What FP8 is, why Pearl is testing it, what the code says today, and what it could mean for your graphics card.
Mining Pearl means doing the same kind of math that runs AI: multiplying huge tables of numbers.
Today Pearl uses whole numbers, like 7 × 9 = 63. FP8 uses small decimal numbers. "FP" means floating point (a number with a decimal point) and "8" means each number takes 8 bits, a quarter of the usual 32. Smaller numbers let a graphics card crunch far more of them at once, and they are what modern AI actually uses.
When a miner finds a block, every computer on the network checks the work and must get exactly the same answer.
"Close enough" is not allowed: a cheater could nudge an answer slightly, again and again, to get free lottery tickets. Pearl's plan is to check the work against the exact rounding of specific chips [1, 2].
We checked Pearl's GitHub on 23 October 2026:
fp8-scheme, started 9 September. It is switched off on the main network and only on for local test networks [4].Keep in mind: Pearl's own miner has only ever supported data-center H100/H200 cards [3]. Gaming cards mine Pearl today because independent developers (BzMiner, tnn-miner, redPeaRL and others) wrote software for them.
None of these has happened yet. Until they do, there is no reason to panic, sell or buy hardware.
From the redPeaRL team: we make redPeaRL, a Pearl miner for AMD RX 9070 XT and RX 9060 XT cards. These cards have FP8 hardware, but we have not tested yet whether they can match Pearl's checks. We will post here when we know. For a lighter take on the same news, read our satire GPU Mining, 2009–2026: An Obituary. Questions: GitHub Discussions or Telegram @redpearlsupport.
Educational overview based on public specifications and open-source code as of 23 October 2026. Drafts and test branches can change. Not financial advice.
node/chaincfg/params.go on the fp8-scheme branch (Fp8ForkHeight not set for mainnet; 1 on regtest and simnet). github.com/pearl-research-labs/pearlzk-pow/src/v4/api/public_params.rs on the fp8-scheme branch (devices H100 and B200). github.com/pearl-research-labs/pearlminer/miner-base/src/miner_base/devices.py on the fp8-scheme branch (RTX 50 / SM120 mapped to the B200 device). github.com/pearl-research-labs/pearl