Questions operators ask
What ARAM works with, what it does alone, and where it fits.
What hardware and vendors does ARAM work with?
BMCs over Redfish, including OpenBMC, Dell iDRAC, HPE iLO, Supermicro, NVIDIA and AMD. NVIDIA and AMD GPUs through DCGM and ROCm. CDUs from Vertiv, Schneider Electric and CoolIT over Modbus, or over Redfish for CDUs that follow DMTF’s Redfish profile for CDUs. Switches running SONiC, or any that expose gNMI or SNMP. ARAM is OpenRMC compatible. Each connector is built to the vendor’s published interface and tested against simulators and emulators; a pilot is where it is validated on your hardware.
How does ARAM work, end to end?
Adapters read each component through the interface its vendor ships. ARAM turns every component into a typed resource in one semantic graph of the rack, and its models detect faults and trace each one to its root cause across GPU, cooling, power and network. Runbooks act within your policies and verify the fix, and everything is reported through a Redfish-compliant API.
What failures do ARAM’s models cover?
Purpose-built models detect, predict, root-cause and remediate failures in AI rack components: GPUs, PSUs, DIMMs, fans, NVLink and UALink fabrics, thermals and coolant.
How does ARAM know a fix worked?
After a runbook acts, ARAM checks the component again and only closes the incident after consecutive clean readings. If the readings don’t recover, it escalates to a person.
Can we see what ARAM did, and why?
Yes. Every detection, decision and action is logged with the evidence behind it and the policy that allowed it.
How much does ARAM do on its own?
You decide, policy by policy. Start with recommendations only, move to approving each action, then let the policies you trust run on their own while everything else still waits for a person.
What happens when a vendor changes its interface or firmware?
ARAM’s adapters use OEM profiles that reload without a restart, so a new profile goes in while readings keep flowing. Pulled components such as a PSU or a compute tray are marked removed, not failed, and their replacements are found and checked.
Does ARAM support liquid-cooled racks?
Yes. ARAM reads the CDU over Modbus or Redfish alongside the GPUs it cools, so it can trace GPU throttling back to the coolant loop. It covers coolant loops and leak detection, and can stop a pump or cap power within your policies.
Does ARAM replace our DCIM?
No. DCIM plans and tracks the data hall: space, power, assets and capacity. ARAM runs each rack and reports rack health and every action it takes through a Redfish-compliant API that your DCIM and other tools can read.
How is ARAM deployed?
ARAM ships as one container image. In a pilot, an ARAM engineer plans with you where it runs and which racks it reads.
Ask us about your racks
Racks you choose, vendors you already run, and policies that decide what ARAM may fix on its own.
Request a pilot