BTCXMining correctness
GX9 hash, arithmetic and block-validation tests.
| Test | Recorded result | Coverage |
|---|---|---|
| CPU / GPU agreement | 20,487 exact matches | 7 fixed vectors + 20,480 repeatable test cases |
| Integrated mining | 1,000 blocks · zero GPU/reference mismatches | Miner and node integration · August 29, 2026 |
| Arithmetic checks | 200,000 comparisons · zero mismatches | Exact FP32 and FP64 operations |
| Block verification | Independent node verification | Hash and target checked before acceptance |
Test dates: August 28–29, 2026.
Hardware benchmarks
| Hardware | Measured result | Run conditions |
|---|---|---|
| NVIDIA GeForce RTX 5080 | 93,890 H/s | 60-second run · Windows x64 |
| Intel Core Ultra 9 285K | 12,861 H/s | 60-second run · 24 threads · Windows x64 |
| Intel Core i5-14400F · verification | 4.38 ms at the 99th percentile | 20,000 verifications · one thread · cached header context |
Mining: September 4, 2026. Verification: August 28, 2026. Short runs on individual devices.
ASIC resistance testing
GX9 uses dependent arithmetic operations to limit specialized-hardware advantage. The estimate combines GPU power measurements and circuit synthesis.
| Check | Result | What it establishes |
|---|---|---|
| Estimated ASIC advantage in hashes per joule | 3.002× under the stated assumptions | Includes statistical uncertainty and power-sensor accuracy |
| Physical ASIC comparison | Not performed | Fabricated hardware is needed to verify the estimate |
| Power accuracy allowance | ±5 W | Applied in favor of the specialized implementation |
| Measurement controls | Clocks observed; not locked | No calibrated inline board-power meter was used |
Calculated August 28, 2026.
SHA3 / GX9 construction
| Component | Specification |
|---|---|
| Header binding | Domain-separated SHA3-256 of the header without its nonce |
| Nonce input | 32-byte header binding + 64-bit nonce |
| Initialization | SHA3-512; SHAKE256 expands the master tape |
| Mutable tape | 8,192 bytes · 512 sixteen-byte entries · fresh copy per nonce |
| Logical working state | 9,948 bytes |
| Serial work | 1,536 macrosteps × 8 payload stages = 12,288 stages |
| Arithmetic | MULHI · FFMA32 · DFMA64 · IMMA INT8 |
| Internal mixing | Specified raw-Keccak duplex reinforced every 128 macrosteps |
| Final digest | 32 bytes |
| Persistent dataset | None |
GX9 specification: August 28, 2026. Byte order, arithmetic and padding determine consensus.
Calculation and assumptions
The calculation assigns no overhead cost to the ASIC and uses the more favorable of two cost comparisons.
| Input | Value | Role in the calculation |
|---|---|---|
| MULHI coefficient | 1.537929 | Instruction-family upper coefficient |
| FFMA32 coefficient | 1.539041 | Instruction-family upper coefficient |
| DFMA64 coefficient | 1.055397 | Instruction-family upper coefficient |
| IMMA INT8 coefficient | 1.128208 | Instruction-family upper coefficient |
| Serial-chain coefficient | 1.8975 | 1.65 × 1.15 |
| Payload energy fraction | 0.632007639 | Lower bound after the power-accuracy allowance |
| Conservative denominator | 0.333073855 | Smaller result from the two cost comparisons |
| Upper efficiency estimate | 3.002337126× | Reciprocal of the conservative denominator |
A = 1 / min(Σ(sⱼ / cⱼ), s_payload / 1.8975). sⱼ is each instruction family’s energy share; cⱼ is its upper coefficient.
Power measurement conditions
An instrumented run measuring serial-work energy, separate from the optimized-miner benchmark.
| Measurement | Recorded value |
|---|---|
| Serial lengths exercised | 0, 4, 8 and 12 · consensus setting: 8 |
| Warmup | Two 90-second windows |
| Measurement windows | Four 30-second repeats per setting, in counterbalanced order |
| Accepted load samples | 830 |
| Rate at consensus setting | 43,962.343 H/s |
| Board power at consensus setting | 106.735241 W |
| Energy before idle subtraction | 2,427.883 microjoules per hash |
| Idle-bracket drift | 0.362036% |
August 28, 2026. These power readings do not apply to the faster September miner benchmark.
Masternode finality
Masternodes certify valid chain state. Finality activates separately from mining.
| Mainnet parameter | Planned setting |
|---|---|
| Selected committee | 10 masternodes |
| Certificate threshold | 7 signatures |
| Signature scheme | Ed25519 |
| Bond per masternode | 2,500 BTCX |
| Minimum eligible population | 30 masternodes |
| Epoch length | 576 blocks |
| Masternode compensation | 40% of post-burn transaction fees when activated |
| Additional block subsidy | None |
| Slashing / delegation / governance voting | Not part of the current design |
Safety assumes no more than three faulty or malicious committee members. Valid proof of work and independent node verification remain required. Conflicting valid certificates halt finality.
Mainnet monetary parameters
| Parameter | Planned setting |
|---|---|
| Supply policy | 21 million + Security Tail |
| Block interval | 150 seconds |
| Difficulty adjustment | ASERT · one-hour half-life |
| Block size limit | 4,000,000 weight units |
| Emission ramp | 17,280 blocks |
| Initial full block subsidy | 12.5 BTCX |
| Halving interval | 840,000 blocks |
| Security tail starts | Block 4,200,000 |
| Long-term issuance | Below 0.45% of gross scheduled supply per target year |
| Premine / developer fee / presale | None |
For gross scheduled supply G, annual tail issuance is max(9 × 10¹², floor(3G / 700)) base units over 210,240 target blocks.