title:subscribe
aliases:[newsletter, mailing-list]
tags:[index]
subscribe
one essay a week, in your inbox. usually kubernetes, gpu infrastructure, sre, gcp, and the parts of operating systems that are easier to learn after someone has drawn the map.
enrol
just your email. nothing else.
double opt-in · plain-text friendly · unsubscribe in every issue.
##what you'll get
- one essay when a new piece ships
- occasional hand-drawn explainers and short videos
- practical notes on architecture, operations, and debugging
- early access when a longer series comes together
##recent issues
cuda toolkit: compilers, libraries, and the host-device model
the software stack that makes gpu programming possible - nvcc compiler pipeline, cublas/cudnn libraries, the host-device memory model, cuda execution flow, and the profiling tools every sre should know.
mar 06, 20269 min · gpu, cuda
inside the sm: warps, partitions, and how gpus schedule work
a deep dive into gpu streaming multiprocessors: sm partitions, warp execution, the simt model, thread divergence, and nvlink interconnect - the hardware mechanics behind gpu compute.
mar 01, 202610 min · gpu, sm
gpu performance: bandwidth, throughput, and what the specs actually mean
how to read gpu specs without getting misled. memory bandwidth, tflops, data type precision, tensor cores, and compute capability - what each metric actually tells you about gpu performance.
feb 27, 202610 min · gpu, performance