Define the production envelope
Confirm GPU model and count, node topology, NVLink or fabric, CPU and RAM, storage throughput, network paths, deployment location, start date, term and acceptance thresholds.
Sourcing and operations
One path from demand to running capacity.
Verranum coordinates the commercial, technical and operational layers required to turn available global supply into dependable customer infrastructure.
From contract to first workload
How an allocation is brought online.
The exact runbook follows the selected facility and runtime, but every deployment moves through the same six controlled gates.
Establish network and identity
Exchange customer CIDRs, egress rules and DNS requirements; establish WireGuard, IPsec or private connectivity; then load named RBAC roles, MFA policy and customer-owned SSH keys.
Install the agreed runtime
Apply the accepted Linux image and pin firmware, NVIDIA driver, CUDA and container runtime. Configure either direct bare-metal access, a Slurm partition or isolated Kubernetes GPU workers.
Prove the infrastructure
Run DCGM diagnostics, GPU topology checks, NCCL collective tests, storage and network benchmarks, fault review and the agreed burn-in window before customer acceptance.
Deliver control and observability
Provide endpoints, tenant accounts, scheduler or cluster access, registry path, dashboards, alerts, support contacts, escalation policy and the approved maintenance process.
Launch the first workload
The customer runs a smoke job, verifies data paths and output, confirms monitoring and completes service acceptance. The production operating cadence then begins under the SLA.
Example handover checks
The first commands depend on the selected runtime.
These examples show the shape of a handover, not universal credentials or a promise that every allocation exposes every scheduler.
BARE METAL / BASTION
ssh -J gpu-admin@<bastion> ubuntu@<gpu-node>
SLURM SMOKE CHECK
srun --partition=h100 --nodes=1 --gres=gpu:8 nvidia-smi
KUBERNETES GPU DISCOVERY
kubectl get nodes -l nvidia.com/gpu.present=true
GPU TOPOLOGY
nvidia-smi topo -m
Operating model
We carry the requirement through the full lifecycle.
The engagement stays anchored to the original performance and service objectives from initial sourcing through ongoing operations.
Translate
Convert business and workload needs into a decision-ready infrastructure specification.
Qualify
Evaluate global supply options against technical, location and commercial constraints.
Accept
Coordinate deployment and validate the environment before production handover.
Own
Maintain the service path across support, incidents, changes and renewal planning.
Supply discipline
Asset-light by design. Contract-led by default.
We source from established facilities and supply partners, reserving capacity against contracted customer demand. Infrastructure ownership is considered only where the contracted book and financing case justify it.
A
Back-to-back alignmentSupply commitments are structured around customer term and
demand.
B
Qualified counterpartiesTechnical and operating fit is assessed before a deployment
is proposed.
C
Deliberate headroomGrowth, resilience and change are considered during the
initial design.
Lifecycle
Clear ownership at every stage.
Before service
Requirements validation, supplier qualification, commercial structure, implementation planning and acceptance criteria.
During service
Access administration, service monitoring, change coordination, support routing, incident management and customer communication under the agreed service model.
At change or exit
Capacity expansion, renewal planning, workload transition and secure data handling are coordinated as part of the lifecycle—not left until the final day.
Source with purpose
Bring us the constraint.
We will turn it into a scoped capacity brief and a qualified global sourcing path.