20 focus on support tasks. An even smaller team of 10 engineers does design work, with another five focused on proactively deciding project requirements. With this thoughtful make-up, the whole team is able to collaborate across all stages of software development.
The release cadence for new software at AMD is usually very tightly regimented and scheduled. The regular release cycle for the AMD Adrenalin graphics-software package has historically been six months on average. After this initial phase, there is a second, less-intensive period of support that continues for an additional three to six months. This comprehensive strategy ensures that AMD can deliver the highest quality updates, while delivering timely fixes and support for its products.
AMD is being ambitious with its plans to apply generative AI across its software development workflows. This strategy dramatically increases bottom-line efficiency. AMD is committed to solving the day-to-day problems of its engineers. By using cutting edge technology, the company has placed itself at the forefront of their industry’s fast-changing landscape.