“`html
The latest TX-Generative AI Next (TX-GAIN) computing framework at the Lincoln Laboratory Supercomputing Center (LLSC) stands as the most robust AI supercomputer in any U.S. higher education institution. Following its recent placement in the TOP500, which biannually releases a ranking of the best supercomputers in diverse categories, TX-GAIN is now among other formidable systems at the LLSC, all aiding research and innovation at Lincoln Laboratory and throughout the MIT campus.
“TX-GAIN will empower our scientists to realize advancements in science and engineering. The system will significantly contribute to generative AI, physical simulations, and data analysis across all research fields,” states Lincoln Laboratory Fellow Jeremy Kepner, who leads the LLSC.
The LLSC serves as a vital asset for expediting innovation at Lincoln Laboratory. Thousands of researchers utilize the LLSC for data analysis, model training, and simulation tasks for federally funded research initiatives. For instance, the supercomputers have simulated billions of aircraft interactions to establish collision-aversion systems for the Federal Aviation Administration and to train models in the intricate processes of autonomous navigation for the Department of Defense. Over time, the capabilities of LLSC have been crucial in numerous award-winning innovations, including enhancements in airline safety, halting the proliferation of new pathogens, and assisting in hurricane response strategies.
As its title implies, TX-GAIN is particularly suited for the development and implementation of generative AI. While conventional AI concentrates on classification tasks, such as determining whether an image features a dog or a cat, generative AI creates completely novel outputs. Kepner characterizes it as a mathematical fusion of interpolation (bridging gaps between known data points) and extrapolation (projecting data beyond established points). Currently, generative AI is widely recognized for utilizing expansive language models to generate human-like replies to user prompts.
At Lincoln Laboratory, teams are leveraging generative AI across various sectors beyond expansive language models. They are employing the technology, for example, to assess radar signatures, enhance weather data coverage where it is lacking, detect irregularities in network traffic, and investigate chemical interactions for the formulation of new medications and materials.
To facilitate such intense calculations, TX-GAIN is driven by over 600 NVIDIA graphics processing unit accelerators specifically engineered for AI tasks, in addition to conventional high-performance computing hardware. With a peak output of two AI exaflops (two quintillion floating-point operations every second), TX-GAIN is the leading AI system at a university and in the Northeast. Since TX-GAIN became operational this summer, researchers have taken note.
“TX-GAIN is enabling us to model not just significantly more protein interactions than ever before but also much larger proteins with more atoms. This advanced computational capability is transformative for protein characterization efforts in biological defense,” remarks Rafael Jaimes, a researcher in Lincoln Laboratory’s Counter–Weapons of Mass Destruction Systems Group.
The LLSC’s emphasis on interactive supercomputing renders it especially advantageous for researchers. For years, the LLSC has innovated software that enables users to access its powerful systems without the necessity of being experts in configuring algorithms for parallel processing.
“The LLSC has always aimed to make supercomputing feel akin to working on your laptop,” Kepner shares. “The volume of data and the complexity of analysis techniques required to remain competitive today far surpass what can be achieved on a laptop. However, with our user-friendly approach, individuals can operate their models and receive results promptly from their workstations.”
In addition to supporting initiatives exclusively at Lincoln Laboratory, TX-GAIN is enhancing collaborative research with MIT’s campus. These collaborations include the Haystack Observatory, Center for Quantum Engineering, Beaver Works, and Department of Air Force–MIT AI Accelerator. The latter project is quickly prototyping, scaling, and applying AI technologies for the U.S. Air Force and Space Force, optimizing flight scheduling for global operations as a demonstrated example.
The LLSC systems are situated in an energy-efficient data center and facility in Holyoke, Massachusetts. Research personnel at the LLSC are also addressing the immense energy demands of AI and spearheading research into various power-reduction strategies. One software tool they created can cut the energy required for training an AI model by up to 80 percent.
“The LLSC offers the capabilities necessary for conducting cutting-edge research while doing so in a cost-effective and energy-conscious manner,” states Kepner.
All supercomputers at the LLSC carry the “TX” designation as a tribute to Lincoln Laboratory’s Transistorized Experimental Computer Zero (TX-0) from 1956. TX-0 was among the first transistor-based systems in the world, and its 1958 successor, TX-2, is renowned for its contribution to advancing human-computer interaction and AI. With TX-GAIN, the LLSC perpetuates this legacy.
“`