The Alps supercomputer, recognized as one of the most powerful globally, is being repurposed for AI model training. Traditionally employed for scientific research, supercomputers are increasingly accommodating lower precision datatypes like FP8, significantly enhancing performance. ETH Zürich and Lausanne researchers presented open large language models developed on Alps, which merges its high-performance computing capabilities with AI. Constructed with Nvidia's Grace-Hopper Superchips, Alps demonstrates impressive computing power, achieving 42 exaFLOPS with sparse FP8 operations, illustrating a revolutionary shift in supercomputer utilization for AI tasks.
Supercomputers are increasingly being used to train AI models due to better support for lower-precision datatypes, marking a shift in their traditional scientific workloads.
ETH Zürich researchers introduced open large language models trained on the Alps supercomputer, which showcases its capabilities in both AI and traditional HPC applications.
The Alps supercomputer, built on Nvidia's Grace-Hopper Superchips, excels in sparse compute by achieving nearly four petaFLOPS with FP8 performance, enhancing the AI training process.
With 42 exaFLOPS of sparse FP8 performance, Alps has redefined the supercomputer landscape, enabling significant advancements in AI model training.
Collection
[
|
...
]