Bifalgorithm post

If you're not dumb in 2026, you would have definitely heard about GPUs, CPUs and TPUs. But do you even know the difference between a CPU, GPU, TPU and NPU? Let me try to make it more simple and interesting for you. If you can see it over here, AI entirely works on massive matrix multiplication, which runs over and over over neural networks. Now AI does all of these matrix multiplications because it's an efficient way to compute large number of mathematical operations to be done simultaneously. Now be it GPU, CPU, TPU, it is exactly... the same math, just a different specialization and different circumstances. So let's start with CPU. We are reading the CPUs from childhood. So it's a master chef and it's a very skilled master chef. It can cook anything, for example, pasta, biryani, pizza, whatever you want. But the limitation that it has is that it can only do one thing at a time. CPU is built to handle anything, for example, loading app, running desktop versions, running any codes, even clicks, which means it can do a lot of things, but it cannot do a lot of computations simultaneously. So if you want to make one pizza, you can make it. But if you want to make a thousand pizzas, it will do it one by one, which is going to take forever. Which is exactly where the Jeep... GPUs come into the picture. Imagine GPUs equal to the kitchen assembly line. Now imagine thousands of simple workers just basically on an assembly line, one chopping the tomato, one spreading the cheese, one just cutting other vegetables. So one person doing one task. Same thing over and over again and all of them doing the stuff simultaneously. GPUs were originally built to render video games via thousands of tiny cores applying the same operation to millions of pixels at once. Even though originally GPUs were run to just render video games but now what is actually guessed it right AI. It does not have few powerful cores. It actually has thousands of simpler ones that all run same operation at thousands of pieces of data at the same time. It still blows my mind to just imagine that the whole future of artificial intelligence is relying on the hardware that was originally just built to render video games. Now imagine you want to use a calculator. So you can buy a 100-Rupiah calculator or you can buy a 1,000,000,000 They thought this is very flexible, AI just needs matrix multiplication. Why not just make devices that can do just that? And that's when TPU comes to the picture, which is a Tensor Processing Unit. It is a specialized factory, which is one task, which is faster matrix multiplications. It uses systolic array architecture that pipelines data through a grid of multiply accumulating units with extreme efficiency. Now think of this like a pizza factory, a factory that makes only pizza, but makes the fastest producing pizza. When Google actually analyzed their GPUs, they found that there are 15 to 30 times faster and it is 80 times power efficient than gpus and cpus tpus are only available via google cloud and they are used to train some of the largest ai models and when i was researching about all of these things one question did come to my mind which is if tpus are so much more power efficient than the gpus why are they not dominating the entire market and i did get an answer to it and if you want me to make a video on that just let me know in the comments and follow me for all of

If you're not dumb in 2026, you would have definitely heard about GPUs, CPUs and TPUs. But do you even know the difference between a CPU, GPU, TPU and NPU? Let…