Installation and Deployment

NCP-AIO · 62 questions

• A city IT ops lead needs one surface that shows which municipal GPU clusters host overnight public-health batch jobs versus daytime 311 inference, without SSHing into each head node. Which approach best fits that need?
• A transit authority wants one pane for rack health across bus-depot edge boxes and the downtown training cluster. Staff debate Mission Control versus only Base View on a single BCM head node. What should they choose?
• County emergency management asks whether Mission Control replaces BCM for adding a flood-model compute node to the civic cluster. What distinction should ops make?
• City GPU ops need real-time node health and utilization while 311 classification jobs spike each morning. Which BCM capability should they open first?
• Public-health inference nodes show amber in Base View while training nodes stay green during the same window. What is the most accurate ops interpretation to pursue first?
• A water-utility analytics rack drops offline overnight. Before rolling a truck to reseat cables, what should civic ops check first?
• Parks department computer-vision jobs finish early, yet Base View still shows high memory on two nodes. What should ops conclude about those charts?
• Housing authority fine-tuning jobs contend with 311 inference for the same GPUs. How should city ops enforce fair shares through BCM?
• Transit ETA model training must wait until after evening-rush inference. How should that blackout window be encoded?
• County assessor imagery training oversubscribes a node category and starves other municipal teams. What BCM action addresses the allocation problem?
• Emergency-management surge models need GPUs immediately during a flood event, while low-priority parks jobs still hold capacity. What should ops apply via the BCM-managed scheduler?
• After a vendor GPU firmware advisory, city BCM must prepare nodes before Monday’s 311 peak. Which BCM action set is appropriate?
• Two water-analytics nodes diverge from the golden software image before the next SCADA model run. What is the correct remediation focus?
• Library edge boxes need a security patch, but Friday catalog-vision inference must stay up. How should BCM patching be planned?
• Transit depot nodes report mixed CUDA driver levels after a partial update. What should BCM ops do to restore consistency?
• Only public-health ML engineers should submit PHI-touching jobs, while parks CV students must not. How should BCM access be designed?
• A contractor supporting the 311 model needs to watch Base View but must not sync software images. Which BCM permission approach is appropriate?
• A school-district intern accidentally received cluster-wide reboot capability on the education GPU lab. What should admins do?
• Housing authority data scientists share one project login for BCM. Security requires named users with stronger authentication. What change aligns with BCM user administration?
• City BCM must keep RDMA paths healthy for public-health multi-node training across nodes, DPUs, and switches. Where should those network settings be configured and monitored?
• Transit InfiniBand fabric shows flaps after a depot switch change. What should civic ops check first?
• BlueField DPUs on county edge nodes must be ready before DOCA services go live. What BCM-focused step should precede DOCA bring-up?
• 311 inference nodes and training nodes need different VLAN and QoS treatment on the city cluster. How should that be applied?
• A water utility adds a storage VLAN and must update node network settings without breaking GPUDirect-oriented AI data paths. What is the right BCM-oriented approach?
• A flood-forecast job fails with exit code 1 across three nodes. Where should civic ops begin root-cause analysis?
• After UPS maintenance, two permitting-model nodes appear dark. How should civic ops use BCM to diagnose the outage before deciding re-image versus power checks?
• Public-health job backlog grows even though GPUs look idle. What BCM diagnosis should civic ops pursue next?
• Transit ETA jobs fail only on category B nodes. How should civic ops use BCM to isolate the fault?
• The courts OCR queue stalls and BCM shows the head node’s disk is full for scheduler state. What should civic ops do?
• A city wants A100 inference nodes kept separate from H100 training nodes for 311 versus research workloads. What BCM capability should ops use?
• Parks computer-vision jobs and public-health training must not land on the same small-MIG pool. How should BCM enforce separation?
• A county adds L40S nodes for edge VLM pilots. What should ops do in BCM before admitting jobs to the new hardware?
• A school-district GPU lab needs some nodes reserved for district operations and others available as preemptible student capacity. Which BCM approach fits?
• The CIO asks for monthly GPU utilization figures for the smart-city AI budget. Where should civic ops generate usage, performance, and issue summaries?
• After a 311 outage, auditors ask for after-action records. How should civic ops preserve the story in BCM?
• A municipal AI platform team needs Kubernetes on NVIDIA GPU hosts. Which install path best matches NCP-AIO Domain 1 practice?
• After a BCM Kubernetes install, what should civic ops verify before exposing the first public-health inference Service?
• County edge depot sites must match downtown Kubernetes control-plane versions. What install approach should civic ops use?
• A school-district AI lab’s first pods will not schedule GPUs after a BCM Kubernetes install. What incomplete step should ops suspect?
• A city wants BlueField DPU-Arm services to accelerate east-west traffic for multi-node health models. What should ops deploy?
• Transit wants DPU telemetry in place before Run:ai edge workers join depot nodes. Which deployment step should come first?
• Water SCADA analytics hosts show BlueField DPUs up, but DOCA services are missing. What should civic ops correct?
• Emergency-management designs need DOCA on DPU-Arm for storage and network services without changing host GPU MIG layouts. What distinction should architects keep?
• A city platform wants Kubernetes AI workload governance for 311 versus research queues. Which Domain 1 install action fits?
• After Run:ai install, housing-authority projects cannot see GPUs. What should civic ops validate?
• A county is sequencing Run:ai onto a BCM-initialized Kubernetes GPU cluster. When should Run:ai be installed?
• A public-health training farm needs multi-node fine-tuning under BCM. Which scheduler install path matches Domain 1?
• Transit keeps overnight training on Slurm while Kubernetes serves daytime inference. What Domain 1 stance fits under BCM ops?
• After a school-district’s first BCM Slurm install, student jobs see CPUs only. What incomplete configuration should ops suspect?
• A new city AI rack’s first operational step is standing up cluster management. What should civic ops install and configure first?
• Civic NOC wants 311 inference nodes to keep running if one BCM head fails. What should ops configure?
• A county still runs a legacy Bright-style footprint and wants production models soon. What should civic ops finish first?
• An air-gapped city cluster needs a Triton image for 311 inference. How should civic ops deploy from NGC?
• Public-health GPU training needs a supported stack. Which container choice should civic ops make?
• A library CV pilot fails after pulling an NGC container. What mismatch should civic ops check first?
• County assessor imagery burst-trains in cloud while keeping parity with on-prem BCM nodes. What should ops deploy?
• On-prem GPUs are saturated during a wildfire surge. What should emergency management use for extra capacity?
• Transit ops must keep NVIDIA stack parity between depot nodes and cloud burst hosts. What should they prefer?
• City sizes storage for multi-node public-health training checkpoints. What must civic design emphasize beyond raw TB?
• 311 embeddings need fast reads while old transcripts can sit cold. How should civic ops tier storage?
• Water-utility training starves reading large sensor corpora over plain NFS. What storage path should civic design consider?
• Courts OCR farm needs durable datasets, fast scratch, and a model registry. What storage layout should civic ops follow?