System Capabilities
Optimization Settings
Based on benchmark data, we've implemented specific optimizations for operations where WebGPU was slower than CPU.
Optimization Strategies:
- NTT: Stockham algorithm with shared memory, multi-batch processing
- Homomorphic Operations: Batched computations, minimized data transfers
- Polynomial Multiplication: FFT-based approach with coalesced memory access
- General: Persistent buffers, workgroup tuning, compute-shader-friendly algorithms
Benchmark Controls
Matrix Multiplication
Core operation for lattice-based cryptography and neural network processing.
CPU Time:
-
GPU Time:
-
Run benchmark to see results
NTT (Number Theoretic Transform)
Essential for lattice-based schemes like Kyber and Dilithium. Optimized!
CPU Time:
-
GPU Time:
-
Run benchmark to see results
Homomorphic Vector Operations
Fundamental operations for homomorphic encryption schemes. Optimized!
CPU Time:
-
GPU Time:
-
Run benchmark to see results
Polynomial Multiplication
Key operation in Ring-LWE based cryptography. Optimized!
CPU Time:
-
GPU Time:
-
Run benchmark to see results