On a cluster with two autoscalers defined (one with node flavour p3.8xlarge - 4 gpus per node - and one with node flavour g4dn.xlarge - 1 gpu per node-) I've tried to trigger a scale up by requesting 1 GPU.
The node that was scaled up was the multi-gpu one (p3.8xlarge) even though the smaller variant would've been enough to satisfy my request.

OCP 4.11

Always in my testing

1.Define two autoscalers, in my case one with single-gpu nodes and one with multi-gpu nodes
2. Try to trigger a scale up by requesting 1 gpu

multi-gpu node is scaled up to fulfill my request of 1 gpu

single-gpu node should've been scaled up, since that would've fulfilled my request