Page Header

Resources

Resource Content

GPU Compute Infrastructure

The infrastructure is engineered to support cutting-edge artificial intelligence research and coursework. Students gain managed, enterprise-grade access to NVIDIA’s premier accelerators through Google Colab Enterprise — eliminating setup barriers and enabling seamless training of modern deep learning and large language models.

Infrastructure Overview

Unified Cloud Development Platform

Our cloud architecture bridges theoretical AI foundations with industrial-scale execution. Powered by Google Cloud Platform (GCP) and Colab Enterprise, the platform gives every student immediate, dedicated access to top-tier GPU runtimes directly within standard Jupyter Notebook environments.

Zero-Setup Runtime

Pre-configured with CUDA, PyTorch, TensorFlow, HuggingFace Transformers, and vLLM for immediate coding.

Enterprise Security

Authenticated via Google Account within isolated Google Cloud project environments.

Dynamic Scaling

Seamlessly scale hardware runtimes from lightweight T4 instance testing to flagship H100 distributed clusters.

Persistent Workspaces

Automatic synchronization with Google Cloud Storage buckets to seamlessly store datasets, model checkpoints, and weights.

Hardware Ecosystem

Available GPU Computing Tiers

Access NVIDIA’s computing stack at every stage of your academic journey — from core machine learning assignments to advanced capstone projects.

Standard Tier • Turing Architecture

NVIDIA T4

The ideal workhorse for foundational AI concepts, exploratory data analysis, traditional computer vision, and lightweight natural language processing.

VRAM Memory 16 GB GDDR6
Bandwidth 300 GB/s
Tensor Cores 320 Turing Cores
Optimal Use-Case CNNs, RNNs & Data Pipelines
Default Allocation GCP t4-standard-4
Generative AI Tier • Ada Lovelace

NVIDIA L4

Delivers 4x the AI performance of T4. Purpose-built for modern Generative AI, small-to-medium LLMs, multi-modal workloads, and high-throughput inference.

VRAM Memory 24 GB GDDR6
Bandwidth 300 GB/s (FP8 Engine)
Tensor Cores 240 4th-Gen Cores
Optimal Use-Case LLM Fine-Tuning & GenAI
Core AI Elective Quota GCP g2-standard-8
Enterprise Training Tier • Ampere

NVIDIA A100

Industry standard for heavy deep learning training. Features high-bandwidth HBM2e memory to support large tensor datasets and multi-GPU distributed training.

VRAM Memory 40 GB / 80 GB HBM2e
Bandwidth 1,935 GB/s (1.9 TB/s)
Tensor Cores 432 3rd-Gen Cores
Optimal Use-Case Large Model Pre-Training
Research & Project Allocation GCP a2-highgpu-1g
Flagship Compute
Frontier Architecture • Hopper

NVIDIA H100

The pinnacle of AI acceleration. Equipped with the Transformer Engine and 80GB of ultra-fast HBM3 memory to train multi-billion parameter foundation models.

VRAM Memory 80 GB HBM3
Bandwidth 3,350 GB/s (3.35 TB/s)
Special Engine Transformer Engine (FP8/FP16)
Optimal Use-Case LLMs, MoE & Capstone Research
Capstone & Research Priority GCP a3-highgpu-8g
Student Onboarding

How Students Access Compute Resources

Access to our high-performance infrastructure is fully automated upon enrollment in the MSc AI programme.

1

Project Provisioning

Provide your Google account details at the start of the semester to be assigned to your dedicated Google Colab Project.

2

One-Click Launch

Open your assigned course notebooks directly in Google Colab Enterprise with pre-authenticated GPU quotas.

3

Select Hardware Tier

Choose T4/L4 for standard labs, or request access to A100/H100 runtimes for large-scale capstone experiments.

4

Save & Deploy

Store trained weights to Google Cloud Storage or export models directly for web/mobile deployment.