In "Tapping the matrix" [1], a talk given by Carlos Justiniano (founder of Chessbrain), GPU is described as an active project with one big drawback. It is implemented, we quote, in an "obscure Pascal dialect". Ok, this was like a trigger for
... [More]
nanobit:C/C compatibility achieved for plugins----------------------------------------Since the beginning, GPU was intended to load standard dynamic link libraries. We used pointer of chars instead of the very practical Delphi Strings for the communication between core and plugins. Using Bloodshed Dev-C [2], an amazing C IDE written in Delphi (ehm) and under GPL, nanobit was able to compile a C-DLL. The source code for the C plugin is under /src/dllbuilding/c_example. We had to change a little the TStack definition, so that the parameters were understood by the plugin. This prompted us to recompile all plugins, and recompiling everything was easy step thanks to Samuel Herzogs "DelphiPackageTool". So, we can now easily plug-in any sort of C/C -code into the GPU system.As final remark, frontends written in any language can communicate with the core, if they use the UDP/IP channel mechanism. We will try to provide some of them in some crazy languages, if we manage to get them working.Water Cooling-------------Running a GPU node for long time is a challenge, one has to fight with dust, noise and neighbourhood.A big thank to Stefano Godenzi (Ninja), who managed to install the Thermaltake [3] Watercooling system inside spartacus, one of the titled computers of the network. The pump as delivered by factory was wrong cabled, and some adjustement was needed. We plan to cover his work into a webpage, which we will made available on this site.Nova-TS Team------------Laurent Wattieaux put up to 6 computers on the network, so that his Nova-TS [4] team often was top team in the GPU Cluster statistics [5]. By the way, we are collecting statistics since about a year, and the network itself is operating since one and a half years. We probably had an average of 6 nodes always on the network, with tops of 16-18 nodes. Ah, nodes with the LNX flag, are GPUs running under wine emulator.Version 0.926 is the current stable release. It is compiled for C/C compatibility, has a whiteboard able to display nanobit's seismograph (more on this in a later news ;-), a netmapper which resizes better and let's the cluster spin on the center of mass. Bugfixes to autopdate, to the core as usual. The source release now includes a Changelog file, such that developers can better follow changes. Do not hesitate and ask our mailing list [6] for developer CVS access![1] Tapping the Matrix, http://bof.linuxforum.dk/2005/slides/Carlos/tappingthematrix.html[2] Bloodshed Dev-C , http://www.bloodshed.net/devcpp.html[3] Thermaltake, http://www.thermaltake.com[4] Nova-TS team, http://www.nova-ts.com/[5] GPU Cluster Statistics, http://www.tea.ch/gpu/statistics/gpu_statistics.html[6] GPU mailing list, [email protected] (0 comments) [Less]
|
