AI on Oscar

Building modern deep learning models requires specialized hardware called graphics processing units (GPUs). Originally designed for graphics rendering in computer games, GPUs' massively parallel architecture make them exceptionally well suited to the linear algebraic operations that are ubiquitous in modern deep learning models. As such, GPUs have become an essential component of deep learning and AI.

We have more than 800 NVIDIA GPUs on Oscar. These are available to anyone with an Oscar account. There are a wide range of GPUs on Oscar, from Titan RTX (24 GB VRAM) to B200 (180 GB VRAM).

Ollama is the easiest way to run large language models directly on Brown’s Oscar high‑performance computing cluster. Available as a preinstalled module, Ollama uses a simple client–server approach: you start a server on your allocated GPU node, then connect a second terminal as the client to chat with models or launch jobs.

We at CCV host more than 100 public, open‑weight models in a shared repository—such as Llama 4, DeepSeek‑r1, gpt‑oss, Mistral, and Gemma3—so you don’t have to download anything; you simply point the OLLAMA_MODELS environment variable to CCV’s shared path and run. We are also open to adding more models, so you can always contact us to request one!