My actual “solution” was to do nothing and just let Kubernetes restart it when it OOMs 😅

My latest example of this was a memory leak in some old software I wrote. It grew linearly with traffic, so it wasn’t until traffic randomly spiked that I started seeing problems.

Laziness is a lazy diagnosis, complexity and ignorance are the more common causes.

An LLM made the logo, if OP is attaching their ego to it that’s their problem.

The dogshit logo let me know this was an LLM project before I read the disclosure.

I’d expect the performance to be awful but it still has relatively niche usecases, especially where performance isn’t a concern. I’m imagining legacy apps that don’t speak S3.

If their progress on GPUs is anything to go by, it will take them at least a decade just to catch up to current DRAM state of the art. And it’s not like there was no incentive to build good GPUs for the last 10-15 years.

Of course they could get lucky, or could steal technologies to accelerate things, but I wouldn’t count on it happening quickly either way.