Especially for BOINC.RU a few questions answered one of the project managers Slicker [TopGun](September 2011) (Russian version)
1. What difficulties arise in the work of the project?
When developing an application for CPUs, once you get it working, it works well on all machines. Developing applications which run on GPUs is a much larger challenge. If the application is written to make use of the features of the newest GPUs, it will not run properly on older GPUs. An application written to work on all GPUs won’t run at its optimal speed on newer GPUs. After an application has been developed, GPU vendors may release new drivers which cause the GPU application to fail. I have several unreleased versions of Collatz that work well for one driver or GPU but not the majority. There is a lot of trial and error.
2. Are you satisfied with the number of participants in your project?
Yes, but it would always be nice to have more!
Collatz struggled for a while when it first start as it became popular because the hardware wasn’t good enough and the workunits were too small for newer, faster GPUs. A faster server with much more memory and increasing the size of the work units has worked well. I often see an increase in the number of participants when other GPU projects are off-line, but the server has handled those increased quite well in the last year and a half.
3. Whether your expectations from this project have justified? Are you satisfied with the results?
At the time the Collatz project began, Milkyway@Home was the only project that supported ATI graphics cards and was shut down or out of work more often than not. At a minimum, I wanted participants to have an alternative if Milkyway was down or out of work. I also wanted to show that a person working part time on the project and with no outside funding could build and run a BOINC project as well or even better than many of the other projects which receive government or university funding and have a large full-time staff. Thanks to Crunch3r’s work on the BOINC client and Gipsel’s work on optimizing the ATI applications, Collatz was the first project to support ATI graphics cards without requiring an app_info.xml file. Gipsel’s hand written GPU assembly code show just how much faster GPU processing can be. I’m very happy with those results and the opportunity to work with Gipsel and Crunch3r. As far as proving or disproving the Collatz conjecture itself, the participants are checking about 40 quadrillion (a.k.a. 40,000 billion or 40,000,000,000,000,000) numbers each day which is far more than I ever expected. When considering the size of the numbers used in calculations in the PrimeGrid project, ours are still very, very small and we have a long way to go which means we’ll probably never run out of work. But, I’m very happy with the results so far.
4. Are you planning to continue working on the project?
Absolutely. Working on the project is a reminder to me that computers can be fun. As an IT professional, I work on computers all day every day. But, as we all know, jobs are not all fun. If they were fun, they would call it a hobby and not work! Regardless, I’ll keep the project running as long as it is fun and I remain gainfully employed.
5. In what direction it is planned to develop the project?
OpenCL is next on the list of things to do. The hard part is making the AMD/ATI GPUs run as well with OpenCL as they do with Gipel’s hand-written assembly code. That’s not easy. Since OpenCL is compiled at runtime, one can’t control the GPU code. And, since OpenCL is really just a flavor of CUDA and not even close to AMD’s CAL Stream SDK, making it work on AMD GPUs is like putting a Lada’s engine in a school bus and then expecting the school bus to run exactly like a Lada. Because it is generic, OpenCL requires more tweaking and tuning to get the most performance out of it. That takes time to learn. But, it appears that is the direction both AMD and nVidia are headed so we need to keep current.
6. What would you like to wish the participants of your project?
Participants are volunteers. It costs them money to run their computers. Participants don’t get paid (except in cobblestones which can’t actually buy anything). That sounds like a hobby to me and hobbies should be fun. If they want to have fun while contributing to Collatz, that’s even better. But, if they prefer it is more fun to crunch on some other project, I’m OK with that too. Join a team. Post to message boards. So long as people have fun with it, they will continue to participate. The success at foldit (http://fold.it/) with the AIDs enzyme found recently couldn’t be done without the participants and the more fun they have, the longer they participate. So, I wish that the participants would have fun.
7. In a press there were messages that the theorem of Collatz is proved. What do you think of it? As it can affect the project?
People have been trying to solve the Collatz conjecture for years. There have been at least three announcements stating someone had claimed to solve the Collatz Conjecture in the past two years. When viewed under scrutiny, all three have been flawed and are not valid. If a proof is ever discovered and accepted by the mathematical community as valid, that would be the end of the Collatz project. But so far, the only accepted proof so far is the proof that shows it cannot be solved other than by brute force..