Huge pages

Memory address translation

Processors use translation structures to map application addresses to physical memory. Larger pages let a translation cover more memory, which can reduce overhead for a workload that repeatedly accesses a large dataset.

XMRig uses the name huge pages across platforms even where the operating system calls them large or super pages. The correct setup depends on both the miner and the host.

References: XMRig huge-page documentation

Check allocation rather than assuming it

A configuration request and successful allocation are different observations. Inspect the miner output after startup or a configuration change. A process can continue mining even when the desired allocation was unavailable.

The RandomX dataset is shared read-only data within an implementation. It is distinct from the miner threads and their working memory. Avoid estimating total memory needs by multiplying the full dataset by the thread count without checking the miner architecture.

References: XMRig huge-page documentation · RandomX algorithm specification

Example

XMRig reports huge-page allocation at startup. If allocation is incomplete, an operator checks memory availability and operating-system permissions before comparing hashrate with an earlier run.

References: XMRig huge-page documentation · RandomX algorithm specification

Limits and cautions

Huge pages do not guarantee a fixed speedup. Support, privileges and page sizes vary by operating system. Reserving memory can reduce what other applications can use, so a mining-machine setting is not a universal server recommendation.

References: XMRig huge-page documentation · RandomX algorithm specification