“BOM verification and full-load diagnostics pass on the replacement” addresses mechanical electrical and thermal qualification are not interchangeable across models, but the current hardware or infrastructure issue is replacement completion was mistaken for thermal acceptance. The option “fan speeds and device inlet temperatures remain stable under sustained operation” is more directly tied to the required outcome.

“power redundancy and firmware inventory meet approved platform baseline” addresses replacement component qualification includes its firmware and inventory state, but the current hardware or infrastructure issue is replacement completion was mistaken for thermal acceptance. The option “fan speeds and device inlet temperatures remain stable under sustained operation” is more directly tied to the required outcome.

“postrepair port tests demonstrate errors cease on the actual faulted path” addresses root-cause evidence was not established before physical card replacement, but the current hardware or infrastructure issue is replacement completion was mistaken for thermal acceptance. The option “fan speeds and device inlet temperatures remain stable under sustained operation” is more directly tied to the required outcome.

“perform targeted media and adapter isolation before choosing a replaceable component” addresses root-cause evidence was not established before physical card replacement, but the current hardware or infrastructure issue is electrostatic handling controls were omitted during replacement. The option “follow manufacturer ESD isolation and packaging process for GPU swaps” is more directly tied to the required outcome.

Useful for physical component replacement omitted its configured software and networking role, but it does not remove the specific condition: electrostatic handling controls were omitted during replacement.

“check surviving feed health and follow qualified PSU replacement procedure” addresses hot-swap maintenance did not confirm remaining power capacity, but the current hardware or infrastructure issue is electrostatic handling controls were omitted during replacement. The option “follow manufacturer ESD isolation and packaging process for GPU swaps” is more directly tied to the required outcome.

“BlueField host and network-offload tests pass and inventory matches the new card” addresses physical component replacement omitted its configured software and networking role, but the current hardware or infrastructure issue is electrostatic handling controls were omitted during replacement. The option “post-service hardware diagnostics run without new PCIe or device errors” is more directly tied to the required outcome.

“postrepair port tests demonstrate errors cease on the actual faulted path” addresses root-cause evidence was not established before physical card replacement, but the current hardware or infrastructure issue is electrostatic handling controls were omitted during replacement. The option “post-service hardware diagnostics run without new PCIe or device errors” is more directly tied to the required outcome.

“post-replacement nvidia-smi UUID agrees with approved hardware repair ticket” addresses service intervention loses asset traceability needed for future diagnostics, but the current hardware or infrastructure issue is electrostatic handling controls were omitted during replacement. The option “post-service hardware diagnostics run without new PCIe or device errors” is more directly tied to the required outcome.

Useful for replacement component qualification includes its firmware and inventory state, but it does not remove the specific condition: physical component replacement omitted its configured software and networking role.

“follow manufacturer ESD isolation and packaging process for GPU swaps” addresses electrostatic handling controls were omitted during replacement, but the current hardware or infrastructure issue is physical component replacement omitted its configured software and networking role. The option “apply qualified DPU firmware mode and port configuration during reinstallation” is more directly tied to the required outcome.

“perform targeted media and adapter isolation before choosing a replaceable component” addresses root-cause evidence was not established before physical card replacement, but the current hardware or infrastructure issue is physical component replacement omitted its configured software and networking role. The option “apply qualified DPU firmware mode and port configuration during reinstallation” is more directly tied to the required outcome.

“power redundancy and firmware inventory meet approved platform baseline” addresses replacement component qualification includes its firmware and inventory state, but the current hardware or infrastructure issue is physical component replacement omitted its configured software and networking role. The option “BlueField host and network-offload tests pass and inventory matches the new card” is more directly tied to the required outcome.

“BOM verification and full-load diagnostics pass on the replacement” addresses mechanical electrical and thermal qualification are not interchangeable across models, but the current hardware or infrastructure issue is physical component replacement omitted its configured software and networking role. The option “BlueField host and network-offload tests pass and inventory matches the new card” is more directly tied to the required outcome.

“post-service hardware diagnostics run without new PCIe or device errors” addresses electrostatic handling controls were omitted during replacement, but the current hardware or infrastructure issue is physical component replacement omitted its configured software and networking role. The option “BlueField host and network-offload tests pass and inventory matches the new card” is more directly tied to the required outcome.

“profile and tune parallel dataloader processes and read-ahead for the observed access pattern” addresses storage I/O and host page cache behavior constrain accelerator feeding, but the current hardware or infrastructure issue is nUMA memory locality increases transfer latency and CPU-to-GPU traffic. The option “bind data-loader processes and memory to the GPU-local NUMA domain” is more directly tied to the required outcome.

“choose qualified PCIe and NUMA-local NIC-to-GPU affinity” addresses nIC and GPU placement creates avoidable cross-socket transfers, but the current hardware or infrastructure issue is nUMA memory locality increases transfer latency and CPU-to-GPU traffic. The option “bind data-loader processes and memory to the GPU-local NUMA domain” is more directly tied to the required outcome.

“select and qualify CPU C-state policy with facility power constraints” addresses processor idle-state configuration conflicts with workload latency objectives, but the current hardware or infrastructure issue is nUMA memory locality increases transfer latency and CPU-to-GPU traffic. The option “bind data-loader processes and memory to the GPU-local NUMA domain” is more directly tied to the required outcome.

“RDMA and collective throughput recover with lower cross-socket traffic” addresses nIC and GPU placement creates avoidable cross-socket transfers, but the current hardware or infrastructure issue is nUMA memory locality increases transfer latency and CPU-to-GPU traffic. The option “profiler shows lower remote memory access and improved batch delivery” is more directly tied to the required outcome.

“context switches decline and end-to-end batch rate rises” addresses host software concurrency exceeds useful CPU processing capacity, but the current hardware or infrastructure issue is nUMA memory locality increases transfer latency and CPU-to-GPU traffic. The option “profiler shows lower remote memory access and improved batch delivery” is more directly tied to the required outcome.

“packet receive rate improves and no single core saturates at peak traffic” addresses interrupt handling placement bottlenecks ingress traffic, but the current hardware or infrastructure issue is nUMA memory locality increases transfer latency and CPU-to-GPU traffic. The option “profiler shows lower remote memory access and improved batch delivery” is more directly tied to the required outcome.

“tune thread count and affinity based on profiled CPU topology” addresses host software concurrency exceeds useful CPU processing capacity, but the current hardware or infrastructure issue is interrupt handling placement bottlenecks ingress traffic. The option “distribute or pin NIC interrupt queues according to measured CPU locality” is more directly tied to the required outcome.

“apply the supported high-performance CPU power profile within thermal constraints” addresses host power-management policy is starving the application input pipeline, but the current hardware or infrastructure issue is interrupt handling placement bottlenecks ingress traffic. The option “distribute or pin NIC interrupt queues according to measured CPU locality” is more directly tied to the required outcome.

“select and qualify CPU C-state policy with facility power constraints” addresses processor idle-state configuration conflicts with workload latency objectives, but the current hardware or infrastructure issue is interrupt handling placement bottlenecks ingress traffic. The option “distribute or pin NIC interrupt queues according to measured CPU locality” is more directly tied to the required outcome.

“RDMA and collective throughput recover with lower cross-socket traffic” addresses nIC and GPU placement creates avoidable cross-socket transfers, but the current hardware or infrastructure issue is interrupt handling placement bottlenecks ingress traffic. The option “packet receive rate improves and no single core saturates at peak traffic” is more directly tied to the required outcome.

“context switches decline and end-to-end batch rate rises” addresses host software concurrency exceeds useful CPU processing capacity, but the current hardware or infrastructure issue is interrupt handling placement bottlenecks ingress traffic. The option “packet receive rate improves and no single core saturates at peak traffic” is more directly tied to the required outcome.

“input wait time declines and accepted training throughput rises” addresses storage I/O and host page cache behavior constrain accelerator feeding, but the current hardware or infrastructure issue is interrupt handling placement bottlenecks ingress traffic. The option “packet receive rate improves and no single core saturates at peak traffic” is more directly tied to the required outcome.

“select and qualify CPU C-state policy with facility power constraints” addresses processor idle-state configuration conflicts with workload latency objectives, but the current hardware or infrastructure issue is host power-management policy is starving the application input pipeline. The option “apply the supported high-performance CPU power profile within thermal constraints” is more directly tied to the required outcome.

“profile and tune parallel dataloader processes and read-ahead for the observed access pattern” addresses storage I/O and host page cache behavior constrain accelerator feeding, but the current hardware or infrastructure issue is host power-management policy is starving the application input pipeline. The option “apply the supported high-performance CPU power profile within thermal constraints” is more directly tied to the required outcome.

“bind data-loader processes and memory to the GPU-local NUMA domain” addresses nUMA memory locality increases transfer latency and CPU-to-GPU traffic, but the current hardware or infrastructure issue is host power-management policy is starving the application input pipeline. The option “apply the supported high-performance CPU power profile within thermal constraints” is more directly tied to the required outcome.

“RDMA and collective throughput recover with lower cross-socket traffic” addresses nIC and GPU placement creates avoidable cross-socket transfers, but the current hardware or infrastructure issue is host power-management policy is starving the application input pipeline. The option “steady GPU occupancy increases as preprocessing throughput recovers” is more directly tied to the required outcome.

“input wait time declines and accepted training throughput rises” addresses storage I/O and host page cache behavior constrain accelerator feeding, but the current hardware or infrastructure issue is host power-management policy is starving the application input pipeline. The option “steady GPU occupancy increases as preprocessing throughput recovers” is more directly tied to the required outcome.

“chosen policy meets throughput and p99 latency targets reproducibly” addresses cPU scheduling configuration was assumed to be universally beneficial, but the current hardware or infrastructure issue is host power-management policy is starving the application input pipeline. The option “steady GPU occupancy increases as preprocessing throughput recovers” is more directly tied to the required outcome.