maxnflaxl/MXBM
Watch
1
0
Fork
You've already forked MXBM
 0
Code Issues Pull requests Projects Releases 1 Packages Wiki Activity Actions
1 release 5 tags
  • v0.8.452 f9cd08ee01

    MXBM v0.8.452
    Some checks failed
    release / linux-x86_64 (push) Has been cancelled
    Details
    release / windows-x86_64 (push) Has been cancelled
    Details
    ci / linux (push) Has been cancelled
    Details
    ci / windows (push) Has been cancelled
    Details
    Stable

    maxnflaxl released this 2026-09-24 21:12:02 +00:00 | 11 commits to master since this release

    Signed by maxnflaxl
    GPG key ID: 18CEADAC856CDB53

    RTX 50-series cards were running a scheduling trick that helps on 40-series and hurts on Blackwell. It's now off there: the RTX 5080 went from 60.8 to 77.9 sol/s, ahead of lolMiner's 74.2 sol/s. Nothing changes on other cards.

    Thanks to Randel for the report and the testing.

    https://git.maxnflaxl.dev/maxnflaxl/MXBM

    Speed

    One RTX 5080 on Windows with the owner's overclock, 120 s --report:

    miner sol/s ms/solve W
    MXBM v0.8.452 77.9 25.7 277.5
    MXBM v0.8.443 60.8 32.4 234
    lolMiner 1.98a 74.2 251.3

    v0.8.443 never reached the card's power limit and read 60.8 sol/s at every cap from 250 to 400 W. At a 250 W cap v0.8.452 matches lolMiner's speed and is 6.4% more efficient. The two miners count sol/s differently, so treat the gap as indicative. The v0.8.452 row was measured on v0.8.451, which has the same code. Full curve: performance.md.

    Changes since v0.8.443

    • CUDA: speculative entry is off on compute capability 12.x (RTX 50 series). It cost the RTX 5080 27.6%. On a 4070 Ti SUPER it's still a 3% win, so it stays on everywhere else.
    • An idle RTX 50-series card no longer reads as memory-locked at startup. GDDR7 idles at 810 MHz, above the old fixed 600 MHz floor. The floor is now a fifth of the card's maximum memory clock.
    • --report prints a device block: compute capability, native cubin or JIT, SM resources, and the occupancy each kernel reached.

    Downloads

    • mxbm-0.8.452-linux-x86_64.tar.gz: Linux x86-64, glibc 2.38 or newer
    • mxbm-0.8.452-windows-x86_64.zip: Windows x86-64

    CUDA needs compute capability 7.5 or newer (RTX 20 / GTX 16 and up). Older NVIDIA cards and AMD run OpenCL. The minimum is 3 GB of VRAM; 8 GB gets the fastest layout.

    Verifying

    gpg --verify SHA256SUMS.asc SHA256SUMS
sha256sum -c --ignore-missing SHA256SUMS    # Linux
certutil -hashfile mxbm-0.8.452-windows-x86_64.zip SHA256    # Windows: compare with SHA256SUMS

    Each archive also has its own .asc. Signing key: A6EF 4605 FCE6 FD23 0262 EAA2 18CE ADAC 856C DB53 (Maxnflaxl me@maxnflaxl.dev).

    Dev fee

    1%: one 36-second round per hour of mining. lolMiner charges the same on BeamHash III.

    Fee rounds get their own row in the stats and never mix with your shares. --dev-fee raises it.

    What would help

    Numbers from cards not listed here. Run

    sudo mxbm --report

    It benchmarks your card, measures the power curve and prints a markdown block. Root is only for the power curve; without it you still get the benchmark. On Windows, use an Administrator terminal. Nothing is uploaded. Send me the block on any of my socials.

    Downloads