FS3 Alpha: The Value of 'Just Doing It'
While researching HPC (high performance computing) and supercomputing with GPUs today, it became readily apparent short-term action is needed on the FS3 project.
It would be easy to spend weeks, months and maybe even years reading about, thinking about and talking about the advantages of GPU-based supercomputers over CPU-based supercomputers. Intel is working on HPC and so is AMD. Cray is building a new supercomputer for Edinburgh University for $222,000,000. Intel is rumored to be working on high end graphics cards. Etc, ad infinitum.
An MIT Technology Review article mentioned all the work being done at that university related to parallel programming models. In an HPC Wire editorial, there is more discussion about what programming languages are best for supercomputer programming.
I have a friend who spent years researching what personal computer to buy. He kept telling me that if he bought a particular computer, he knew it would be outdated very soon, plus he could save money if he waited a few more months and the price dropped on the older models. And so it went for quite a few years...
The FS3 project is likely to move ahead much more quickly if we build the six box, 12-GPU Alpha system suggested by David T than if we wait until we raise enough money to build the 50-GPU Beta system. Once the Alpha hardware is put together, the team will start finding out what the challenges are and will start developing the solutions. Coding for FS3 will be much more challenging than hardware, and every day we wait to start is another day of lost opportunities. We probably won't have the Alpha for BarCampMadison on 3 - 4 March 2007, but it's still possible to get the system put together for one or two university road trips before MinneBar on 21 April.
During the next FS3 weekly teleconference, I'm going to propose planning be started for the software/coding part of the project -- if it hasn't already been started. The FS3 Alpha system can be assembled for $15K - $30K, depending on its final design. Concurrent with software planning are a number of other activities, including putting the organizational structure for the FS3 project in writing (LLC or other entity, with all the important details spelled out), following up with AGW on the $250K grant applications, developing a professional FS3 website, and making connections with potential project members.
Lots of activities, lots of things to plan for. FS3 Alpha -- here we come.