Get the latest tech news
Learning about GPUs through measuring memory bandwidth
About GPUs Through Measuring Memory Bandwidth At Traverse Research, we need to have a deep understanding of GPU performance to develop our benchmark, Evolve. Additionally, we sometimes do projects for very specific hardware where we need to know all the ins and outs of this hardware.
First we will be going over some background information about GPU hardware relating to loading from and storing to memory, then we will take a look at how our microbench is built, and finally we will look at some GPUs of which we measured the bandwidth and what we learned from that. When writing code we would recommend to use whichever resource type is most natural, but when optimizing for the Meta Quest 3 you might want to replace specific large buffers with textures to get that higher bandwidth. A zoomed in view of the Blackwell Streaming Multiprocessor (SM)The NVIDIA GeForce RTX 5070 Ti we tested uses the same architecture but a smaller version of the chip.
Or read this on Hacker News