Management Switching

A dedicated port on the head node connects to a top-of-rack (TOR) switch internal to the cluster, which links to each compute or storage node. Its purpose is to:

  • Send user jobs to your compute resources.

  • Provision compute nodes (management/deployment of OS images).

  • Enable the head node to monitor the cluster’s health and status.

The cluster is also connected to your existing enterprise or private (dedicated) network.

The network:

  • Carries all traffic to and from the cluster.

  • Provides all users and administrators with access to the cluster.

back to top

Internode Communications Switching

High-performance internode communications are provided by either InfiniBand, 10GbE, or 40GbE switching, which can be made available to all nodes.

These switches:

  • Enable compute nodes to use MPI and RDMA.

  • Provide low latency, scalability, and high throughput.

  • Can be supplied in a variety of switch types:

    • 10GbE copper (RJ45, SFP+, or single/multimode fiber optics)

    • 40GbE (QSFP+)

    • InfiniBand QDR (40Gb/s), or FDR (56Gb/s)

back to top

Head Node

One server node (2 for redundant configurations) provide the entry point for cluster administration, monitoring, and management. From the head node, administrators can:

  • Manage the cluster through a unified interface (command-line or GUI).

  • Define hardware usage by deploying job-specific images to node groups or categories.

  • Set up automated monitoring and health checks using BMC/IPMI to monitor variables such as hardware health, heat dissipation, node utilization, and many others.

  • Get insights for planning from metrics/graphs that identify anomalies, high-usage periods, and other key data points.

  • Optimize workloads to keep jobs running smoothly and help ensure internally defined SLAs are met.

  • Avoid incompatibilities by developing and testing applications on the cluster.

back to top

Compute Resources

For compute resources we use x86 cores and can design solutions with GPUs and/or coprocessors. We:

  • Design the solution to meet application requirements, taking into account core-to-RAM ratios, clock speed, power consumption, and more.

  • Can accelerate performance by using GPUs and/or coprocessors where possible, to:

    • Offload repetitive calculations, leaving complex or dynamic jobs to “smarter” standard x86 cores.

    • Execute entire jobs natively on coprocessors.

back to top


Storage is typically provided by NFS servers. We design your solution to:

  • Meet your specific needs for capacity, physical space, backup strategy, future growth, and data tiering.

  • Provide redundancy if necessary (1 or 2 mirrored nodes).

  • Share your existing storage or supplement with new storage.

  • Manage user authentication and authorization.

  • Take advantage of BMC/IPMI and automated health checks.

back to top


Our knowledgeable engineers build your system for the highest levels of availability. They:

  • Size and configure power distribution units (PDUs) for each component and circuit type.

  • Build in redundancy for easy failover in the event of power or PDU failure with:

    • 2 power supplies per server, and 2 PDUs in each rack.

    • Connections from each server to both PDUs.

  • Can design solutions with uninterruptible power supplies (UPS) that can provide power for several minutes, enabling the cluster to shut down properly and avoid data loss.

back to top

The Silicon Mechanics Difference

What’s different about Silicon Mechanics?
We make HPC easy. We know HPC, so you don’t have to.

  • We’ve been building HPC clusters for more than a decade, across many applications, and across multiple industries.

  • We’re brand-agnostic, so we can deliver the best solution for your environment.

  • Our “expert included” support gives you a minimum of three years hardware support.

The biggest difference?  Our focus is on you. It's your HPC cluster—and we'll do everything we can to deliver your turnkey solution, precisely the way you want it Contact us today!

back to top

Commercial Sales:
Education & Research Sales:
Government Sales: