GRAPE Software Package

GRAPE Software Package is a collection of user libraries, utilities, documents, and sample programs for GRAPEs. It supports not only KFCR's GRAPE-9, GRAPE-DR and GRAPE-7, but also GRAPE-6A, GRAPE-6BX, and Phantom-GRAPE-5^note1. In addition, CUDA G5/G6 Optional Package provides G5/G6 APIs for NVIDIA's CUDA devices.

Contents of the Package:

A GRAPE-5-compatible user library (for GRAPE-9, GRAPE-DR, GRAPE-7, GRAPE-6A, GRAPE-6BX, Phantom-GRAPE-5)
A GRAPE-6-compatible user library (for GRAPE-9, GRAPE-DR, GRAPE-6A, GRAPE-6BX)
The Host Interface Bridge (HIB) library and its device driver for GRAPE-9, GRAPE-DR and GRAPE-7.
The HIB library and its device driver for GRAPE-6BX.
A device driver for GRAPE-6A.
Sample programs.
User's guide and other documents.

Note

All libraries have C-language interface. Libraries for GRAPE-9, GRAPE-DR and GRAPE-7 have Fortran interface, too.
All programs are provided with source codes.

Download

The GRAPE Software Package can be downloaded from the hyperlink below.

Old Versions:

version 1.5.x (2013/10/09)
version 1.4.x (2013/5/10)
version 1.3.x (2012/7/20)
version 1.2.x (2010/6/22)
version 1.1.x (minor changes for ver. 1.1)
version 1.0 (2009/7/17, no English documentation)

License and Copyright

The MIT software license is applied to the GRAPE Software Package (hereafter "the Software"), except for files in 'driver' directory, to which GNU General Public License (hereafter GPL) is applied.

The copyright of the software belongs to K&F Computing Research, except for device drivers, Phantom-GRAPE-5 library, and user libraries of GRAPE-6BX and GRAPE-6A, whose copyrights belong to alternate individuals.

See 00readme for the detail.

Related Information

CUDA G5/G6 Optional Package

CUDA G5/G6 Optional Package provides G5/G6 APIs for NVIDIA's CUDA devices. By adding this package into the GRAPE Software Package, application programs developed for conventional GRAPEs can be run on CUDA devices.

Contents of the Package:

GRAPE-5-compatible user library (for CUDA device)
- Via the user library, a user can handle multiple CUDA devices (i.e. GPUs) as independent devices, or as a single device. Devices to incorporate a run can be specified by an environment variable.
- Fortran interface is provided, as well as C interface.
- A cutoff function is supported for the P³M Method.
- A new API is provided for the Multiple Walks Method^note2.
- Two libraries are provided to support different numerical precisions. One is for single precision (32-bit floating-point), the other one is for GRAPE-5-upper compatible precision (64-bit floating-point for addition, 32-bit for multiplication).
GRAPE-6-compatible user library (for CUDA device)
- Via the user library, a user can handle multiple CUDA devices (i.e. GPUs) as independent devices, or as a single device. Devices to incorporate a run can be specified by an environment variable.
- Fortran interface is provided, as well as C interface.
- The library performs addition and multiplication in 64-bit and 32-bit floating-point format, respectively.

All programs are provided with source codes.

Products Line up

We offer the following line up for CUDA G5/G6 Optional Package. Please contact info@kfcr.jp for purchase. We accept the following payment methods : PayPal, IPMO, and Wire Transfer.

CUDA G5/G6 Personal Edition (International Price: 315,000 JPY)

A product for up to 10 users. There is no limitation for the number of local copies installed into different PCs.

The product includes one-year free version up service, and one-year technical support.

CUDA G5/G6 Institutional Edition (International Price: 945,000 JPY)

A product for more than 10 users. There is no limitation for the number of local copies installed into different PCs. There is no limitation for the number of users as long as the user belongs to the user group.

Even if a user belongs to a group that consists of more than 10 members (e.g. a university, an enterprise), the user can purchase the personal edition, as long as the number of actual users does not exceed 10.

The product includes one-year free version up service, and one-year technical support.

CUDA G5/G6 Evaluation Edition

We are offering a 30-day free evaluation for the software. Please contact info@kfcr.jp for request.

Measured Performance

CUDA G5 Library (GRAPE-5 equivalent precision, direct summation, shared timestep)
number of particles	GTX 480 1GPU	GTX 295 1GPU 2GPU 3GPU 4GPU				GTX 285 1GPU
1k	170	89	93	87	80	104
8k	498	247	423	521	590	301
64k	576	276	535	774	995	341
256k	587	280	554	818	1073	346
CUDA G5 Library (single precision, direct summation, shared timestep)
number of particles	GTX 480 1GPU	GTX 295 1GPU 2GPU 3GPU 4GPU				GTX 285 1GPU
1k	247	149	133	115	100	166
8k	918	526	825	947	1022	615
64k	1157	625	1194	1688	2123	740
256k	1192	640	1256	1837	2384	759
CUDA G6 Library (GRAPE-6 equivalent precision, direct summation, individual timestep)
number of particles	GTX 480 1GPU	GTX 295 1GPU 2GPU 3GPU 4GPU				GTX 285 1GPU
1k	26	16	15	13	12	14
8k	152	85	122	138	145	94
64k	261	136	257	361	451	142
256k	252	135	250	366	480	151

Wallclock time spent for a calculation from nj particles to ni particles is measured for various number of particles. Figures are Gflops values (38 and 57 floating-point operation counts are assigned to a single pairwise interaction for shared and individual timestep scheme, respectively).
The host computer for GTX 480 and GTX 295 is Intel Core i7-920 (X58 Express chipset) That for GTX 285 is Intel Core i5-750 (P55 Express chipset).

License and Copyright

Permission for use of the CUDA G5/G6 Optional Package (hereafter the "Software") is granted only to owners of a copy of the Software. The Software may not be redistributed. The Software may be modified by the owner, as long as the modified ones subject to this license agreement.

The copyright of the Software belongs to K&F Computing Research. Co.

Related Information

NVIDIA Co. CUDA ZONE

note1: Phantom-GRAPE-5 is a software based on Phantom-GRAPE Limited accuracy version for BH-tree code developed by Keigo Nitadori (RIKEN). Phantom-GRAPE is neither a dedicated hardware nor a control library for a hardware. It is a library for gravitational calculation that works on conventional x86_64 CPUs. Its code is highly optimized at assembly-langulage level, taking full advantage of the Intel SSE technology.
note2: The Multiple Walks Method is an algorithm named and integrated into the Barnes-Hut Treecode on GPUs by Dr.Hamada (Nagasaki University) and Dr.Nitadori (RIKEN). A many-body simulation perormed using the algorithm won the Gordon Bell Prize in 2009.
Contact : info@kfcr.jp