AMD's RMPOPT v16 patch set was queued into the tip/tip.git x86/sev branch on 1 October 2026, ahead of the Linux 7.4 merge window that opens later that month. The patches target EPYC server CPUs running SEV-SNP confidential virtual machines, and the filing carries no commit hash or patch identifier beyond the branch name and version number.

A reduction of the Reverse Map Table, the hardware structure SEV-SNP uses to track which physical pages belong to which encrypted guest. RMPOPT trims that table when the host's system RAM is not saturated with confidential VMs, freeing address-space overhead that would otherwise sit idle in the RMP. The feature is aimed at Zen 6 'Venice' platforms, where the larger memory maps make the RMP footprint more visible to the hypervisor. The patch carries no wattage, node, or SKU measurement, and no RMP page count is printed.

No RMP size delta, no page-count reduction, no latency figure appears anywhere in the filing. Sixteen rounds of revision on a table-size optimization suggests the SEV-SNP security model constrained the patch shape more than the performance target did. For a buyer sizing a Venice server for mixed confidential and non-confidential workloads, the document does not prove how many gigabytes of RMP overhead disappear. That number is the one that decides whether RMPOPT matters at rack scale, and it is absent from every page in the dossier.

What remains unmeasured: the exact commit hash, the stable release date (only 'around late December' is printed), and any before-and-after RMP size. The x86/sev branch on git.kernel.org was inaccessible at filing time due to a proof-of-work challenge, so the commit log could not be verified independently. Watch the 7.4 stable tree in late December for the actual RMP delta and any follow-up tuning patches that might adjust the threshold at which RMPOPT engages relative to total system memory.

Sixteen revision rounds on a table-size patch indicates the SEV-SNP security model, not the performance target, drove the patch shape. The filing proves no RMP size reduction, so a Venice buyer cannot quantify the RAM savings from RMPOPT.

The 16-round revision history on a single RMP-trimming feature points to security-review friction in the SEV-SNP path, not to a difficult performance optimization. Without a printed page-count or byte delta, the architectural savings are unquantified; the buyer must wait for the 7.4 stable tree to measure actual RMP shrinkage on a Venice platform.