How GPUs work
Covers GPU vs CPU: The Core Difference, Parallel Processing and Thousands of Cores, GPU Memory Hierarchy and Bandwidth and more.
“How GPUs work” is a 4-lesson course at Teen level, created with soclever, a personal AI teacher. Each lesson takes a few minutes and ends with a check-in question; finish the curriculum and you can take a certificate test to earn a diploma. Try the first lesson without an account — or make your own course on any topic. Shared by @meirrosenschein.
What you'll learn
- GPU vs CPU: The Core Difference. Why does your phone's camera app open instantly but rendering a video takes minutes, even though both do math on the same device?
- Parallel Processing and Thousands of Cores. Without the ability to perform the same operation on many data elements at once, rendering a single video frame would take seconds.
- GPU Memory Hierarchy and Bandwidth. Think of a busy restaurant kitchen.
- SIMT Warps and Tensor Cores Explained. Picture a school bus driver who takes a whole group of kids to the same stop at the same time. It is far easier than sending each kid off on their own route.
Lesson 1: GPU vs CPU: The Core Difference
Why does your phone's camera app open instantly but rendering a video takes minutes, even though both do math on the same device?
A CPU (central processing unit) is the main processor in any computer. It has a small number of cores—typically 4 to 16—and each core is highly optimized to execute one task at a time, very quickly, in a strict sequence. A GPU (graphics processing unit) is a specialized processor designed for a different job: it has thousands of small cores, each slower than a CPU core, but all working in parallel on the same type of problem simultaneously.
The key difference comes down to what each processor was built for. CPUs excel at branching logic and sequential decisions—the kind of work where you must do step A, then check the result, then decide whether to do step B or C. Your phone opening an app involves countless conditional checks: Is the file there? Is it corrupted? What permission does the user have? A CPU handles this perfectly because it can make quick decisions one after another.
GPUs were originally designed to render graphics, which involves a repetitive operation: take a pixel, apply a color formula, write it to the screen, repeat for millions of pixels. Every pixel gets the same operation, just with different inputs. This is called parallel processing—thousands of cores doing nearly identical work at once.
Video rendering is essentially millions of pixel operations per frame. A GPU can split that work across thousands of cores and finish in seconds what a CPU would take minutes to complete alone. The camera app, by contrast, is mostly decision-making and file management—CPU work—so the GPU sits idle while your CPU handles it.
Now you can answer the opening question: the camera app is fast because it's sequential logic, which CPUs dominate. Video rendering is slow on a CPU because it's repetitive math across massive datasets, which GPUs were built to crush.
Questions this course answers
What is a CPU?
A CPU is the main processor in any computer that has a small number of cores and is highly optimized to execute one task at a time, very quickly, in a strict sequence.
What is a GPU?
A GPU is a specialized processor designed with thousands of small cores, each slower than a CPU core, but all working in parallel on the same type of problem simultaneously.
What does parallel processing mean for a GPU?
Parallel processing is when thousands of cores do nearly identical work at once by executing identical instructions across processor cores in lockstep.
What is VRAM?
VRAM is the GPU's main memory, acting like a big pantry that holds things like game textures, images, and the numbers of an AI model.
More courses to explore
- How Does ChatGPT Actually Work — 4 lessons (Teen level).
- How Does Bitcoin Actually Work — 4 lessons (Teen level).
- Israeli elections — 4 lessons (Teen level).
- Who is lebron james — 4 lessons (Teen level).