Last Monday, ngspice made announcement on their site for release of CUSPICE. CUSPICE is ngspice on CUDA platform. Following is the words taken from their official release:
CUSPICE is the revolutionary ngspice on CUDA platforms. The ngspice simulator has been modified to exploit the parallelism offered by CUDA platforms. The code has been uploaded in the source repository. CUSPICE now supports only a small (growing) set of ngspice devices: BSIM4v7, Capacitor, Self and Mutual Inductor, Current Source, Resistor and Voltage Source.
CUSPICE requires an NVIDIA video card with Fermi (or newer) architecture and a working CUDA enviroment installation.
CUDA is a platform from NVIDIA. It is parallel computing platform and programming model which works on their GPU. Using CUDA, GPU can be used for general purpose computing and their architecture allows them to run multiple threads. ngspice on CUDA platform will enhance the parallelism of the simulation.
While discussing on a group in Linkedin, ngspice developer Francesco Lannutti reports speedup upto 9x which looks promising for a SPICE simulator and results are same or nearly same to the CPU version for ngspice. At this point of time, a limited set of devices are supported which includes BSIM4v7, Capacitor, Self and Mutual Inductor, Current Source, Resistor and Voltage Source. The code is still in Beta and is under development.
I would suggest the interested people to test it out as it will also be contribution on our side for an open source EDA. The user guide and release notes can be found here.