From chapter-1 I learnt a few things which I will be sharing (i made notes in @hackmdio which I will be sharing soon after I complete the book) > it talks about why we need to understand CUDA and not sit behind hidden box calls like tensor.cuda(), it’s like learning a manual car vs automatic car, both work equally well but if one breaks down we know what and how to fix > got to know about the power wall and how post 2005 no free lunches were served as any increase in GPU clock speed meant increase in heat and power wasn’t constant and hence the cooling systems didn’t catch up, which made the switch to multi processor systems > CPU processors and GPU processors work differently, one is made for fast latency and another for more throughput > GPUs have smaller ALUs and more of them and smaller controls and cache and high memory bandwidths to transfer things up and down the lane > data parallelism vs task parallelism, GPU tackles the first as CUDA follows SPMD model where one program operates on multiple data > group of 32 threads is called wraps, GPUs make sense when the data they are working with is large as for smaller ones the transportation costs are larger > learnt about amdahl’s law on how the sequential portion limits speedup when the problem size is fixed. Gustafson’s law shows that parallelism can be harvested by growing the problem size to utilise it

anthropic still not adopting agents.md is straight nonsense lol

Summer has been extended.. On ddocs.new 🌴💁‍♀️

Do you remember these? • index.html • styles.css • script.js

Can we all just pause for a sec and reflect on how fucking important Tailscale is, and how nobody ever talks about them? I don't think I've ever seen a more useful service with a more invisible corporate footprint.

Book Mode is rebuilt add, reorder, and nest notes straight from the sidebar no syntax to memorize, no refresh to remember the notes were never the hard part, the table of contents was