Interactive Benchmark

Python Implementations

From naive loops to JIT compilation and AADC acceleration

Performance Comparison

Trades
101001K
Scenarios
10K100K500K
Greeks Tax

Why AAD Scales Better

Bump-and-revalue cost grows linearly; AAD stays constant

Greeks Scaling: NumPy vs AADC

AAD computes ALL Greeks in constant time regardless of count

Greeks Bump Evaluations AAD Evaluations AAD Advantage
3 4 ~1.4 3x
10 11 ~1.4 8x
50 51 ~1.4 36x
100 101 ~1.4 72x

Bump-and-revalue requires N+1 evaluations for N Greeks, while AAD computes all Greeks in ~1.3 evaluations (1 forward + ~0.3 adjoint overhead).

Bump-and-Revalue (NumPy)

Requires N+1 evaluations for N Greeks. At 50 Greeks, NumPy needs 51 model evaluations.

AAD (AADC)

Requires ~1.4 evaluations regardless of Greek count. One forward pass + one adjoint sweep.

Integration Effort

Code Changes Required

AADC requires minimal changes to your existing Python code

Basic → AADC
+176 lines added
781 → 957 total lines

Add AADC imports, wrap inputs as idouble, and call kernel recording APIs. Your model logic stays unchanged.

What Changes
  • Import aadc module
  • Wrap inputs: spot = idouble(spot)
  • Record kernel: kernel = aadc.record(...)
  • Request Greeks: kernel.greeks(['delta', 'rho', 'vega'])
  • No changes to pricing logic
  • No manual differentiation
Learn More

Related Resources

How AADC Works

AADC extracts a computational graph (DAG) from existing analytics at runtime and compiles it to optimized machine code, achieving 100-1000× for Python and 6-100× for C++ with automatic derivatives.

Learn More

AADC for Python

Revolutionary Python accelerator delivering 1000x performance improvements for quantitative modeling and Monte Carlo simulations with automatic differentiation. Proven 10x+ faster than JAX, PyTorch, and TensorFlow for quantitative finance workloads.

Learn More

All Benchmarks

Independent benchmarks demonstrating AADC's performance advantages over traditional AAD approaches, GPU implementations, and ML frameworks.

Learn More
Test Environment

Benchmark Environment

All benchmarks executed on enterprise-grade server hardware

System Configuration

CPU2x Intel Xeon Platinum 8280L @ 2.70GHz
Cores56 physical (28 per socket), 112 threads
Architecturex86_64, Cascade Lake
L3 Cache77 MiB (38.5 MiB per socket)
RAM283 GB DDR4
OSLinux kernel 6.1.0-13-amd64 (Debian)
CPU Features: AVX-512, AVX2, FMA, AES-NI

Test Configuration

ModelAsian Option Monte Carlo
DynamicsGeometric Brownian Motion (GBM)
Timesteps252 (daily over 1 year)
GreeksDelta, Rho, Vega (3 sensitivities)
Threads8 (configurable)
SIMDAVX2 (4 doubles/instruction)
Note: AVX-512 (8 doubles/instruction) provides ~1.7x additional speedup on supported hardware

Compilers & Versions

GCC12.2.0 (Debian)
Clang14.0.6 (Debian)
Python3.11.2
NumPy1.26.x
AADC2.0.0
C++ compiled with -O3 -march=native -std=c++17