# You can also use Deployment.options() to change options without redefining
# the class. This is useful for programmatically updating deployments.
SimpleDeployment.options(num_replicas=2).deploy()
```
By default, each call to `.deploy()` will cause a redeployment, even if the underlying code and options didn't change.
This could be detrimental if you have many deployments in a script and and only want to update one: if you re-run the script, all of the deployments will be redeployed, not just the one you updated.
To prevent this, you may provide a `version` string for the deployment as a keyword argument in the decorator or `Deployment.options()`.
If provided, the replicas will only be updated if the value of `version` is updated; if the value of `version` is unchanged, the call to `.deploy()` will be a no-op.
When a redeployment happens, Serve will perform a rolling update, bringing down at most 20% of the replicas at any given time.
(configuring-a-deployment)=
## Configuring a Deployment
There are a number of things you'll likely want to do with your serving application including
scaling out or configuring the maximum number of in-flight requests for a deployment.
All of these options can be specified either in {mod}`@serve.deployment <ray.serve.api.deployment>` or in `Deployment.options()`.
To update the config options for a running deployment, simply redeploy it with the new options set.
Similarly, when Ray Serve scales down and terminates some replica actors, it may result in some nodes being empty, at which point the Ray autoscaler will remove those nodes.
:::
(serve-cpus-gpus)=
### Resource Management (CPUs, GPUs)
To assign hardware resources per replica, you can pass resource requirements to
`ray_actor_options`.
By default, each replica requires one CPU.
To learn about options to pass in, take a look at [Resources with Actor](actor-resource-guide) guide.
For example, to create a deployment where each replica uses a single GPU, you can do the
Deep learning models like PyTorch and Tensorflow often use multithreading when performing inference.
The number of CPUs they use is controlled by the OMP_NUM_THREADS environment variable.
To [avoid contention](omp-num-thread-note), Ray sets `OMP_NUM_THREADS=1` by default because Ray workers and actors use a single CPU by default.
If you *do* want to enable this parallelism in your Serve deployment, just set OMP_NUM_THREADS to the desired value either when starting Ray or in your function/class definition:
```bash
OMP_NUM_THREADS=12 ray start --head
OMP_NUM_THREADS=12 ray start --address=$HEAD_NODE_ADDRESS
```
```python
@serve.deployment
class MyDeployment:
def __init__(self, parallelism):
os.environ["OMP_NUM_THREADS"] = parallelism
# Download model weights, initialize model, etc.
MyDeployment.deploy()
```
:::{note}
Some other libraries may not respect `OMP_NUM_THREADS` and have their own way to configure parallelism.
For example, if you're using OpenCV, you'll need to manually set the number of threads using `cv2.setNumThreads(num_threads)` (set to 0 to disable multi-threading).
You can check the configuration using `cv2.getNumThreads()` and `cv2.getNumberOfCPUs()`.
