* **alpha**: NVIDIA GPU support ([#24836](https://github.com/kubernetes/kubernetes/pull/24836), [@therc](https://github.com/therc))
  * Adding loadBalancer services and nodeports services to quota system

## Known Issues and Important Steps before Upgrading

* Add an Azure CloudProvider Implementation ([#28821](https://github.com/kubernetes/kubernetes/pull/28821), [@colemickens](https://github.com/colemickens))

The CustomCFSQuotaPeriod alpha feature enables node administrators to change the default period used to enforce CFS quota on a node.  This can improve performance for some workloads that experience latency while using CFS quota with the default measurement period.  Finally, the SIG continues to focus on improving reliability by fixing bugs while working out design and implementation of future features.

For more information, see [#53040](https://github.com/kubernetes/kubernetes/issues/53040).

* Object quotas are not consistently charged or updated. Specifically, the object count quota does not reliably account for uninitialized objects. Some quotas are charged only when an object is initialized. Others are charged when an object is created, whether it is initialized or not. We plan to fix this issue in a future release.

- Fixed a potential deadlock in the resource quota controller. Enabled recording partial usage info for quota objects specifying multiple resources, when only some of the resources' usage can be determined. ([#74747](https://github.com/kubernetes/kubernetes/pull/74747), [@liggitt](https://github.com/liggitt))

* kube-controller-manager can now start the quota controller when discovery results can only be partially determined. ([#67433](https://github.com/kubernetes/kubernetes/pull/67433), [@deads2k](https://github.com/deads2k))

* kubelet initial flag parse now normalizes flags instead of exiting.  ([#61053](https://github.com/kubernetes/kubernetes/pull/61053), [@andrewsykim](https://github.com/andrewsykim))

* Fixed regression where kubelet --cpu-cfs-quota flag did not work when --cgroups-per-qos was enabled ([#61294](https://github.com/kubernetes/kubernetes/pull/61294), [@derekwaynecarr](https://github.com/derekwaynecarr))

if supported by the kernel. You should either adjust your CPU limit, or set CPU
request only, if you want to avoid hardcapping. If the kernel does not support
CPU Quota, NodeStatus will contain a warning indicating that CPU Limits cannot
be enforced.

### API Change

- 1. PodAffinityTerm includes a namespaceSelector field to allow selecting eligible namespaces based on their labels. 

* Moved qos to api.helpers. ([#44906](https://github.com/kubernetes/kubernetes/pull/44906), [@k82cn](https://github.com/k82cn))

* Fix issue with the resource quota controller causing add quota to be resynced at the wrong ([#45685](https://github.com/kubernetes/kubernetes/pull/45685), [@derekwaynecarr](https://github.com/derekwaynecarr))