20.10.2025

Auf Einladung von Prof. Dr. Sebastian von Mammen findet der folgende Vortrag statt:

Montag, 20. Oktober 2025, 14:15 Uhr, Raum 1.012, Zentrales Hörsaal- u. Seminargebäude Z6, Am Hubland

Pirmin Pfeifer und Fabian Wildgrube
AMD, München

One Thread...? Thousands! An Introduction to Modern GPUs

Abstract

Most of us have started their journey into software engineering with high-level, single-threaded languages and a mental model of a singular CPU. However, we quickly learned that modern CPUs actually run multiple threads in parallel. As applications process more and more data (think, e.g., databases) making use of the processing power available becomes ever more important. However, for certain workloads, such as creating 3-D worlds or crunching matrix after matrix in an LLM, even the many cores (and threads) available in modern CPUs aren't nearly enough — and everybody has probably heard that graphics cards (GPUs) are the solution with "thousands of cores" and "incredible performance".

In this talk we will explore what these thousands of "cores" look like under the hood, why they aren't the same as CPU cores and especially what kind of software engineering challenges arise from having to deal with and program for what is effectively a distributed system within your computer. We'll cover the hardware architecture of modern GPUs, traditional approaches for programming them, and introduce a novel programming paradigm that we work on here at AMD.

Join us to take a look behind the buzzwords that usually surround GPUs and gain a better understanding of what they actually are. And get a peek at the kind of work we as software engineers do at a hardware company.