Two Gaming Graphics Cards Started the AI Boom

Why does AI run on graphics cards? Part 5 of How AI Works (Advanced): how two gaming GPUs in a bedroom won ImageNet in 2012 (AlexNet), why AI's maths has the same shape as game graphics, how CUDA opened GPUs up, how 3 GPU machines matched 16,000 CPU cores, what's inside a modern AI chip (tensor cores, H100), why memory speed is now the real bottleneck, and chips built only for AI (TPU). ⏱ Chapters 0:00 Two gaming cards in a bedroom (AlexNet, 2012) 1:11 1. Made for pixels 2:21 2. AI is a mountain of onions (matrix maths + CUDA) 3:31 3. 16,000 cores, or 3 machines 4:41 Quick one 4:52 4. Inside a modern AI chip (H100, tensor cores) 6:11 5. The memory wall 7:55 6. Chips made only for AI (TPU) 8:53 What this means for you 9:28 Your turn 9:35 Recap ✅ GPUs and AI in one minute 1. GPUs were built to compute millions of independent pixels at once 2. AI is giant grids of multiply-add, also independent: the same shape 3. CUDA (2007) opened GPUs to any maths; AlexNet (2012) proved it 4. Modern AI chips add tensor cores that multiply whole grids at once 5. The real bottleneck is memory speed, not maths: the "memory wall" 📚 How AI Works (Advanced) playlist:    • How AI Works (Advanced)   📚 How AI Works (Intermediate) playlist:    • How AI Works (Intermediate)   📰 Sources Krizhevsky, Sutskever & Hinton, "ImageNet Classification with Deep Convolutional Neural Networks" (2012); Computer History Museum (2025) NVIDIA, CUDA announcement (2007); H100 product page and architecture whitepaper (2022) Le et al., "Building High-level Features Using Large Scale Unsupervised Learning" (2012) Coates et al., "Deep learning with COTS HPC systems" (2013) Gholami et al., "AI and Memory Wall" (2024) Jouppi et al., "In-Datacenter Performance Analysis of a Tensor Processing Unit" (2017) 💬 What graphics card do you have, and have you tried running AI on it? ▶ Watch next: How to shrink an AI model:    • Stanford Copied an AI for Under $600. Here...   🔔 One honest AI test every week:    / @taiphambuilds   Notes: The tokens-per-second figure is our rough worked example. Bedrooms, kitchens, pipes, card stunts and chip diagrams are illustrations. Narration is an AI voice. Music: "Stake Out", "Majestic-12" and "Fire In The Sky" by Alex Jones / Xander Jones (YouTube Audio Library). 🌐 https://phamphutai.com #ai #gpu #nvidia #alexnet