Novel Mathematics. Real Quantum Hardware. Results That Speak for Themselves.
Novel mathematical algorithms and frameworks that outperform
30-year industry standards on real quantum hardware — and deliver
measurable advantages on classical systems. All results independently verifiable.
Novel Mathematics. Multiple Breakthroughs.
Phoenix Quantum Labs develops novel mathematical algorithms and frameworks that produce verified results on both quantum and classical hardware. The same core mathematics drives our quantum search algorithm, our state observation methods, and our classical computation tools.
We discovered mathematical structures that unlock computational advantages on any hardware they're applied to — from consumer GPUs to quantum processors. Not simulations. Not incremental improvements. Fundamentally new mathematics.
Phoenix Attention — Long-Context Inference Kernel
Drop-in attention replacement for frontier LLMs. 26× faster than FlashAttention at 1M context on H100 (Nemotron per-GPU config); 154× at 1M context on L40S (Qwen 7B). Validated end-to-end on Meta Llama 3.3 70B (9 of 80 attention layers bit-exact, 2.57× e2e at 4K context) and NVIDIA Nemotron-3-Super-120B (1 of 8 attention layers bit-exact). 100% token-identical to FlashAttention reference in the substitutable region. No retraining required.
VIEW BENCHMARKS →Phoenix SVD — Weight Decomposition
CUDA SVD for LoRA / PiSSA initialization, weight decomposition, model compression. On real production model weight matrices at their real sizes: 79.9× on Nemotron-3-Super-120B Mamba in_proj 18,560² (10 min 46 sec → 8 sec), 41.9× on Llama 3.3 70B up_proj 16,384² (4 min 6 sec → under 6 sec), 67.7× on Llama 3.3 70B q_proj 8,192², 48.6× on Nemotron q_proj 4,096². Cross-domain proof on 10x Genomics single-cell biology data: 166.9× over NumPy CPU. fp64 singular values match torch to ~10−9–10−11. Drop-in torch.linalg.svd-compatible API.
Quantum Search
Beat Grover's algorithm — the gold standard since 1996 — on real quantum hardware. 32/32 wins across 3-, 4-, and 5-qubit systems. 98.2% vs 79.9% at 3q. 97.7% vs 52.3% at 4q. 97.3% vs 7.3% at 5q. Gets better as noise increases, not worse.
VIEW RESULTS →Non-Destructive Observation
A novel observation method that extracts quantum state information while preserving 93.9% of the original. Independently confirmed as real measurement. 20 recursive observations with no exponential decay.
VIEW RESULTS →VOIS — O(1) Search
GPU-native similarity search with constant-time retrieval. 2.5–4.2x faster than Meta's FAISS at every recall level. 197,000 queries/second on a consumer RTX 4060. Same math, classical hardware.
VIEW BENCHMARKS →PX Compute — GPU + CPU Memory Pool
Systems-level GPU + CPU memory pool allocator. 2,706× over cudaMalloc on 1 GB block reuse (single H100); 1,500× avg across 8× H100s in parallel (driver-lock contention amplifies the win). 22.6× CPU vs libc malloc on 1 GB reuse. Targets workloads that re-cycle large buffers in tight loops — LLM KV caches, attention scratch, gradient buffers, embedding tables.
Seeking Research Partners & Funding
We are actively seeking SBIR/STTR funding, research partnerships, and strategic investment. U.S. patents filed. Select results available under NDA.
CONTACT US →