Upcoming Livestream: GPUs for AI (Shaped by You)

You can choose the topics for a Live Session on GPUs in AI

Hey everyone,

In the upcoming weeks, I’ll be hosting a live session on GPUs for AI together with Miguel Otero Pedrido (from The Neural Maze), where we aim to unpack the role of GPUs in AI.

GPUs sit at the core of modern AI systems, but for many, they remain a black box.

Every AI workload there is, is powered by an accelerator, GPU, NPU, TPU, LPU, or any other chip, ASIC or not. But how they actually work, what really matters when choosing a GPU, how to optimize the AI workloads and models, are all valid questions that an AI Engineer should ask.

Instead of guessing what to cover, I want to build this session around your interests.

In this article, I’ll be attaching two things:

1. A video overview of the topics I thought about

2. A set of polls, to reason about your interests

Before diving into the polls, which I kindly ask you to take 2-3 minutes to complete, I want to shortly go over the diagrams and charts I’ve prepared, to probe the topics we could discuss, related to AI and GPUs.

---

Starting Points

In this video, I go through a few diagrams and sketches I’ve prepared, in the order I plan to cover them in the upcoming live session.

(Your feedback can change this)

The topics already outlined:

• NVIDIA GPUs

  • What the hardware looks like

  • What is PCIe, and how does it differ from SXM

  • Reading a datasheet, CUDA, NVCC, Cuda Capability

• ASICs (Application-specific integrated circuit)

  • Google TPUs

  • Cerebras Waffer-Scale

  • Groq LPU

(Additional) Topics which might be interesting:

• Optimizing Models on GPUs

  • TensorRT, GGUF

  • ONNX, OpenVino, Apple MLX

---

Your Feedback (helps greatly)

Below are the questions I’ve shared as polls with the audience. Your picks can directly shape the structure, depth, and focus of the livestream.

We’re aiming to transform the livestream into an open discussion, where we could answer all of your questions.

So, feel free to ask ANY type of question during the session.
There are no stupid questions; we want to hear from everyone!

---

1. Understanding your current experience with Software and/or AI.

2. Your understanding of GPUs

The aim is to understand what GPU you’ve been mostly using, to dive into the specific details. The way CUDA works for NVIDIA GPUs, for example, is different than how MPS works on Apple M-series chips, or how ROCm works on AMD GPUs.

3. Topics focused on using GPUs for AI

Most AI Engineers will build applications, optimize and deploy models, or optimize already running pipelines and clusters, if they also work on the infrastructure side.

The goal here is to find out if we should touch on Inference Engines, how LLM (or AI) Inference works, and what the popular Inference Engines are, with brief details on how key components work.

---

One more thing

If you want other answer options that are not present in the polls, please leave a comment on this article with your thoughts, options, and impressions.

I’ll be reading and replying to each comment in part.

I’m sure you’ll learn a lot about how GPUs work, their role and scope in AI development, both during building and deploying models to user-facing applications.

Let’s prepare for an amazing learning session!

See you soon,
Alex