Pearl (PRL) plans a hard fork (PIP-3, certificate v4) that switches PearlHash mining from INT7 to
FP8 arithmetic. No mainnet fork height is set yet, but the code is public and a
lready running on testnet, so you can measure today what the fork means for your GPU.
Our open-source miner
bpm now has a built-in test for this:
--test-fp8.
Quickstart (Linux)git clone --recursive https://github.com/pushsoftnow/blackwell_pearl_miner.git
cd blackwell_pearl_miner
cargo build --release
./target/release/bpm --test-fp8
# then mine with the same binary:
./target/release/bpm --pool prl-eu.kryptex.network:7048 --wallet YOUR_KRYPTEX_ID --worker rig1
Quickstart (Windows, PowerShell), after installing the build tools listed in the README:
git clone --recursive https://github.com/pushsoftnow/blackwell_pearl_miner.git
cd blackwell_pearl_miner
cargo build --release
.\target\release\bpm.exe --test-fp8
# then mine with the same binary:
.\target\release\bpm.exe --pool prl-eu.kryptex.network:7048 --wallet YOUR_KRYPTEX_ID --worker rig1
Requirements: NVIDIA Blackwell GPU with compute capability 12.0 (RTX 50-series, RTX PRO Blackwell), CUDA Toolkit 12.8+, Rust. Windows builds are new and not yet tested on real hardware
.
What the test does- Checks the FP8 kernel bit for bit against a reference job from the official Pearl code
- Runs today's algorithm (INT7) and the FP8 algorithm in alternating rounds on your GPU
- Prints a side-by-side comparison: hashrate, efficiency, power, clock, temperature
- Runs offline, needs no pool and no wallet, and takes about 5 minutes
- It's part of the normal miner binary, so you can start mining with the same build right after the test
Example result (RTX PRO 4500 Blackwell, 150 W power limit):
TODAY (INT7, v3) AFTER FORK (FP8, v4) CHANGE
Hashrate 176.5 TH/s 146.1 TH/s -17.2 %
Efficiency 1.177 TH/J 0.974 TH/J -17.2 %
Power draw 150.0 W 150.0 W
SM clock 2196 MHz 1803 MHz -17.9 %
GPU temperature 75 °C 74 °C
On our card, FP8 costs about 17 % hashrate at 150 W. The FP8 data has to be streamed from VRAM, which costs power, so the GPU runs at a lower clock under the same power limit. A higher
limit narrows the gap (-13 % at 200 W). Every GPU has to switch to FP8, so network hashrate and difficulty will move too; a lower TH/s does not mean lower earnings 1:1.
Please note: the results are
preliminary. PIP-3 is not final, and the protocol and pool rules can still change. The FP8 kernel is experimental: it matches the reference b
it for bit, but no pool accepts FP8 shares yet.
bpm is free and open source (ISC), with no dev fee. Every share is checked with the official Pearl verifier before it's submitted.
https://github.com/pushsoftnow/blackwell_pearl_minerSo far we've only tested the RTX PRO 4500.
Please post your result in this thread: GPU, power limit and the RESULT table. Numbers from RTX 5090/5080/5070 and RTX PRO 6000 would
be especially interesting.