Can I scale multiple deployments with the same labels/selectors at once

[tpfau]

So here is my situation:
I have a somewhat heterogeneous kubernetes cluster with nodes with several different types of GPUs (lets assume A and B). I would like to host some LLMs on that cluster and scale those depending on demand.
However, some of the models I want to provide require more memory than provided by some of the GPUs (e.g. A has only 48GB, while the model needs 60GB )but can be run on multiple of those GPUs, while they can run on a single GPU of type B (which provides 96GB).
I could easily create two deployments for those models, one that specifies the resources and requirements for multiple GPUs of type A and another for type B which only requires one GPU.

And here is my problem/question:
Is there a way to have the Auto-scaler work, such that it scales the total replica count of the combination of Deployments of type A and type B ?
And/or does someone know whether there is a good way to achieve this or whether this is just out of scope for kubernetes?

[linkvt]

Hi @tpfau , not sure if this is still relevant, but I guess this is more of a question about how KEDA scales rather than the HTTP Addon.
I'm not a KEDA expert but as far as I know there is always a single HorizontalPodAutoscaler behind each KEDA ScaledObject, I don't think that this is able to scale multiple deployments.
Maybe the KEDA guys have an idea how you could solve this problem.

Note: Closing this discussion as I'm doing a cleanup right now, let me know if something should be reopened.