To deploy an LLM inference server on Kubernetes, run a serving engine such as vLLM in a Kubernetes workload, make the model files available to it, request the CPU or GPU resources the workload needs, and expose the server through a Kubernetes Service. Configure startup and readiness probes for model-loading time, then verify that the pod becomes ready and can answer an API request. For a direct setup, use native Kubernetes Deployment and Service resources; choose KServe’s LLMInferenceService when you want a declarative serving resource with additional routing and scheduling options.
Choose a deployment path
The right path depends on how much serving infrastructure you want Kubernetes to manage. The vLLM Kubernetes guide documents native Kubernetes deployment as well as Helm and integrations. KServe LLMInferenceService provides a higher-level custom resource for model serving, with options for routing and scheduling. The vLLM production stack is another packaged route, using Helm and documenting Grafana observability.
As an Amazon Associate I earn from qualifying purchases.
| Path | Primary interface | Useful when | Documented capabilities |
|---|---|---|---|
| Native vLLM on Kubernetes | Kubernetes Deployment and Service | You want direct control over a relatively compact serving setup. | The vLLM guide covers CPU and GPU examples, probes, and troubleshooting. It does not establish universal resource sizing or comparative performance. vLLM Kubernetes guide |
| KServe LLMInferenceService | Kubernetes custom resource | You want a declarative model-serving API and the option to configure routing or scheduling features. | KServe documents parallelism, scheduling, gateway and route fields, and multi-node topics. Its example is illustrative, not a general sizing recommendation. KServe overview |
| vLLM production stack | Helm chart | You prefer a packaged vLLM deployment path with dashboard-oriented operations. | The quickstart describes a Helm installation and Grafana observability; it is not proof that a particular setup is production-suitable. vLLM production stack |
These options are not performance rankings: the cited project documentation does not provide a comparable benchmark across them. The rest of this guide uses native vLLM deployment to explain the core Kubernetes workflow.
Check the cluster and model prerequisites
Before creating a workload, confirm that the cluster has nodes able to satisfy the resources you intend to request, and that the chosen vLLM image, model, tokenizer, and accelerator path are compatible with one another. GPU serving also depends on the cluster’s ability to expose and schedule the relevant GPU resource. The KServe runtime overview describes GPU and CPU runtime details; vLLM’s Kubernetes guide documents its CPU and GPU deployment paths.
#1 Best Overall
- 【DeskPi RackMate T2】It's made of aluminum alloy and acrylic frame mini chassis which you can setup your own cluster or home assistant server. For 10 inch 4U Server Cabinet (DeskPi RackMate T0), please refer to ASIN B0DPGZPTPP . For 10 inch 8U Server Cabinet (DeskPi RackMate T1), please refer to ASIN B0CSCWVTQ7 .
- 【10-inch width】The cabinet has a width of 10 inches, which is a relatively small size that saves space while accommodating sufficient equipment. With dimensions of 11.02x10.23x23.22 inches, it is suitable for small offices, home environments, and large enterprises looking to save space.
- 【Open Design】The cabinet adopts an open design, allowing easy access to all devices inside. This design facilitates equipment installation and maintenance, aids in device cooling, and maintains optimal working conditions.
- 【12U Standard】The cabinet has a height of 12U, which is a standard unit size. With 1U equaling 1.75 inches, 12U implies a height of 21 inches.
- 【Translucent Design】Both sides are made of translucent acrylic, providing dust resistance and reduced weight. This design allows direct observation of the cabinet's interior, and users can add ambient lights for decoration.
- Model access: decide how the server will obtain model files, such as from a model URI or prepared storage. If access requires credentials, arrange their availability to the workload using your cluster’s established secret-handling approach.
- Resources: select CPU, memory, and—if applicable—GPU requests based on the model and anticipated workload. The available examples do not establish a universal GPU count or hardware recommendation.
- Version alignment: choose the vLLM release and corresponding image instructions for your deployment. Avoid treating a moving
latestimage tag as a stable production version pin. - Startup time: allow for model initialization before declaring the server ready. Set probe thresholds based on observed startup behavior in your environment rather than assuming the process is ready as soon as its container starts.
Deploy vLLM with native Kubernetes resources
The native pattern is a vLLM server in a Deployment, with a Service providing a stable in-cluster endpoint. The upstream guide includes CPU and GPU examples, but its CPU path is for demonstration and testing: vLLM says, “The use of CPUs here is for demonstration and testing purposes only and its performance will not be on par with GPUs.” Use the current release’s image and source YAML instructions instead of relying on an unversioned copied manifest.
- Prepare model access. Make the model and tokenizer available to the server through the storage or model-access method supported by your chosen deployment. Add credentials if the model source requires them.
- Create the server workload. Follow the vLLM Kubernetes guide’s Deployment example for the selected CPU or GPU path. Set the model and container configuration according to that release’s instructions, and request the resources the workload is intended to use.
- Configure startup and readiness probes. Use the probe approach documented by vLLM, allowing enough time for the model to initialize. Tune the thresholds against actual startup behavior; premature failures can prevent a healthy but still-loading server from becoming ready.
- Expose the workload. Create a Kubernetes Service that selects the vLLM pods and exposes the server port specified by the chosen upstream manifest. The Service is the stable endpoint other workloads in the cluster can address.
- Apply and inspect the resources. Use your normal Kubernetes deployment workflow to apply the manifests, then inspect pod scheduling, readiness, and server logs. If pods remain pending or restart, resolve the reported resource, configuration, or probe issue before adding replicas or routing layers.
Validate scheduling, readiness, and API responses
Do not treat a created Deployment or Service as proof that inference is working. First wait for the pod to schedule and pass readiness after model initialization; then review the server logs for successful startup. The vLLM production-stack quickstart demonstrates checking Kubernetes pod status and sending an OpenAI-compatible API query after installation. Use the endpoint and request format documented for your selected vLLM release, and send a small request to confirm that the server responds.
Rank #2
- 【DeskPi RackMate T1】It's made of aluminum alloy and acrylic frame mini chassis which you can setup your own cluster or home assistant server. For 10 inch 4U Server Cabinet (DeskPi RackMate T0), please refer to ASIN B0DPGZPTPP. For 10 inch 12U Server Cabinet (DeskPi RackMate T2), please refer to ASIN B0DT2XM22G.
- 【10-inch width】The cabinet has a width of 10 inches, which is a relatively small size that saves space while accommodating sufficient equipment. With dimensions of 11x7.8x16 inches, it is suitable for small offices, home environments, and large enterprises looking to save space.
- 【Open Design】The cabinet adopts an open design, allowing easy access to all devices inside. This design facilitates equipment installation and maintenance, aids in device cooling, and maintains optimal working conditions.
- 【8U Standard】The cabinet has a height of 8U, which is a standard unit size. With 1U equaling 1.75 inches, 8U implies a height of 14 inches.
- 【Translucent Design】Both sides are made of translucent acrylic, providing dust resistance and reduced weight. This design allows direct observation of the cabinet's interior, and users can add ambient lights for decoration.
- Pod will not schedule: check node capacity and whether the requested CPU, memory, or GPU resources are available and exposed to Kubernetes.
- Pod is restarting or not ready: inspect container logs and probe events. Check model access and give initialization time consistent with observed startup behavior.
- Pod is ready but requests fail: confirm that the Service selects the intended pods, that the target port matches the server configuration, and that the client is using the API path and request format for that server.
When KServe LLMInferenceService is a better fit
Use KServe’s LLMInferenceService when the team benefits from expressing serving configuration as a model-serving resource rather than managing every serving detail through lower-level workload objects. Its documented resource can represent the model, replicas, container resources, and routing or scheduling configuration. That abstraction adds capabilities, but also means the KServe API and configuration must match the installed KServe version and environment.
Recommended Free Tools
KServe’s overview includes an example with three replicas and one NVIDIA GPU requested per replica, along with model URI and managed gateway, route, and scheduler fields. Those are example values, not a sizing prescription. The same overview describes tensor, data, and expert parallelism and points to autoscaling and multi-node configuration topics; these are separate design choices, not settings every deployment needs.
Rank #3
- Adjustable Depth: 23-40'' adjustable depth is used for servers and network equipment, ensuring enough space for AV equipment, components, and cabling, while allowing you to access ports and equipment from multiple sides.
- Strong Load Capacity: Ground-Mounted Load Capacity: 500 lbs, Wall-Mounted Load Capacity: 150 lbs. The av rack is made of carbon steel for better weldability performance and can help save space while meeting your need to place multiple devices.
- User-friendly Design: Ergonomic design makes the open frame av rack easier to use. The additional top panel is able to place other items with more available space. Roller design moves anywhere and anytime, is convenient, and is more energy-saving.
- Complete Accessories: We provide the accessories you need, including 2 x Pallets, 145 x M5*10 Cross Head Screws, 4 x Casters, 4 x M10*50 Expansion Screws,10 x M6*12 Cage Nuts, 1 x Grounding Wire, 1 x User Manual.
- Wide Application: The server rack wall mount maximizes the use of available space, suitable for retail venues, classrooms, offices, and other places where space is limited.
Scale only after choosing the serving design
Adding replicas and distributing a model across devices or nodes solve different problems. Multiple replicas can provide more serving capacity when the model fits the resources assigned to each instance. Tensor, data, or expert parallelism changes how model work is distributed; multi-node serving introduces further configuration. KServe documents these options, but the appropriate approach depends on model footprint, latency and throughput goals, and measured behavior on the cluster.
Likewise, autoscaling, routing, and observability are platform decisions rather than universal defaults. Start with a working endpoint, then add the controls your operational requirements call for. Use the relevant upstream documentation for the installed versions, and base replica counts and resource requests on measurements from the actual model and workload.
Quick Recap
Best Value
- Adjustable Depth: Depth adjustable from 23" to 40", this open frame server rack accommodates servers and network equipment while providing ample space for A/V gears and cable management. Enjoy easy access to ports and devices from multiple angles.
- High Weight Capacity: Supports up to 300 lbs on the floor (200 lbs when adjusted to maximum depth) and 200 lbs when wall-mounted (depth cannot be adjusted in wall-mounted mode). Made from carbon steel for superior welding performance and durability, this open frame rack is designed to save space while accommodating multiple devices.
- User-Friendly Design: Designed with your convenience in mind, this open frame server rack features an top shelf for extra storage and improved space utilization. The rolling casters let you move it effortlessly wherever you need it, making setup and movement a breeze.
- Widely Applicable: Maximize your space with this adaptable open frame server rack, designed to make the most of every inch. Ideal for retail spots, classrooms, offices, and any area where space is at a premium, it delivers practical solutions for your storage needs.
- Everything You Need: Our open-frame rack comes with fully equipped accessory kit for easy setup and secure installation: 2 x Trays, 4 x Casters, 1 x set of Screws, 16 x M6*12 Cage Nuts, 1 x Grounding Wire, 1 x Internal & External Hex Wrenches, and 1 x User Manual.
Rank #4
- 【Powerful load-bearing】12U Network Rack Open Frame is constructed from durable Cold Rolled Steel; Rack Shelf Back Support enhances stability; load-bearing capacity of 260lbs
- 【Sliding&Considerate】Open-frame layout, including four wheels easy to move, a top panel adding space, anti-slip shelf stops fixing devices and compatible racks for stack and expansion to meet requirements of home server rack
- 【Complete Accessories】A 12U open frame server rack, two ventilated shelves, four shelf stops, four casters, four velcro straps and a set of equipment mounting screws
- 【Versatile Application】Ideal for space-efficient multi-device setups in warehouses, retail, classrooms, offices and more; Excellent choices as AV Rack/IT Rack
- 【Effortless Setup】Server rack with wheels includes hardware, a comprehensive manual, mounting hole drilling template and an online assembly video to simplify setup
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minute
