Research Compute Resources

The Institute of Science Tokyo provides the computing infrastructure through TSUBAME4.0, with access to up to 128 NVIDIA H100 GPUs. Our lab allocates compute credits to support open-ended research and collaborations.

There are two ways to roughly estimate our lab’s compute allocation:

128 × H100s for 128 days (standard priority)Our lab’s compute allocation supports about 128 days of continuous use at standard priority. (Highest priority uses 4× the compute credits for the same GPU count and runtime.)

24 × H100s for every day (highest priority)At highest priority, our lab’s compute allocation supports continuous training on 24 H100 GPUs every day through March 31, 2027.

These estimates describe our lab’s compute allocation as of October 21, 2026. GPUs are allocated through a shared queue, so availability and start times depend on scheduling. Larger runs use the budget faster; smaller runs can run longer.

Valid through March 31, 2027. Our lab’s compute allocation will be renewed after that date, and these estimates will be updated for the new budget.

TSUBAME scheduling and priority details

If you are interested in collaborating, please get in touch using the contact links on my profile.

Xuhan SHENG盛 栩涵

Research Compute Resources

Xuhan SHENG盛栩涵