Inside a modern GPU: from architecture to trillions of calculations

We often talk about graphics card performance in staggering numbers—trillions of calculations per second and a capacity that just keeps growing. But what is really happening beneath the surface? A video from Branch Education dives into the machinery itself, using the GA102 chip in the Nvidia RTX 3090 as an example to break down how a modern GPU actually works. The starting point is as simple as it is fascinating: how can a single component perform tens of trillions of operations every second? In the most advanced models, it's over 36 trillion calculations per second—and development continues to point upward. The explanation lies in how GPUs are built. Where traditional CPUs work step by step, GPUs are designed to do many things at once. Instead of one powerful core, it's about thousands of smaller units working in parallel. The result is an enormous capacity to handle large amounts of data in real time. This working method is often described as Single Instruction, Multiple Data (SIMD)—a model where the same instruction is executed across many data points simultaneously. Combined with high-speed memory like GDDR6X, an architecture is created that is particularly suited for data-intensive tasks. It's not just about graphics. The same principles today power everything from game rendering to advanced AI calculations and simulations. Rather than a single technical breakthrough, the video shows that it is the interplay between many parts that makes the difference. It is in the whole that the GPU's true strength emerges, explaining its increasingly important role in modern high-performance computing