Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

Low GPU utilisation is a serious issue.

Either you massively over-specced your hardware and should have chosen something cheaper and with less power consumption, or your graphics quality is far below what it could be.

Let's not beat around the bush: wasted resources are expensive, in some way or another.



view as:

> Low GPU utilisation is a serious issue.

Maybe? Maybe not? If you have a low GPU utilisation while loading a new level and only displaying a loading bar, that's not a serious issue. If a CAD software has a low GPU utilisation after loading that's not a serious issue. (It just means that the software GUI is written efficiently and can handle complicated assemblies.) If a chat application has a low GPU utilisation that is not a problem, it is simply not the application which calls for full utilisation of the GPU. If your inside-out-tracker has a low GPU utilisation that is not a problem, it just means that you are leaving more space for the user's applications.

But this is not the issue with the analogy. This conversation between you and me, whether or not 5% gpu utilisation is bad or good, or it-depends doesn't just happen here. It happens in everyones head who reads his post. He wants people to think about the organisational inefficiencies of Meta. And a significant portion of his audience is thinking "what is a GPU?". Because you can absolutely be a useful member of the Meta company without knowing that. And then a smaller portion of the audience is thinking "Is 5% GPU utilisation bad?" You could totally understand his points about organisational inefficiency without having to have any understanding of GPU performance metrics.

He lost clarity, for no good reason.


Meta has been heavily criticized on the poor quality of the graphics in VR. In this context poor GPU utilisation is a Very Serious Issue.

Carmac has been working with GPUs at a low level for 25 years. He's going to make a GPU analogy . Frankly, based on organizational dynamics, GPUs map fairly well. Work distribution, caches, warp fronts (aka 'sprints'), instruction sets.


Legal | privacy