GPU-accelerated AI, in your browser
Powered by NVIDIA DGX Spark and Kubernetes, the KRMU AI Centre of Excellence provides a centralized, GPU-accelerated environment for AI research, development and experimentation.
What we've built
A centralized, multi-user environment for students, researchers and faculty to develop and test AI and machine-learning projects. Each user works in their own notebook workspace, while Kubernetes manages the shared computing resources.
On-premises infrastructure
NVIDIA DGX Spark systems running at KRMU, orchestrated by Kubernetes. Notebook workloads and resources are managed across the GPU cluster.
See the cluster →GPU acceleration
NVIDIA GB10 GPUs. Multiple students can run GPU-accelerated notebooks at the same time, using shared GPU resources managed through Kubernetes.
GPU access by account type →Personal workspaces
An isolated Jupyter environment you open from any browser, with persistent storage for your files.
CPU and GPU notebooks →Three DGX Spark nodes, one GPU pool
Three NVIDIA DGX Spark nodes, each with an NVIDIA GB10 GPU, are joined into one Kubernetes cluster. Notebook sessions are placed across the nodes and share the GPU resources.
Five capabilities at your fingertips
From a first notebook to tracked experiments and stored results.
Personal notebook workspace
JupyterLab in your browser. Choose a CPU or a GPU notebook each time you start your workspace.
Notebook profiles →AI model development
Start developing with a prepared Python and PyTorch environment, GPU acceleration, and tools for machine learning experimentation.
Validated results →Experiment tracking
Track machine learning experiments, record metrics and compare training runs using the integrated MLflow service.
How it works →Storage and archive
Your workspace keeps your files, and S3-compatible object storage supports dataset and model artifact workflows. An archive of your work is available read-only in your workspace.
How it works →Language models
Run and experiment with language models on the platform's GPU infrastructure. Your instructor will announce hosted model access and how to connect.
Read the FAQ →40 accounts, two kinds of GPU access
All 40 platform accounts have been created. Standard accounts get a fixed compute allocation, and super-user accounts can draw more from the shared GPU pool.
| Account type | Accounts | GPU compute |
|---|---|---|
| Standard user | 32 | Fixed compute allocation: one logical GPU allocation in the GPU notebook (up to 4 CPU cores and 16 GiB RAM) |
| Super user | 8 | Can use as much GPU compute as is available in the total shared pool at the time |
Get started in three steps
Your account has already been created. There is nothing to register: use the username and password provided to you by your platform administrator.
Open the login page
The login address for the KRMU AI platform is provided by KRMU IT. Open it in your web browser.
Sign in
Enter the username and password provided to you by your platform administrator.
Choose a notebook
Select a CPU or GPU notebook profile to start your workspace.
CPU notebook or GPU notebook
When you start your server you pick one of two profiles. Use the CPU notebook for coding, data work and small experiments, and the GPU notebook for deep learning.
| Profile | Resources | Best for |
|---|---|---|
| CPU Notebook | Up to 2 CPU cores · 8 GiB RAM · 20 GiB workspace | Python programming, data analysis, preprocessing and lightweight ML workloads |
| GPU Notebook | Up to 4 CPU cores · 16 GiB RAM · 1 logical GPU allocation · 20 GiB workspace | PyTorch, deep learning, model training and GPU-accelerated computation |
- Your workspace keeps your files between sessions (20 GiB capacity).
- An archive of your work is available read-only in the
archivefolder of your workspace. - A PyTorch environment with CUDA support is ready to use in the notebook.
Tested with 30 users at the same time
We started 30 GPU notebook sessions together and ran a continuous GPU workload in all of them for 10 minutes.
Measured with a synthetic GPU benchmark (large matrix multiplications) on the three DGX Spark GPUs. TFLOPS means trillions of operations per second. Each session received a steady share of the GPU, roughly 5 to 12 TFLOPS depending on how many sessions shared the same GPU. Real workloads will differ; this shows how the platform behaves under heavy simultaneous use.
How the system works
From your browser to the GPU, in six parts.
MLflow tracking
Records experiments, parameters and metrics so you can compare runs.
Object storage
S3-compatible storage for datasets and model artifacts.
Archive storage
NAS-backed archive of your workspace, readable from your notebook.
Frequently asked questions
Will my work be lost if I log out?
Can other students see my work?
What is the difference between standard and super-user accounts?
I can't sign in
What if I run out of GPU memory?
How long does training take?
Can I download my trained models?
What is the archive folder?
Can I use a hosted language model?
What if something doesn't work?
Start building
The KRMU AI Centre of Excellence gives you GPU-accelerated notebooks, experiment tracking and storage in one place.
Sign in
Open your workspace on the KRMU AI platform.
Sign in →Create
Develop and train models with GPU acceleration.
Choose a notebook →Track
Record and compare your experiments with MLflow.
How it works →Store
Keep datasets and model files safe in your workspace and object storage.
How it works →