CuPy
E97065
CuPy is an open-source array library for Python that accelerates numerical computing by providing a NumPy-compatible interface backed by GPU execution.
Statements (50)
| Predicate | Object |
|---|---|
| instanceOf |
GPU-accelerated array library
→
Python library → open-source software → |
| category |
GPU computing framework
→
Python scientific computing library → |
| compatibleWith |
NumPy
→
|
| designedFor |
numerical computing
→
|
| documentation |
https://docs.cupy.dev/
→
|
| enables |
GPU-accelerated numerical computations
→
|
| hasFeature |
NumPy-like linalg module cupy.linalg
→
NumPy-like random module cupy.random → |
| integratesWith |
Chainer
→
Dask → PyTorch (via array interoperability) → RAPIDS ecosystem → |
| license |
MIT License
NERFINISHED
→
|
| openSource |
true
→
|
| programmingLanguage |
Python
→
|
| provides |
NumPy-compatible API
→
drop-in replacement for NumPy on GPU → |
| repository |
https://github.com/cupy/cupy
→
|
| supports |
CUDA
→
CUDA streams → Fourier transforms → GPU acceleration → JIT compilation of kernels → NVIDIA GPUs → NumPy broadcasting semantics → NumPy indexing semantics → NumPy ufunc semantics → RawKernel interface → RawModule interface → array computing → broadcasting → cupy.ndarray core array type → custom CUDA kernels → distributed computing via Dask integration → linear algebra operations → memory pool for GPU memory management → multi-GPU computation → multi-dimensional arrays → random number generation → sparse matrices → universal functions → |
| targetPlatform |
CUDA-enabled systems
→
|
| typicalSpeedup |
faster than NumPy on compatible GPU workloads
→
|
| usedFor |
deep learning workloads
→
high-performance computing → machine learning workloads → scientific computing → |
Referenced by (1)
| Subject (surface form when different) | Predicate |
|---|---|
|
NumPy
→
|
influenced |