AAYU Silicon Performance & AI Benchmarks
Real measured results across the 3 core pillars: flat-stack zero-copy function calls, native silicon Tensor/GEMM neural operators, and Liquid Data 2.0 molecular compression.
7.2x Recursive Speedup
Replaced 2KB CallFrame cloning with a preallocated flat-stack base pointer. Eliminated 28 GB of heap allocations in recursive workloads, dropping fib(32) to sub-second speed.
Native Cache-Blocked GEMM
32x32 cache-tiled matrix multiplication, batched dot-product attention, Taylor-expanded GELU/SiLU activations, and Vector RAG cosine similarity.
91.1% - 95.2% Byte Compaction
Git CAS interning + Columnar FoR deltas + 7-layer cryptographic police inspectors with O(1) random cell access and 100% byte-flip quarantine.
Empirical Execution Speed & Architectural Ground Truth
| Workload | Native C / Rust (-O) | AAYU (Silicon VM) | Node.js (V8 JIT) | Python 3.12 | Architectural Ground Truth |
|---|---|---|---|---|---|
| Cold Boot Time | ~2 ms | 13.9 ms | ~35 ms | ~50 ms | Instant startup for CLI tools and serverless containers |
| Recursion fib(32) | 34 ms | 383 ms (bytecode) | 173 ms | 780 ms | 2x faster than Python; frame zeroing optimization avoids 11.1B writes |
| 50M Hot Loop | 94 ms | 2,432 ms | 329 ms | 11,442 ms | 4.7x faster than Python; ~25x behind native C (standard for bytecode VMs) |
| 5M Function Calls | 34 ms | 666 ms | 183 ms | 1,908 ms | 2.9x faster than Python; flat stack dispatch without heap allocs |
thread::spawn) with Connection: close (~1,500 - 2,500 req/s in real networks; 36,000 req/s in loopback mock). Epoll + keep-alive pooling is scheduled for the next release. Windows PE binary (aayu.exe) is 2.27 MB with embedded SQLite & crypto; minimal stripped Linux ELF is ~890 KB.jit.rs via can_compile) selectively compiles verified pure integer loops and recursion directly into native instructions, keeping dynamic code safely in the VM.Native Silicon Tensor & AI Engine Benchmarks
| Workload / Operation | Dimension / Inputs | Expected Result | Status | Verification |
|---|---|---|---|---|
| Cache-Blocked GEMM (MatMul) | [2, 2] x [2, 2] | [[19, 22], [43, 50]] | VERIFIED | 32x32 Cache Tiled Kernel |
| Transformer Attention(Q, K, V) | Seq=2, D=4 | [[0.58, 0.25, 0.15, 0.68], ...] | VERIFIED | Scaled Dot-Product Forward |
| GELU Activation | [-2.0, -1.0, 0.0, 1.0, 2.0, 3.0] | gelu(1.0) = 0.8412 | VERIFIED | Taylor Approximation Constants |
| Vector RAG Cosine Similarity | 4D Embedding Vectors | sim1 = 0.9993 | VERIFIED | O(1) L2-Norm Normalized |
| Adam Optimizer Step | Weights, Grads, M, V, t=1 | Bias-Corrected Gradients | VERIFIED | std/ai.aayu Pure Action |
AAYU Core Workload Matrix (20 Programs)
Twenty distinct benchmark programs from the currently supported AAYU core subset, all run at N=10,000. Every case checks its output.
| Workload | Input size | Expected output | Median ms | Min ms | Max ms |
|---|---|---|---|---|---|
| 01 · Integer addition accumulator | 10,000 | 49995000 | 73.546 | 70.314 | 199.249 |
| 02 · Integer subtraction countdown | 10,000 | 0 | 74.215 | 70.876 | 155.731 |
| 03 · Integer multiply-accumulate | 10,000 | 149985000 | 76.679 | 74.669 | 91.060 |
| 04 · Integer divide-accumulate | 10,000 | 24995000 | 112.868 | 87.982 | 163.001 |
| 05 · Float half-step accumulation | 10,000 | 5000 | 75.452 | 70.692 | 77.032 |
| 06 · Float multiply-accumulate | 10,000 | 24997500 | 73.279 | 70.933 | 77.946 |
| 07 · Less-than true branch | 10,000 | 5000 | 72.851 | 71.212 | 77.565 |
| 08 · Greater-than true branch | 10,000 | 4999 | 72.784 | 72.227 | 77.527 |
| 09 · Equality branch | 10,000 | 1 | 73.867 | 70.723 | 78.430 |
| 10 · Inequality branch | 10,000 | 9999 | 74.745 | 71.805 | 78.093 |
| 11 · Less-or-equal branch | 10,000 | 5001 | 72.496 | 71.005 | 76.236 |
| 12 · Greater-or-equal branch | 10,000 | 5000 | 73.431 | 70.292 | 77.896 |
| 13 · Nested loop counter | 10,000 | 10000 | 76.886 | 70.722 | 80.723 |
| 14 · Nested loop weighted sum | 10,000 | 495000 | 73.169 | 69.788 | 81.360 |
| 15 · Mixed multiply and subtract | 10,000 | 49995000 | 73.878 | 72.859 | 99.137 |
| 16 · Multiply then divide | 10,000 | 249975000 | 74.592 | 70.753 | 82.061 |
| 17 · Descending sum | 10,000 | 50005000 | 75.985 | 72.070 | 86.114 |
| 18 · Variable copy and accumulate | 10,000 | 49995000 | 73.450 | 69.887 | 99.145 |
| 19 · Nested conditional branches | 10,000 | 19999 | 74.717 | 72.385 | 78.386 |
| 20 · Comparison loop boundary | 10,000 | 50005000 | 71.639 | 70.375 | 77.727 |
Reproduce with .\aayu.exe run tests\test_tensor_ai_math.aayu and .\aayu.exe run tests\test_compiler_security.aayu.