SMI300XE MI300X Expert (SMI300XE) Certification Exam Questions and Answers

When a GPU is not detected in theIPMI GUI reports, the recommended recovery method is topower off the system, reseat the GPU and heatsink, then power it back on and check the IPMI GUI for detection. This process ensures that the GPU is properly connected and making full contact with the system board.

Option A (pressing a GPU button for 5 seconds)is incorrect because there is no dedicated reinitialization button for GPU detection in Supermicro MI300X systems.

Option B (removing the GPU tray while powered on)is not advisable as it could cause damage to components.

Option D (reseating the motherboard tray)is unnecessary, as the issue is likely limited to GPU connectivity rather than a full system board issue.

Option C (powering off the system, reseating the GPU, and verifying in IPMI GUI)follows proper troubleshooting guidelines as documented in Supermicro’stroubleshooting and system health monitoring sections​.

The ROCm softwarerequires at least one GPU to function. It can stilloperate with up to 2 missing GPUs, butif 3 or more GPUs are missing, ROCm commands will fail.

Option A (Incorrect):ROCmfails if 3 GPUs are missing.

Option B (Incorrect):The software can function withmore than 1 GPU missing.

Option C (Correct):ROCmcan still execute commands if only 1 or 2 GPUs are missing.

Option D (Incorrect):At least 1 GPUmust be detected for ROCm to work.

TheAGFHC tool for Ubuntumust be installed in the/opt/AMD/AGFHCdirectory.

Option A (Incorrect):The/tmp/directory istemporary storageand is not used forapplication installations.

Option B (Incorrect):/windows/user/desktopis a Windows path and does not apply to Linux-based Ubuntu.

Option C (Incorrect):/usr/home/customer/AGFHCis not the correct standard directory for system-wide software.

Option D (Correct):/opt/AMD/AGFHCis the correct directory for installing AMD-related tools in Ubuntu.

Whenremoving the GPU trayin anMI300X system, the following precautions must be taken:

Two people should remove the trayas itweighs over 70 pounds, making it unsafe for one person to handle.

Keeping the tray straightprevents it fromtilting, which candamage the chassis components.

Unnecessary movement can damage cooling fansdirectly beneath the GPU tray.

Since all the above precautions arecritical for safe removal, thecorrect answer is D (All of the above).

IfROCm commands fail to provide output, the primary issue is likelymissing or undetected GPUs. Thebest way to validate this is by checking the IPMI GUI and reviewing the GPU component sectionto ensure that all GPUs are present.

Option A (Correct):Missing GPUsare the most common causeof ROCm failures. The IPMI GUI provides visibility into which GPUs are detected.

Option B (Incorrect):CPU misalignment would causegeneral system failure, not ROCm-specific issues.

Option C (Incorrect):The GPU baseboard status can be checked in IPMI but isnot the primary causeof ROCm failures.

Option D (Incorrect):ROCm can function with multiple missing GPUs, butthe key issue is confirming GPU presence first.

When working withSupermicro MI300X GPU systems, it ishighly recommendedtomap or take a pictureof cable connections before disconnection.

Reasons:

Ensuring Proper Reconnection→ NVMe backplanes have multiplecritical connections(MCIO cables, power cables, data lanes, and SAS connections). Misplacing any of these cables could cause:

Boot failures

Drive detection issues

Performance degradation

Avoiding Errors in Cable Routing→ Supermicro documentation states thatimproperly routed cables can obstruct airflow, potentially leading tooverheatingin high-performance systems.

Easing Future Troubleshooting→ If issues arise after reassembly, aphoto or cable mapallows engineers toquickly verifywhether cables werereconnected properly.

This best practice isofficially recommendedinSupermicro’s service and installation guidelinesfor NVMe backplanes.