0x137E32
So You Want Low-Latency Object Detection
How a 3-second lag taught me everything I didn't know about real-time video pipelines.
SYS::IDX
Explore all posts
Filter by topic and view posts from newest to oldest.
0x85DABF
Optimizing PyTorch Data Pipelines: From Bottlenecks to 39× Speedups
I went to revisit the PyTorch training loop, and ended up chasing a bottleneck that had nothing to do with the model.
0x9213ED
CUDA Performance: What Actually Matters
Shared memory, coalescing, occupancy — the techniques that actually make GPU kernels fast. And the benchmarks that show when they don't.
0xC3829C
CUDA Is Not About Speed — It’s About Thinking Differently
Understanding GPUs is less about APIs and more about changing how you think about computation.