I'm building this out in the open. CUDA and GPU-programming training is in development, and I'm actively working toward NVIDIA DLI certification (Fundamentals of Accelerated Computing with Modern CUDA C++, and multi-GPU). I'd rather tell you exactly where things stand than overstate a polished course. If accelerated computing is on your roadmap, register interest now — it shapes what I build first, and you get priority access.
GPU-accelerated computing sits under modern AI, HPC, simulation, and high-throughput data work — and engineers who can write and optimize CUDA C++ kernels, not just call frameworks, are scarce. This track is aimed at that gap: real parallel programming skill, from the GPU execution model up through multi-GPU scaling. It complements the AI engineering work on the rest of this site — see all training or corporate AI training for team upskilling.
Planned Curriculum
Draft outline — being finalized against NVIDIA DLI's accelerated-computing tracks
1 · Why GPUs
The parallel-computing model, when a GPU wins over a CPU, and where accelerated computing fits in real workloads.
2 · CUDA C++ basics
Host vs device, kernels, threads, blocks and grids — writing and launching your first CUDA C++ kernel.
3 · Memory model
Global, shared, and constant memory, coalescing, and the data-movement patterns that make or break performance.
4 · Profiling & optimization
Nsight profiling, occupancy, memory-bound vs compute-bound kernels, and practical CUDA optimization.
5 · Multi-GPU
Scaling CUDA C++ applications across multiple GPUs — data distribution, streams, and synchronization.
6 · GPU for AI/HPC
Where CUDA meets deep learning and HPC libraries, and how to reason about accelerated pipelines end to end.
Planned Formats
Indicative — final formats confirmed once the track launches
1:1 Private Mentoring
Built around the GPU workloads you actually need to accelerate.
- Live online, paced to you
- Write and profile real kernels
- Async follow-up between sessions
Team Workshop
Get a C++ or ML-infra team productive with CUDA.
- Multi-day, live online
- Hands-on labs on your problems
- Reusable kernel patterns
Prerequisites
- Intermediate C or C++
- Comfort with the command line
- No prior GPU experience needed
Frequently Asked Questions
Is CUDA training available now?
Not yet — this GPU-programming track is in development and I'm preparing for NVIDIA DLI certification. Join the early-interest list and you'll be contacted first when dates open.
Who is it for?
C++ engineers, HPC and ML-infrastructure teams, and systems programmers who want to write and optimize GPU-accelerated code — CUDA C++, kernel optimization, and multi-GPU scaling.
What will it cover?
CUDA C++ fundamentals, the GPU execution and memory model, writing and profiling kernels, performance optimization, and multi-GPU scaling — finalized against NVIDIA DLI's accelerated-computing tracks.
How do I register interest?
Email arjun@arjunthakur.dev or book a free intro call. Mention CUDA / GPU training plus your team size and goals to join the early-interest list.
Related Training
All Training →
Full catalog of AI and backend training programs and formats.
Corporate AI Training →
Team upskilling for engineering and L&D — build in-house capability.
Agentic AI Training →
End-to-end agent stack with LangGraph, tools, and evals.
RAG Training →
Production retrieval: chunking, embeddings, pgvector, evals.