Home > AI Solutions > Artificial Intelligence > Guides > Design Guide—Digital Assistant on Dell APEX Cloud Platform for Red Hat OpenShift with Red Hat OpenShift AI > Physical architecture
The following figure demonstrates the physical architecture design of the solution used to run the LLM-based application on Dell APEX Cloud Platform for Red Hat OpenShift.
The following sections describe the various hardware stacks involved in designing this solution.
Dell APEX Cloud Platform for Red Hat OpenShift compute layer setup is configured with four MC 760 nodes for running AI workloads. MC 760 server is a 2U, two-socket fully featured enterprise rack server, designed to optimize even the most demanding workloads like artificial intelligence and machine learning. It provides a broad choice of core densities using Intel's 4th gen Xeon scalable processors.
MC 760 Servers offer:
For more information about Dell MC 760 servers, see the Dell APEX Cloud Platform MC 760 Hardware Requirements and Specifications page. This solution includes MC 760 servers with Intel(R) Xeon(R) Gold 6430 processors which offer 32 cores per socket, operating at a speed of 2.10 GHz. Additionally, servers are equipped with 2 X NVIDIA L40S GPU. The NVIDIA L40S GPU is a powerful data center GPU, which comes with 48 GB GDDR6 memory and supports PCIe Gen4 x16: 64 GB bi-directional interconnect interface, delivering end-to-end acceleration for the next generation of AI workloads such as generative AI, LLM inferencing and training.
Block Storage: Dell PowerFlex is a software-defined infrastructure that delivers consistent predictable outcomes at large scale for the most demanding mission-critical environments. In our setup, four dedicated PowerFlex nodes are configured as storage nodes to provide persistent block storage for workloads, such as Vector store, OpenShift AI workbench and data science pipelines. The Dell Container Storage Module (CSM) enables simple and consistent integration and automation experiences, extending enterprise storage capabilities to Kubernetes for cloud-native state applications.
Object Storage: Dell ObjectScale is a high-performance containerized object storage that is designed for scalable and cost-effective data management and provides an ideal solution for housing massive LLMs, large datasets, and pipeline artifacts. As shown in the solution architecture, Dell ObjectScale is deployed separately which stores the Llama 2 model, other datasets, and Red Hat OpenShift AI pipeline artifacts.
Two 25 GbE Dell PowerSwitch network switches are being used for data plane connectivity, and two 10 GbE Dell PowerSwitch network switches are being used for management plane connectivity. Dell PowerSwitch enhances performance in data center fabrics of all sizes with efficient and flexible switching solutions.
These nodes are used to run VMs, and containers related to client application and tools required to manage the Red Hat OpenShift cluster, Object Storage, and LLM-based application. Also, infra VMs such as AD and DNS are running in these nodes.