We should add compaction logic before deciding whether to allocate new memory for a new GPUBindGroup.