Skip to content

gpu_acceleration.md

Latest commit

 

History

History
52 lines (37 loc) · 1.49 KB

gpu_acceleration.md

File metadata and controls

52 lines (37 loc) · 1.49 KB

GPU Acceleration

SecActPy supports GPU acceleration via CuPy for large-scale analysis.

Setup

from secactpy import secact_activity_inference, CUPY_AVAILABLE

print(f"GPU available: {CUPY_AVAILABLE}")

# Auto-detect GPU
result = secact_activity_inference(expression, backend='auto')

# Force GPU
result = secact_activity_inference(expression, backend='cupy')

Performance Benchmarks

Dataset R (Mac M1) R (Linux) Py (CPU) Py (GPU) Speedup
Bulk (1,170 sp × 1,000 samples) 74.4s 141.6s 128.8s 6.7s 11–19x
scRNA-seq (1,170 sp × 788 cells) 54.9s 117.4s 104.8s 6.8s 8–15x
Visium (1,170 sp × 3,404 spots) 141.7s 379.8s 381.4s 11.2s 13–34x
CosMx (151 sp × 443,515 cells) 936.9s 976.1s 1226.7s 99.9s 9–12x
Benchmark Environment
  • Mac CPU: M1 Pro with VECLIB (8 cores)
  • Linux CPU: AMD EPYC 7543P (4 cores)
  • Linux GPU: NVIDIA A100-SXM4-80GB

CUDA Version Notes

# CUDA 11.x
pip install "secactpy[gpu]"

# CUDA 12.x (do NOT use [gpu] extra)
pip install secactpy
pip install cupy-cuda12x

Important (CUDA 12.x users): Do not use the [gpu] extra on CUDA 12.x systems — it installs cupy-cuda11x, which conflicts with cupy-cuda12x. If you already installed with [gpu], remove the conflicting package first:

pip uninstall cupy-cuda11x
pip install cupy-cuda12x