RandomX dataset

Fast mode and memory use

RandomX specifies a dataset base size of 2,147,483,648 bytes plus 33,554,368 extra bytes. Fast mode keeps this generated data in memory so the virtual machine can access it while executing its work.

Light mode obtains the required dataset items from the smaller cache instead of keeping the full dataset resident. This reduces memory requirements at a computational cost. The modes evaluate the same proof-of-work function rather than separate mining networks.

References: RandomX algorithm specification · RandomX specification

Seeds and startup pauses

The key used to generate RandomX state is distinct from the block nonce being varied by a miner. Mining software must use the appropriate seed for the target block. Daemon mining interfaces expose seed information alongside other job fields.

Dataset initialization can explain a startup pause without implying that the daemon is resynchronizing its blockchain. Diagnose the miner and node separately, using their own logs and status information.

References: RandomX algorithm specification · Daemon RPC reference

Example

A miner initializes the dataset for the current seed, then reuses it across many hash attempts. Changing a block nonce does not require rebuilding the entire dataset.

References: RandomX algorithm specification · RandomX specification

Limits and cautions

The dataset contains generated mining data rather than blockchain or wallet records. Budget for thread memory, caches and the operating system as well as the dataset itself.

References: RandomX algorithm specification · RandomX specification