Shared inference capacity
Planned inference on SmartTec-owned GPU nodes shared by multiple customers. Scheduling, tenant controls, supported models and performance targets will be validated before service.
Workload planner / From requirements to proposal
Define the job. Design the deployment.
Start with the application and the service your customers need. We turn that brief into a workload validation plan, hardware configuration and commercial proposal, with a clear path through procurement and commissioning.
AI hardware concept · Not installed inventory or an OEM configuration drawing.
Explore the compute concept in 3D
SMARTTEC / SPATIAL CONCEPT3D EXPLORER
Prepare your model, tenancy, storage and service requirements below.
Define the hardware ownership, deployment and support responsibilities for a proposal.
Confirm feasibility and throughput with a representative workload benchmark.
Explore the compute concept.
AI hardware concept · Not installed inventory or an OEM configuration drawing.
Choose a topic, or open the 3D model.
Scene controls
Illustrative hall only. The workload guide prepares requirements for a deployment review; equipment shown does not establish inventory or performance.
01Choose the service
Ownership and tenancy come first.
Dedicated GPU servers
Planned SmartTec-owned GPU servers for one tenant per node. The proposal will define isolation, GPU allocation, software access and support.
Colocation
Planned space, power, cooling and connectivity for servers you own, subject to a separate equipment and facility allocation review.
Shared inference and dedicated GPU servers use different scheduling and isolation designs. Tell us whether you need shared inference or a dedicated single-tenant node on SmartTec-owned hardware, or colocation for your own equipment.
02Prepare the brief
The details that shape a useful proposal.
- Application and models
- Inference, rendering or another workloadInclude model/version, license, precision, context length, expected request mix and the software you intend to run.
- Performance
- Define the customer outcomeLatency, throughput, concurrent requests, batch sizes, scheduled hours and peak periods establish the benchmark to run.
- Data and access
- Storage, networking and isolationInclude dataset size, transfer patterns, private networking, access controls, retention and any compliance requirements.
- Hardware ownership
- Your equipment or SmartTec-owned serversFor colocation, include rack dimensions, weight, power connections and the equipment's cooling specification.
- Service responsibilities
- Agree who operates each layerSpecify software administration, patching, monitoring, support coverage and recovery requirements before a service commitment.
- Commercial scope
- Target date, duration and usage commitmentA proposed start date depends on customer terms, equipment delivery and facility acceptance. An inquiry does not reserve capacity.
03Validate before deployment
A benchmark connects the model to the machine.
Memory capacity alone does not determine throughput or service quality. A representative workload test will establish the supported configuration, scheduling approach and performance envelope. Electrical and cooling review must then confirm that the facility can support it.
The current proposal is 64 B300 GPUs in 8 complete Supermicro systems. The 60 modeled saleable GPUs and four held back are a financial allocation, not a complete standby node or a validated tenant layout. Customer discussions remain unsigned and procurement is ahead.
04Start a conversation
Bring the brief to SmartTec.
Share the requirements you know, and we can work through the rest together. Detailed project economics and their assumptions are available separately in the protected investor area.