In a significant move towards bolstering its technological infrastructure, the British government has announced a substantial investment of £225 million in the supercomputer named Isambard-AI, which is expected to become the fastest supercomputer in the UK. Housed at the University of Bristol, Isambard-AI derives its exceptional computational power from 5448 Grace Hopper superchips supplied by Nvidia.
These GH200 chips represent a groundbreaking fusion of an H100 GPU with an Nvidia Grace CPU, packed into a single module comprising approximately 200 billion transistors. Each Grace Hopper superchip boasts an impressive configuration, with 72 Neoverse V2 CPU cores and a remarkable 16,896 CUDA cores. In total, Isambard-AI harnesses over 392,000 CPU cores and a staggering 92 million CUDA cores, further supported by nearly 25 petabytes of storage.
According to reports, Isambard-AI is capable of delivering more than 200 petaflops of fp64 compute power, as assessed by the TOP500 Linpack benchmark. The system also offers an astounding 21 exaflops of AI computing power, equating to a mind-boggling 200 quadrillion calculations per second. On paper, Isambard-AI is estimated to be ten times faster than the UK’s existing fastest supercomputer, placing it within the top ten fastest supercomputers globally.
Scheduled for deployment in the summer of 2024, Isambard-AI is set to be a cornerstone of various AI applications across the UK. These applications will encompass the training of large language models, handling big data, and advancing robotics, underscoring the nation’s commitment to achieving global leadership in the field of artificial intelligence. Hewlett Packard Enterprise, the system manufacturer, stands by this vision.
The supercomputer resides within a state-of-the-art, self-cooled, and self-contained data center located at the National Composites Centre within the Bristol and Bath Science Park. Furthermore, plans are already in motion for the construction of the Isambard-3 supercomputer, for which the British government allocated $10 million earlier this year, ensuring continued advancements in high-performance computing capabilities.