This one is for people who rack, bring up and validate GPU systems. Nearly two thirds of the marks are bring-up and cluster verification. You cannot pass it on reading alone, and the weight on testing tells you what NVIDIA thinks goes wrong.
Take it if
You deploy and validate GPU clusters by hand.
Skip it if
You never touch hardware or firmware. Choose NCP-AIO if you operate rather than build.
Before you start
Check yourself against this list first. If something is missing, close that gap before you pay for an exam.
Two to three years with GPU serversNVIDIA recommends this.
Data center networking and powerAlso recommended.
Hands-on hardware accessYou cannot learn bring-up from reading.
2 to 3 years deploying and validating GPU servers and AI clusters
If you fail
Wait 14 days
What the exam covers
Cluster Test and Verification33%
System and Server Bring-Up31%
Control Plane Installation and Configuration19%
Troubleshooting and Optimization12%
Physical Layer Management5%
What to focus on
Cluster testing is a third of the exam: burn-in, performance benchmarks and fabric checks.
Server bring-up: BMC, firmware, drivers, and GPU validation.
BlueField DPU setup and network configuration.
Control plane installation and cluster management.
Finding faults in hardware, cabling and transceivers.
Everything on one mapEvery topic and key idea in this certification.
mindmap
root((NCP-AII))
Test and verify
DCGM Diagnostics
Load testing
Isolate
Server bring-up
BMC
Firmware and drivers
Missing GPUs
Control plane
Base Command Manager
BlueField DPU
Fabric Manager
Troubleshoot and physical layer
Physical first
Thermals
Records
From rack to handoverThe order of work, and where faults send you back.
stateDiagram-v2
direction LR
[*] --> Racked
Racked --> Firmware: BMC reachable
Firmware --> Drivers: Versions match
Drivers --> NodeTest: GPUs visible
NodeTest --> FabricTest: Node passes
FabricTest --> LoadTest: Links clean
LoadTest --> Handover: Stable at full load
NodeTest --> Firmware: Fault found
FabricTest --> Racked: Cable or optic fault
Lesson 1 of 4, 33% of the exam
Test and verify
A third of the exam.
DCGM Diagnostics
dcgmi diag runs levels 1 to 4 to assess readiness.
Load testing
Some faults only appear at full load and heat.
Isolate
Node passes but pair fails? Look between them.
DCGM diagnostic levelsHigher levels run longer and go deeper.
flowchart TD
A["nvidia-smi shows fewer GPUs"] --> B{"Seen on PCIe?"}
B -->|No| C["Check seating and hardware"]
B -->|Yes| D{"Firmware and driver match?"}
D -->|No| E["Align versions"]
D -->|Yes| F["Check kernel logs"]
Common trap
Reinstalling software for a hardware visibility problem.
Try it yourself
Walk through bring-up from BMC to a passing diagnostic.
flowchart TD
A["Link errors"] --> B["Reseat the cable"]
B --> C{"Fixed?"}
C -->|No| D["Swap the transceiver"]
D --> E{"Fixed?"}
E -->|No| F["Try another port"]
C -->|Yes| G["Done"]
E -->|Yes| G
Exam facts checked on October 31, 2026 against NVIDIA certification page. Lessons are our own explanations, so use the official docs as the final word, and confirm price and dates before booking.