Two-Socket NVMe RAID with Intel VROC: VMD Setup, Arrays, Spares and Rebuilds
These archived examples show Intel VROC with VMD Technology 6.2.0.1034 through the BIOS UEFI interface. Socket numbers, PCIe stacks, drive ports, upgrade keys and BIOS menus are platform-specific. Confirm the installed server’s lane/connector mapping and supported firmware/driver/OS before changing VMD or boot configuration. A BIOS restart or storage-controller change can interrupt access to an existing OS.
Creation, deletion, spare assignment and rebuilding are separate operations. Verify the intended available drives and protect required data first. A Normal controller state is not proof that initialization, OS boot, application access or backups have been tested. The original screenshots are already in English and retain their actual selections.
1. Preparation: enable the appropriate VMD path
The source first sets Advanced → CSM Configuration → Boot Mode to UEFI Mode and sets Network, Storage, Video OPROM Policy and Other PCI devices option-ROM policies to UEFI. CSM Support itself remains Enabled in the screenshot. This is the source configuration, not an instruction to change a working installation’s boot layout blindly:
Open Processor → IIO Configuration → Intel VMD Technology. For the source NF5280M5 wiring, the NVMe connection is associated with CPU1 and the source selects the Socket1 VMD configuration. Preserve the platform’s own numbering rather than assuming a universal CPU-to-socket offset:
In that Socket1 example, PStack0 and PStack1 are Disabled and PStack2 is Enabled; other settings are left as shown. Save with F10 and reboot after confirming the actual NVMe connector/stack map:
The original article says VROC can only be configured in BIOS. This guide uses the pre-OS workflow, but that is not a general restriction: Intel also documents OS management tools for supported releases. Do not substitute instructions from a different VROC generation without checking compatibility.
2. RAID configuration and management
2.1 View the arrays
Under Advanced, open Intel Virtual RAID on CPU. Select the intended volume in Intel VROC Managed Volumes to inspect details and members. The list example shows Volume0 RAID0 10619.16GB and Volume1 RAID0 530.94GB, both Normal. Volume0 details show 128KB strips, 512-byte blocks and Bootable Yes:
2.2 Delete the verified volume
Select the intended volume and choose Delete. This deletion case shows Volume0 RAID0 11178.05GB, a different capacity from the preceding view example. The confirmation warns that all volume data will be lost and members reset to non-RAID. Choose Yes only when the target and data disposition are correct:
2.3 Create RAID0
Open Intel Virtual RAID on CPU → All Intel VMD Controllers. Under Non-RAID Physical Disks, inspect the available drives and choose Create RAID Volume. The source lists three 3726.02GB Intel SSDPE2KX040T8 drives on CPU1/VMD2, ports 3:1, 3:2, 3:3:
Review Name, RAID Level, Enable RAID spanned over VMD Controllers and Select Disks. The source uses Volume0 and RAID0, leaving spanning unchecked. It initially marks the first two drives with X; later settings include all three. Use the keys supported by the actual BIOS; this interface shows X for selected drives:
Review strip size and capacity before Create Volume. The final source form has all three members selected, 128KB strips and 10619.16GB, with a displayed maximum 11178.05GB. The default is below the maximum. The source says unused space can support another volume; Intel also documents reserved capacity for disk coercion in supported releases, so do not force the maximum without checking that release’s requirements. Choose Create Volume only after checking the complete member set:
2.4 Create RAID1
The source repeats Advanced → Intel Virtual RAID on CPU → All Intel VMD Controllers → Create RAID Volume. Set the intended name and choose RAID1(Mirror). This separate case selects the first two drives, leaving spanning unchecked and the third unselected:
Review the capacity. The source shows 3539.71GB and a maximum 3726.01GB; these are configuration values, not guaranteed OS-visible capacity. Choose Create Volume after verifying the members, licensing and any required confirmations. The later spare example shows this RAID1 volume Normal:
2.5 Create RAID5
Use the same navigation, select RAID5(Parity) and mark the three intended available drives. The source keeps spanning unchecked because these members are on one VMD controller:
The settings show 64KB strips, 7079.43GB, maximum 7452.03GB and RWH Policy Disable. Keep that recorded policy distinct from a claim that RAID write-hole protection is enabled. Select Create Volume only after the appropriate policy, licensing and data-loss checks. These images show the form/command; they do not include a separate completed RAID5 creation confirmation:
2.6 Assign a hot spare
This branch returns to the separate RAID1 Volume0, 3539.71GB, Normal, with Upgrade key Premium. Under All Intel VMD Controllers, select the available third 3726.02GB drive on port 3:3, slot 54; it is outside the two-member mirror:
Choose Mark as Spare. Its confirmation explicitly warns that marking the disk as a spare removes all data on that disk. Answer Yes only for the verified available drive. The following detail screen shows Status Spare; spare assignment is not a rebuild-completion test:
3. Common cases
3.1 Missing RAID options or license-dependent levels
The source says RAID0 works without a key in its example and that RAID1/RAID10/RAID5 require a RAID_KEY module. Treat that as its observed configuration, not universal licensing guidance. Intel’s official documents distinguish Pass-Thru, Standard, Premium and Intel-SSD-only configurations; supported levels depend on the release, platform, SSDs and installed key. Even RAID0 availability without a key is not universal. Verify the server vendor’s supported entitlement before ordering or fitting a module:
3.2 A redundant-array member goes offline
The source’s RAID5 Volume0, 7079.43GB case changes to Degraded after a member fault. It says a supported replacement can start automatic rebuilding; the next captured state is Rebuilding, not completed. Follow the actual platform’s supported replacement/power sequence:
If the matching service procedure requires a manual rebuild, select the degraded volume and choose Rebuild. Select the verified destination drive. The source identifies port 3:3/slot 54, 3726.02GB, and warns that selecting it initiates rebuilding, which completes in the OS; all data on the destination disk will be lost. Do not select a surviving member or an unrelated disk:
The next view is Rebuilding. The last view is Normal after the source’s rebuild example. Verify the actual completed rebuild and data access rather than treating an earlier Rebuilding state as completion:
3.3 Multiple offline members
The final screenshot shows the RAID5 volume Failed and unavailable. If required data is not recoverable from a verified backup, preserve the member identities and failure history and obtain qualified recovery support before further writes. The source’s multiple-member RAID5 failure is not a rule that every two-drive fault always destroys every RAID10 layout; survivability depends on RAID topology and which members failed:
