Solutions
High-Performance Computing Clusters
An HPC cluster is a procurement of many identical things plus one hard interconnect decision. Uniqcli handles both — node build, fabric, parallel storage, rack power and the OS subscriptions on top — delivered racked and cabled to a published layout.
- Category
- Clustered compute for research, modeling and simulation
- We quote
- Nodes, fabric, parallel storage, rack power, OS and support subscriptions
- Delivered
- Racked, cabled and labeled to a published elevation
- Boundary
- The scheduler, the codes and the cluster's operation stay with your team
Uniformity is the requirement everyone underestimates
An HPC cluster is mostly a logistics problem wearing a technical costume. Forty nodes have to be genuinely identical — same firmware, same BIOS settings, same DIMM population, same drive model — because the one that is not will show up as an unreproducible result or a job that runs 20 percent slower for reasons nobody can find. Layered on top of that repetition is a single decision with long consequences: the interconnect. Ethernet or a low-latency fabric, what port speed, how many switches, blocking ratio, and whether the cable runs the design assumes are physically possible in the room you have. Uniqcli quotes the whole cluster as one bill of materials and builds it to a single documented configuration.
What an HPC procurement actually contains
Five categories, one delivery. Missing any of them is how clusters end up commissioned three months late.
- Compute nodes — identical build, firmware-leveled, populated to a documented DIMM and drive map
- Interconnect — adapters, switches, optics or direct-attach copper, sized to the topology and blocking ratio
- Parallel or scratch storage — the tier that decides whether nodes compute or wait
- Head, login and management nodes, plus out-of-band management and console access
- Rack power and containment — PDUs, circuits and airflow sized to the real per-rack draw
- Operating-system and support subscriptions quoted as their own recurring lines
Where clusters are made or lost
Tightly coupled codes — computational fluid dynamics, molecular dynamics, weather and structural solvers — spend a great deal of their wall time in collective communication. For those, latency and consistent bandwidth between nodes matter more than another few cores per socket, and the fabric choice deserves the same scrutiny as the compute. Loosely coupled and embarrassingly parallel work, by contrast, often runs perfectly well on well-designed Ethernet, and paying for a low-latency fabric it never exercises is money that should have gone into more nodes or faster storage.
The honest answer depends on the codes you run, which is why we ask about them before quoting a fabric. What we can do either way is quote adapters, switching, optics and physical cable runs as a matched set against a real rack elevation — including the run lengths, because a 40-meter requirement quietly rules out several otherwise attractive cabling choices.
Storage sits on the same axis. A cluster whose nodes idle waiting on a shared filesystem is a cluster that was sized wrong, and the fix is almost always cheaper before delivery than after.
The cluster, line by line
Sourced through authorized US distribution and screened per line for TAA and NDAA §889 before the quote goes out. Naming a manufacturer describes the market, not a Uniqcli partnership or endorsement.
Compute and head nodes
Lenovo rack platforms built to a single documented configuration and firmware-leveled across the group, with NVIDIA accelerators added where the codes are GPU-capable. Memory populated to a published DIMM map rather than to whatever a configurator defaults to.
Fabric and adapters
Network adapters, switching, transceivers and direct-attach copper quoted against the topology and the physical run lengths in your room. High-throughput adapter lines from NVIDIA sit alongside conventional Ethernet options so the comparison is real rather than assumed.
Scratch and parallel storage
Nexsan arrays and drive shelves, plus Micron NVMe for node-local scratch, sized to the read pattern your jobs actually generate. Quoted with the array's own network path rather than as a box that someone else has to attach.
Cluster OS and support
Red Hat and SUSE enterprise Linux subscriptions and support entitlements, quoted per socket or per node against the term you want. Software, subscriptions and support entitlements are quoted and sourced through authorized US distribution rather than held on a shelf, so stock language does not apply to them.
Delivery
Racked, cabled and documented before it reaches your floor.
Nodes built to one configuration, firmware leveled across the group, racked to a published elevation with a delivered cable schedule and asset record. Commissioning the scheduler and running the codes stays with your team.
HPC procurement questions
Do you install and configure the scheduler?
No. Slurm, PBS, the module environment, the codes and the queue policy are your cluster's operating character, and they stay with your team or your research computing group. We deliver hardware built to one documented configuration with the base operating system installed, which is the part that makes their job straightforward.
Can you match an existing cluster build?
Yes, and it is usually the right call. Send the current node specification, firmware levels and DIMM map and we quote to match rather than introducing a second variant. Where a component has gone end-of-life we say so and propose the closest equivalent instead of silently substituting.
What about vendor HPC lines you do not stock on this catalog?
Some manufacturers' server and networking lines are quoted through our sister storefronts rather than this catalog. Ask for what you need and we route the quote to the right place instead of steering you toward whatever happens to sit here.
Do you handle the facility side — power and cooling?
We quote and integrate the equipment: PDUs, circuits, rack hardware, containment and the airflow assumptions the platform vendor publishes. Electrical and mechanical work on the building itself belongs to your facilities contractor, and we coordinate the equipment specification with them rather than claiming it.
The solutions atlas
Every solution, one accountable partner.
UniQ platforms
By technology
By customer
- TAA & NDAA-889 Compliance Screening
- CMMC & CUI Solutions
- Federal & DoD
- State, Local & Education
- Healthcare
- Enterprise
- Rapid Procurement & GPC Buys
- Multi-Vendor Integration Projects
- eProcurement & Custom Catalogs
- FISMA Modernization
- CJIS-Compliant Justice Cloud & Local AI
- Federal Storage Modernization
- Government ERP & Business Systems Infrastructure
- Managed Procurement
- Secure AV & Conferencing
- Fiber Network Infrastructure
- Satellite & Resilient Connectivity
- Wavelength & Optical Transport
- Decentralized Data Centers
- Data Center Design & Build
Send the node count and the codes
Tell us how many nodes, what the jobs look like and what the room can carry. We will quote nodes, fabric, storage and power as one screened bill of materials.