Research Environment and Infrastructure

AI Development and Research Environment

DGX Spark Environment

• NVIDIA DGX Spark: 2 units
• Processing Power: Capable of running large-scale LLMs in the 405B parameter class.
• Usage: Fully on-premise operation of open models such as GPT OSS120B, Meta Llama 3.1 405B, and DeepSeek V2 series.

Note: The supercomputer at Kyoto University uses NVIDIA A100 80GB SXM, which cannot operate open models exceeding 80GB.

Workstations

• PC with RTX 5070 Ti: VRAM 16GB / RAM 128GB
• Provides time-series calculation and high-performance local inference/development environment.

Secure Network Environment

• In addition to Kyoto University VPN, we have built our own VPN (private network).
• High-security remote access to laboratory PCs and AI-dedicated PCs is possible.

LLM Development System

• Closed LLMs: Professional utilization in secure environments.
• Open LLMs: General-purpose utilization and research development.
• Understanding characteristics of each LLM and clarifying pros and cons.