TrainingNRP Training Recordings
Catch up on Docker, Kubernetes, JupyterHub and more from recent sessions.
Shared research computing, run by the people who use it
The National Research Platform pools GPUs, petabyte-scale storage, and hosted open-weight AI models across more than 70 institutions. It is free to U.S. nonprofit research and education, funded by the National Science Foundation and governed by the community that contributes to it.
- 400+
- Nodes
- 70+
- Locations
- 3
- Continents
- 5K+
- Users
What can you do with the NRP?
Use our hosted JupyterHub, or run your own
Sign in with your institution and get a notebook server on shared GPUs — no setup, no Kubernetes. When a course or a lab needs more control, run a private hub with your own images, profiles, and sign-in rules.
- Hosted
- Sign in, pick a profile, get a server
- Your own
- Your images, your profiles, your sign-in rules
- Behind both
- The same shared GPUs and CPUs as everything else
Train and serve models on shared GPUs
Ask for GPUs by the card in a job manifest, or take a whole multi-GPU node — the scheduler holds 4- and 8-GPU machines back for jobs that need them. Pin a specific accelerator when your run needs the memory.
- Interface
kubectl, or a browser IDE- Pick by
- GPU product, CUDA runtime, region, or host
- Also here
- FPGAs and other specialized accelerators
# four GPUs on one node
apiVersion: batch/v1
kind: Job
spec:
template:
spec:
restartPolicy: Never
containers:
- name: train
image: pytorch/pytorch
resources:
limits:
nvidia.com/gpu: "4"
memory: 64Gi
cpu: "16"kubectl apply -f train-job.yaml
Call open-weight models from your own code
A rotating catalog of open-weight models runs on NRP hardware behind an OpenAI-compatible endpoint. Point any client at it with a personal API key, or skip the setup entirely and open the hosted chat interface in a browser.
- Endpoint
- OpenAI-compatible, with per-user API keys
- Models
qwen3·kimi·glm-5·gpt-oss- Clients
- Open WebUI, Chatbox, or your coding CLI
Host services close to your data
Not everything is a batch job. Expose an HTTP service on its own nrp-nautilus.io hostname with TLS, then pin it to a region, a zone, or a single machine when it has to sit next to the instrument or the dataset it serves.
- Runs as
- Deployments and StatefulSets, not just Jobs
- Placement
- By region, zone, or individual host
- Ingress
- HAProxy, with TLS on
*.nrp-nautilus.io
# your own hostname, with TLS
kind: Ingress
spec:
ingressClassName: haproxy
rules:
- host: my-app.nrp-nautilus.io
tls:
- hosts: [my-app.nrp-nautilus.io]
---
# running where the data already is
kind: Deployment
spec:
template:
spec:
nodeSelector:
topology.kubernetes.io/region: pacifickubectl apply -f my-app.yaml
Distributed Infrastructure
A global network of research resources: shared compute, data, and services.
Events and cluster activity
7NRP workshop media is now available
Browse the agenda, slides, tutorial materials, and session recordings from the Seventh National Research Platform workshop at UC San Diego.
Live
Latest cluster news
Pulled from Nautilus Support chat. Join the channel for real-time updates.
From the NRP blog
View all posts →Recent events across the NRP and its collaborating projects.
Empowering AI-Enabled Research and Education: NRP Full-Day Tutorial at PEARC26
NRP presented a full-day hands-on tutorial at PEARC26 showcasing AI workflows, JupyterHub, LLM services, and agentic AI on Kubernetes.
Inference as a Medium: A UCSD Art Class Runs Its Ethical-Dilemma Final on NRP
For the final of VIS 145B at UC San Diego, instructor Jon Paden's students built interactive artworks on a single theme — ethical dilemma — each one wired to NRP's LLM inference gateway and deployed on NRP GitLab Pages.
DOE American Science Cloud experiments with NRP — ESnet tutorial at 7NRP
A recap of the DOE American Science Cloud (AmSC) tutorial that ESnet ran at the 7NRP workshop, covering the Genesis Mission context, the AmSC platform, and hands-on IRI Facility API work using NRP's training JupyterHub.
Seventh National Research Platform (7NRP) workshop at UC San Diego - Media
The media collected at 7NRP workshop
Frequently asked questions
What the NRP is, who can use it, and what it costs.
What is the difference between NRP and Nautilus?
The NRP is the platform and the community. Nautilus is the Kubernetes cluster the NRP runs on, and the name you will see in technical documentation and in the cluster tooling itself.
How much does it cost?
Nothing. We are funded by the U.S. National Science Foundation to provide free access to nonprofit research and education institutions.
Does my project fit here?
If your work is part of a scientific project or educational use at a U.S. nonprofit institution, yes. That includes community colleges, not just R1 universities.
What if I don't know Kubernetes?
You do not need it. Hosted JupyterHub, Coder, and our inference API require no Kubernetes knowledge. Our tutorials and bi-weekly office hours cover the rest.
Why NRP rather than a commercial cloud?
It is free to eligible institutions, it is governed by the research community that uses it, and its geographic distribution supports experiments that a single-region cloud cannot.
How can I help?
Contribute hardware to the cluster, or help by spreading the word. Contributors get priority on their own nodes whenever they want them.
What do I get if I contribute hardware?
Priority access to your own hardware whenever you need it. When your group is not using it, other NRP users get opportunistic access.
Do I have to administer hardware I contribute?
No. We install the OS, configure the node, and handle upgrades ourselves; we just need IPMI access. Occasionally we may ask you to reboot a node or swap a drive.
Start computing on the NRP
Bring your research to a platform built and governed by the people who use it.