Lu(x)iEdge evaluates mathematical expressions with guaranteed identical results on every run, every architecture — CPU, GPU, ARM, x86 — at 2.35 billion operations per joule. Built for systems where reproducibility is not optional.
SHA-256 verified outputs | Cross-platform determinism | Independently validated by TestFort QA Lab
Under 5MB. Runs locally. No cloud required.
† Metrics validated against non-linear function suite. Full engine validation in progress.
NumPy, PyTorch, and TensorFlow optimize for speed. They do not guarantee identical results across runs or platforms. GPU thread scheduling causes floating-point drift that breaks audits, blocks certifications, and makes debugging impossible.
Lu(x)i is the only math engine that guarantees bit-exact, SHA-256 verifiable results on ARM, x86, and GPU, every time.
Lu(x)i completes mathematical workloads faster than standard libraries. Faster completion means less time at peak power.
Extended runtime, reduced thermal throttling
Less heat per workload, lower cooling demand
"Efficiency validated for dense vector expressions (y = f(x)). Results vary by workload and hardware."
Drones, robots, satellites, industrial sensors. 2.35B ops/joule means more compute per charge, less thermal throttling, longer missions.
Learn More →FINRA-compliant • Monte Carlo • Audit trails
Learn More →DO-178C requires deterministic execution. Memory-safe Rust, sub-5MB binary, SHA-256 verified outputs for safety-critical systems.
Learn More →Deterministic computation for safety-critical edge systems
DO-178C certification-ready architecture with SHA-256 verified calculations for targeting systems, sensor fusion, and autonomous navigation.
FINRA-compliant reproducible calculations
Reproducible calculations for trading, risk modeling, and Monte Carlo simulations with complete regulatory audit trail support.
Faster compute means less time at full power
Lu(x)i completes calculations faster, so your hardware spends less time at full power.
Independently verified by TestFort QA Lab during a 1-hour GPU endurance test on NVIDIA H100 SXM, December 2025. Validation scope: non-linear function suite, cross-platform determinism, and GPU endurance. Validation of remaining function categories is in progress.
| Metric | Value | Why It Matters |
|---|---|---|
| Determinism | SHA-256 verified | Bit-exact results for audits & certification |
| Error Rate | 0.00% | Zero failures over 444.4T operations |
| Hash Consistency | 10/10 runs identical | GPU and CPU produce same output |
| Metric | Value | Why It Matters |
|---|---|---|
| Aggregate Throughput | 286.94B ops/sec | Process massive vectors in milliseconds |
| Peak Throughput (sqrt) | 331.13B ops/sec | Function-dependent ceiling |
| P95 API Latency | 1.47ms | Real-time capable under load |
| Metric | Value | Why It Matters |
|---|---|---|
| Energy Efficiency | 2.35B ops/joule | Edge-deployable, battery-friendly |
| Average GPU Power | 117.2W | Below H100 TDP |
| Parameter | Value |
|---|---|
| Hardware | NVIDIA H100 SXM, 80GB HBM3 |
| Duration | 1 hour continuous |
| Concurrent Users | 200 |
| Total Operations | 444.4 trillion |
| Validator | TestFort QA Lab, Dec 2025 |
98bd97026a738671ec7c3d302efa6aa8ff078a5fb9183f7fdf51a1c4ff938321
Identical hash confirms bit-exact determinism across GPU and CPU execution modes.
Independent testing via Phoronix Test Suite on NVIDIA H100 80GB HBM3
System: 2x Intel Xeon Platinum 8468 (96 cores), 1520GB RAM, Ubuntu 22.04, CUDA 12.4
Exceeds NVIDIA spec of 67 TFLOPS
View on OpenBenchmarking.org →
Deviation: 0.59% across samples
View on OpenBenchmarking.org →
2.7x median vs 121 GPUs tested
View on OpenBenchmarking.org →
All benchmarks independently verified via OpenBenchmarking.org | Dec 29-30, 2025
Zero drift across incongruent hardware architectures via the 'Art of Fugue' Polyphonic Stress Test.
| Platform | Architecture | Stress Test | SHA-256 Identity |
|---|---|---|---|
| Apple M1 Pro | ARM64 (Neon) | Polyphonic (40 Threads) | 98bd97...ac19 [MATCH] |
| NVIDIA H100 | CUDA (Ampere) | Polyphonic (40 Threads) | 98bd97...ac19 [MATCH] |
| NVIDIA L4 | Vulkan | Polyphonic (40 Threads) | 98bd97...ac19 [MATCH] |
Same input + same environment = identical output with SHA-256 verification.
GPU mode (CUDA/Vulkan) and CPU mode (Rust f32/f64 with SIMD acceleration: AVX-512, AVX2, ARM Neon).
Demonstrated 286.94B ops/sec aggregate on NVIDIA H100 SXM.
2.35B ops/joule efficiency enables edge and power-constrained deployments.
Patent Pending — Track 1 Non-Provisional Filing, USPTO
Stateless HTTP interface for deterministic expression evaluation. 22 core functions and 6 binary operators combine into 2,900+ expression combinations.
Vectorized expression evaluation with deterministic results.
Request:
POST /evaluate
{
"expr": "exp(-0.5 * ((x - mu) / sigma)^2) / (sigma * sqrt(2 * pi))",
"x": [1.0, 2.0, 3.0],
"mu": 2.0,
"sigma": 0.5
}
Response:
{
"y": [0.1080, 0.7979, 0.1080]
}
expr = mathematical expression | x = scalar or vector input | Deterministic: same request = same response
Transparent methodology, reproducible benchmarks, and citation formats for academic evaluation.
@software{luxiedge2026,
author = {Waller, Eric},
title = {{Lu(x)i}: Deterministic Numeric
Computation Engine},
year = {2026},
version = {0.1.0},
url = {https://luxiedge.com},
note = {Accessed: 2025-12-18}
}
Waller, E. (2026). Lu(x)i: Deterministic Numeric Computation Engine (Version 0.1.0) [Computer software]. https://luxiedge.com
Note: Lu(x)i is commercial software. This citation references the software directly.
Request early access to Lu(x)iEdge. Tell us about your use case and we'll follow up within one business day.