NVwulf | Research Computing & Informatics

NVwulf is Stony Brook's GPU cluster for AI and machine learning, with NVIDIA H200 and RTX PRO 6000 Blackwell GPUs. It's open to Stony Brook instructors, researchers, staff and students.

141 GBHBM3e on each H200 GPU

96 GBon each RTX PRO 6000 Blackwell

8GPUs on the largest nodes

80PFLOPS FP8

Get connected

In your browser

Open OnDemand

Jupyter, RStudio, VS Code, private AI chat, AlphaFold, cryo-EM and imaging apps on GPU nodes. Sign in with your NetID and Duo.

Launch Open OnDemand

From a terminal

SSH

NVwulf has its own storage, separate from SeaWulf. Copy data over with rsync or Globus.

ssh -X <netid>@login.nvwulf.stonybrook.edu

Your first hour, step by step: Getting started on NVwulf.

Specifications

Node type	Details
H200 nodes (4-way)	4 NVIDIA H200 NVL (141 GB HBM3e each), 64 CPUs, about 750 GB RAM
H200 nodes (8-way NVL)	8 NVIDIA H200 NVL, 64 CPUs, about 1,500 GB RAM
H200 nodes (8-way SXM)	8 NVIDIA H200 SXM, 192 CPUs, about 2,200 GB RAM
B40 nodes (4-way)	4 NVIDIA RTX PRO 6000 Blackwell (96 GB GDDR7 each), 64 CPUs, about 512 GB RAM
Scheduling	Multi-tenant Slurm, up to 8 GPUs per node

Queues

Partition	GPUs	Good to know
`h200x4`	4 H200 NVL per node	Also `-long` and `debug-` versions
`h200x8`	8 H200 per node	For the largest models and multi-GPU training
`b40x4`	4 RTX PRO 6000 per node	Open OnDemand apps run here: up to 8 hours, or 48 on `b40x4-long`

Start with a debug- partition for short tests, and ask only for the GPUs you'll use. Every queue's CPUs, memory and time limit, and the priority queues, are on NVwulf queues.

Your first GPU job

#!/bin/bash
#SBATCH --job-name=test-tf
#SBATCH --output=res.txt
#SBATCH --partition=h200x4
#SBATCH --nodes=1
#SBATCH --ntasks=8
#SBATCH --cpus-per-task=1
#SBATCH --mem=25g
#SBATCH --gpus=1
#SBATCH --time=05:00
module load tensorflow/2.19.0
python tf_test_nn_training.py

Save it as test_job.slurm, then run sbatch test_job.slurm and squeue -u $USER. Or build one with the job script builder.

Storage

Space	Size	Backed up	Good to know
Home `/lustre/nvwulf/home/<netid>`	20 GB	Yes	Never deleted
Scratch `/lustre/nvwulf/scratch/<netid>`	20 TB	No	Files older than 30 days are deleted

Popular apps in your browser

No patient data. NVwulf, including its AI chat apps, isn't approved for PHI. Work with PHI belongs on ClinWulf.

Best suited for

AI and machine learning
Biomedical imaging
Molecular modeling and NLP
Scientific computing and data science

Getting access

Join a projectIf your PI already has an active SeaWulf or NVwulf project, open a ticket with the help topic "Request an Account for NVwulf" and include your NetID, your PI's name and the project number.
Getting access to NVwulf
Or start one (faculty)Send a one-page PDF with the help topic "Request a Project Number – NVwulf". A review committee replies by email.
Set up Duo, then log inThen move your data over and run a first GPU job.
Duo setup Getting started on NVwulf