Let Developers Code, Not Configure: CMKL’s Glider Minimizes Infrastructure Headaches
Let’s face it—much of the frustration in AI research doesn’t come from building models. It comes from waiting for computing resources, troubleshooting server compatibility, and fixing broken dependencies that somehow only exist on the remote machine.
At CMKL University, we believe that kind of friction shouldn’t stand in the way of innovation.
Enter Glider: our self-hosted, streamlined platform for GPU-powered AI development, built with one mission in mind:
"I want developers or researchers to be able to focus on writing code rather than managing infrastructure." — Permpoon Boonyarit, CMKL System Engineer
No fluff. No vaporware. Just GPUs, code, and a browser.
Can We Use the Existing HPC System for AI Workflows?
For years, we’ve relied on systems like Slurm, a robust and trusted scheduler for batch computing. It works well for traditional HPC jobs—running simulations, submitting long training scripts, or working with pre-tested software.
But when it comes to interactive AI development—debugging a transformer model, tuning hyperparameters, or performing real-time inference—Slurm is no longer a one-size-fits-all solution.
Are there alternatives?
Cloud platforms such as AWS or GCP offer flexibility, but often at a high cost. They introduce complexity, billing sprawl, and vendor lock-in that many researchers and students can’t—or shouldn’t—have to deal with.
So, we built our own.
Meet Glider: Click. Code. Conquer.
Glider is CMKL’s response to this challenge—a platform designed to reduce infrastructure barriers and let users focus on what matters.
It offers a clean, browser-based interface that abstracts away the complexity of Kubernetes while delivering full GPU-powered environments. Among the tools currently supported:
- CodeServer: Run VS Code directly in your browser
- JupyterLab: Ideal for data science and hands-on teaching
- vLLM: Lightweight inference for large language models
- NeMo: NVIDIA’s end-to-end platform for generative AI development
And most importantly: it works.
Log in, choose your workspace, and within minutes you’re writing code with GPU acceleration. No YAML, no SSH, and definitely no crying.
Built by Developers Who’ve Seen It All
The team behind Glider understands the pain points—because they’ve lived them.
“From my experience working on an infrastructure team, setting up and managing an environment that supports developers’ work is one of our most important responsibilities.” — Permpoon Boonyarit, CMKL System Engineer
Glider was born out of that mission. Instead of endlessly troubleshooting container setups and explaining virtual environments, we built something that just works—for students, researchers, and faculty alike.
This isn't just a platform. It’s a declaration: infrastructure shouldn’t stand in the way of creativity.
Can Students Actually Use It?
Yes—and they already are.
Undergraduates now have access to GPU-backed development environments in seconds. Researchers can deploy models without spinning up cloud instances. Instructors finally have a tool that makes AI education easier, not harder.
And Glider is designed to be inclusive—it’s not just for computer science majors. Anyone exploring AI can use it, regardless of their technical background.
What’s Next?
Glider is still evolving. Features in development include:
- Multi-cluster support
- Sandboxed inference
- Full VM provisioning for enhanced isolation and licensing use cases
This is not a one-off experiment. Glider is steadily becoming the alternatives compute environment at CMKL University, tailored to the real needs of its users.
Final Thoughts
Innovation doesn’t wait for dependency resolution. It shouldn’t be limited by queue systems or buried under sysadmin overhead.
With Glider, CMKL is building an AI development ecosystem where infrastructure becomes invisible—and all that’s left is possibility.
Whether you're a student, researcher, or educator, the future of GPU-backed coding at CMKL is just a browser tab away.
