The most potent AI supercomputer at any American university is the Lincoln Laboratory Supercomputing Center’s (LLSC) brand-new TX-Generative AI Next (TX-GAIN) computing system. TX-GAIN joins the ranks of other potent systems at the LLSC that support research and development at Lincoln Laboratory and throughout the MIT campus, thanks to its recent ranking from TOP500, which publishes a list of the top supercomputers in various categories every two years.
“Our researchers will be able to make scientific and engineering advances thanks to TX-GAIN. Lincoln Laboratory Fellow Jeremy Kepner, who leads the LLSC, said the system will be crucial in enabling generative AI, physical modeling, and data analysis in all research domains.
At Lincoln Laboratory, the LLSC is a vital tool for boosting innovation. For federally supported research initiatives, thousands of researchers use the LLSC to conduct simulations, train models, and evaluate data. For instance, the Department of Defense has trained models in the intricate duties of autonomous navigation, and the Federal Aviation Administration has developed collision-avoidance systems by simulating billions of aircraft interactions using the supercomputers. Many award-winning systems throughout the years, such as those that have enhanced airline safety, stopped the spread of new diseases, and facilitated hurricane responses, have relied heavily on LLSC capabilities.
TX-GAIN is particularly well-suited for creating and utilizing generative AI, as the name implies. Generative AI generates completely new outputs, while traditional AI concentrates on classification tasks, such as determining whether a photo shows a dog or a cat. According to Kepner, it is a mathematical fusion of extrapolation (extending data beyond known locations) and interpolation (filling in the gaps between known data points). These days, generative AI is well-known for producing human-like replies to user input by using massive language models.
Teams at Lincoln Laboratory are using generative AI in fields other than big language models. For example, they are utilizing the technology to analyze radar signals, fill in gaps in weather data, identify network traffic irregularities, and investigate chemical interactions to create novel materials and medications.
In addition to conventional high-performance computing hardware, TX-GAIN is powered by over 600 NVIDIA graphics processing unit accelerators specifically made for AI operations to support such demanding computations. The best AI system in the Northeast and at a university is TX-GAIN, which can perform up to two AI exaflops (two quintillion floating-point operations per second). Researchers have been paying attention since TX-GAIN went online this summer.
“TX-GAIN is enabling us to model substantially larger proteins with more atoms in addition to a lot greater number of protein interactions than previously possible. For protein characterization efforts in biological defense, this new computational power is revolutionary, according to Rafael Jaimes, a researcher in the Counter-Weapons of Mass Destruction Systems Group at Lincoln Laboratory.
The LLSC is particularly helpful to researchers because of its emphasis on interactive supercomputing. The LLSC has long been at the forefront of developing software that enables users to utilize its robust systems without requiring them to be proficient in setting up parallel processing algorithms.
“The LLSC has always tried to make supercomputing feel like working on your laptop,” explains Kepner. “A laptop cannot handle the volume of data and the advanced analysis techniques required to be competitive in today’s market. However, users may run their model and obtain responses fast from their workspace thanks to our user-friendly methodology.
TX-GAIN is expanding research partnerships with the campus of MIT in addition to financing projects exclusively at Lincoln Laboratory. These partnerships include the Department of Air Force–MIT AI Accelerator, Beaver Works, Haystack Observatory, and Center for Quantum Engineering. One fielded example of the latter initiative’s quick prototyping, scalability, and application of AI technology for the U.S. Air Force and Space Force is the optimization of aircraft scheduling for worldwide operations.
The LLSC systems are located in Holyoke, Massachusetts, in an energy-efficient data center and facility. The LLSC’s research team is also addressing AI’s enormous energy requirements and spearheading studies into other power-saving techniques. They created a software solution that can cut the energy required to train an AI model by up to 80%.
“The LLSC provides the capabilities needed to do leading-edge research, while in a cost-effective and energy-efficient manner,” explains Kepner.
All of the LLSC’s supercomputers are referred to as “TX” in honor of the 1956 Transistorized Experimental Computer Zero (TX-0) at Lincoln Laboratory. One of the earliest transistor-based devices in history was the TX-0, and its 1958 sequel, the TX-2, is renowned for having helped to establish AI and human-computer interaction. The LLSC carries on this tradition with TX-GAIN.