NVIDIA NCP-AII Questions & Answers

The option would apply to “During HPL burn-in one GPU's temperature and power-limit counters increase but the aggregate benchmark still finishes”. A completed remediation must demonstrate that Per-device telemetry stabilizes and repeated HPL throughput remains within approved spread. This question instead requires “Replayed tests with matching versions produce stable, interpretable timing and error data”.

The related but distinct situation is “A one-minute HPL run is used for a rack acceptance contract that requires proving sustained computation under design power”. The decisive evidence is that The HPL campaign maintains correct output and accepted thermal and ECC trends throughout the interval. For the stated problem, “Replayed tests with matching versions produce stable, interpretable timing and error data” addresses the central condition.

This alternative addresses the separate case “An HPL stress campaign produces high performance but occasionally fails validation residuals after several hours”. The strongest acceptance result is that Repeated sustained HPL jobs pass both performance and numerical residual acceptance, not merely that a command executed. The deciding requirement here is “Replayed tests with matching versions produce stable, interpretable timing and error data”.

This option serves the scenario “A single-node burn-in completes briefly but temperature rises steadily and GPU clocks drop after thirty minutes”. A completed remediation must demonstrate that The long-running stress maintains expected clock behavior without rising error rates or out-of-policy thermals. This question instead requires “Acceptance fails for any nonzero critical hardware error even if the compute test process exits normally”.

This would be a meaningful test for another issue (A new server passes isolated GPU tests but fails when eight GPUs and all NICs are stressed concurrently), but not for the required outcome: Acceptance fails for any nonzero critical hardware error even if the compute test process exits normally.

The candidate evidence checks Stress every installed accelerator simultaneously and reconcile per-GPU health and performance results in another context. Here the specific acceptance result is Acceptance fails for any nonzero critical hardware error even if the compute test process exits normally.

This option serves the scenario “An HGX server has eight SXM accelerators but a runbook treats it as eight independent PCIe cards for service procedures”. Therefore Use the platform-specific HGX service and identification documentation before installation or replacement. The present scenario is distinguished by “Align the CUDA runtime and NVIDIA driver with the documented compatibility requirements for the workload image”.

This response is applicable to a different failure: A new NVIDIA GPU server boots Linux, but the operator never checks whether all expected PCIe accelerators enumerate. Here the central dependency is that CUDA applications rely on a compatible host driver even when user-space toolkit components are installed, so the specific response is Align the CUDA runtime and NVIDIA driver with the documented compatibility requirements for the workload image.

The option would apply to “An operator runs an AI application on a fresh server but does not record the GPU driver version or actual compute device selection”. The direct fix is Record device identity, supported driver, GPU health and selected CUDA device mapping during software commissioning. This question instead requires “Align the CUDA runtime and NVIDIA driver with the documented compatibility requirements for the workload image”.

This alternative addresses the separate case “A team reports healthy storage throughput from the login node but GPU compute nodes use different mounts and client networking”. The decisive evidence is that A scheduled multi-node storage test returns valid data and expected throughput for every participant. For the stated problem, “An end-to-end checkpoint test reaches the recovery-time target under production-like concurrency” addresses the central condition.

The option would apply to “All compute nodes can list training files but concurrent training readers experience extreme latency spikes from the shared storage array”. The strongest acceptance result is that A concurrent reader benchmark meets approved throughput and tail-latency targets without I/O errors, not merely that a command executed. The present scenario is distinguished by “An end-to-end checkpoint test reaches the recovery-time target under production-like concurrency”.

The proposed validation does not isolate the reported fault. An effective closure test requires evidence that An end-to-end checkpoint test reaches the recovery-time target under production-like concurrency.

The related but distinct situation is “A data center imports server settings from a reference rack without checking whether the new deployment uses different network and storage attachments”. This case illustrates that Identical server models do not imply identical cabling, fabric identities or storage security boundaries at each site. For the stated problem, “One completed workload may avoid weak links and says little about redundant paths or intermittent component failures” addresses the central condition.

This option serves the scenario “A contractor declares a new GPU cluster ready based only on successful PXE boot while half the server NICs lack documented fabric connections”. The controlling explanation is that A successful OS boot proves provisioning reachability but does not prove all AI workload data paths have been commissioned. The present scenario is distinguished by “One completed workload may avoid weak links and says little about redundant paths or intermittent component failures”.

This alternative addresses the separate case “An AI factory rack has completed electrical installation, but technicians schedule cluster burn-in before BMC addresses and power checks are signed off”. The controlling explanation is that A benchmark cannot establish deployment readiness while foundational power, management and hardware dependencies remain unverified. The deciding requirement here is “One completed workload may avoid weak links and says little about redundant paths or intermittent component failures”.

The related but distinct situation is “A BlueField upgrade finishes on the host OS but its Arm environment and NIC firmware remain at a different unsupported release”. Therefore Use the validated BlueField software and firmware compatibility process across host, Arm environment and adapter components. The deciding requirement here is “Collect protected metrics from the authorized BlueField-side management interfaces with documented host limitations”.

This alternative addresses the separate case “A GPU server's DPU has an approved high-speed physical link but no functioning OOB path for emergency firmware recovery”. The appropriate intervention is Document and test the BlueField management path through its authorized Arm or controller access independently of data traffic. For the stated problem, “Collect protected metrics from the authorized BlueField-side management interfaces with documented host limitations” addresses the central condition.

For this finding, Apply supported BlueField host-restriction settings from the authorized Arm or BMC management side is less direct than Collect protected metrics from the authorized BlueField-side management interfaces with documented host limitations, because Zero-trust restrictions may intentionally block host-side hardware counters or tracing functions.

This would be a meaningful test for another issue (An installer forgets one GPU power connector after servicing a PCIe server and the GPU appears intermittently under load), but not for the required outcome: PCIe readback and an adapter throughput test match supported hardware capabilities without recurring errors.

This alternative addresses the separate case “A node's BIOS reports a NUMA topology different from the approved design after a firmware reset”. The decisive evidence is that Topology reports and controlled bandwidth tests align with the approved baseline after configuration correction. The deciding requirement here is “PCIe readback and an adapter throughput test match supported hardware capabilities without recurring errors”.

The option would apply to “A newly deployed node passes a CPU-only stress test but its GPU accelerators fail when exposed to simultaneous high-memory loads”. A completed remediation must demonstrate that Simultaneous GPU memory tests complete within errors and thermal limits across all installed devices. This question instead requires “PCIe readback and an adapter throughput test match supported hardware capabilities without recurring errors”.

Another failure mode is possible in general, but the stated condition most directly creates the exposure that Unverified boot settings can allow untrusted components to load without visible evidence.

The option would apply to “An installer disables the BMC firewall to make Redfish accessible from every office subnet during rack acceptance”. The direct failure mode is that Excessive BMC exposure can enable unauthorized shutdown, firmware changes or sensor tampering. The present scenario is distinguished by “Unverified boot settings can allow untrusted components to load without visible evidence”.

The related but distinct situation is “A technician changes the BMC clock manually but never configures time synchronization, causing alert timestamps to disagree with cluster logs”. Without correction, Clock drift can prevent correct sequence analysis during power and firmware incidents. The deciding requirement here is “Unverified boot settings can allow untrusted components to load without visible evidence”.

This response is applicable to a different failure: A server with MIG partitioning passes independent tenant GPU kernels but the team assumes every MIG instance supports arbitrary cross-instance NVLink collectives. Here the central dependency is that Algorithm bandwidth and bus bandwidth represent related but different collective throughput normalizations, so the specific response is Use the metric definitions appropriate to the collective and compare with the same topology and message sizes.

For this finding, Use rank-by-rank and pairwise topology validation to isolate the failing accelerator or NVLink/NVSwitch path is less direct than Use the metric definitions appropriate to the collective and compare with the same topology and message sizes, because Algorithm bandwidth and bus bandwidth represent related but different collective throughput normalizations.

This alternative addresses the separate case “NCCL single-node all-reduce bandwidth falls after maintenance even though each GPU still passes a separate CUDA sample”. The appropriate intervention is Compare GPU topology, NVLink/NVSwitch health and NCCL software versions against the known baseline. The deciding requirement here is “Use the metric definitions appropriate to the collective and compare with the same topology and message sizes”.

The related but distinct situation is “Two BCM head nodes both attempt to mount a shared writable filesystem during a simulated split-brain incident”. The acceptance board should flag that Split brain can corrupt management and job metadata during apparent redundancy. The deciding requirement here is “Premature takeover can damage cluster state and prevent reliable subsequent recovery”.

Another failure mode is possible in general, but the stated condition most directly creates the exposure that Premature takeover can damage cluster state and prevent reliable subsequent recovery.

The option would apply to “A Base Command Manager cluster relies on one head node and loses both DHCP provisioning and Slurm scheduling whenever that node powers off”. Without correction, An untested single head node can become a cluster-wide control-plane outage. The present scenario is distinguished by “Premature takeover can damage cluster state and prevent reliable subsequent recovery”.

This alternative addresses the separate case “An operator checks spine switches only by ping and does not inspect temperature, transceiver status or port error counters”. This case illustrates that A responsive control-plane address says little about port ASIC or transceiver performance under GPU traffic. The deciding requirement here is “Switch software interoperability and feature parity can affect routing and RDMA behavior under load”.

The option would apply to “A fabric switch reboot clears a temporary configuration and leaves performance counters and congestion policies different from the acceptance baseline”. The operational evidence follows from the mechanism that A working running state is not guaranteed to survive reboot without correct saved startup configuration. This question instead requires “Switch software interoperability and feature parity can affect routing and RDMA behavior under load”.

This is an incorrect prioritization of the mechanism. The reported fault points to Switch software interoperability and feature parity can affect routing and RDMA behavior under load, not the separate condition involving A switch is reachable over SSH but the installed firmware does not support the configured high-speed port breakout profile.