Package 'tulpaGPU'

Title: CUDA Backend for the 'tulpa' Bayesian Engine
Description: Runs the device operations of the 'tulpa' engine on an NVIDIA graphics processing unit: dense and sparse Cholesky factorizations, factorizations batched across the cells of a nested Laplace grid, basis products, observation scatter and fused likelihood kernels. The CUDA driver and the 'cuSOLVER', 'cuBLAS', 'NVRTC' and 'cuDSS' libraries are loaded at run time, so installing the package needs only a C++ compiler. Once loaded, 'tulpa' sends each operation to the device when it lies above its measured size crossover, or always under control$device = "cuda".
Authors: Gilles Colling [aut, cre, cph] (ORCID: <https://orcid.org/0000-0003-3070-6066>)
Maintainer: Gilles Colling <[email protected]>
License: MIT + file LICENSE
Version: 0.1.0
Built: 2026-11-21 23:30:43 UTC
Source: https://github.com/gcol33/tulpagpu

Help Index


CUDA device availability

Description

Report the CUDA driver, the selected NVIDIA device, and the runtime libraries tulpa's device routes use. Installation requires a C++ compiler only; the libraries are loaded when a device operation is first requested.

Usage

gpu_info()

## S3 method for class 'tulpa_gpu_info'
print(x, ...)

Arguments

x

A gpu_info() result.

...

Ignored.

Value

An object of class tulpa_gpu_info with availability, the reason for unavailability, device properties, and library paths and reasons.

Examples

gpu_info()