Dr. Kasson posted:
However, those numbers must include all Windows SMP and OSX SMP systems in addition to all SMP Linux systems.kasson wrote:The conversation about what the role of BA is in the project is a good one to have. The BA program has been of particular scientific benefit to my research group, but the FAH project as a whole has needed to position BA within a broad range of scientific needs.
Data that helped inform where to place the new thresholds:
~5% of active FAH machines with SMP>2 are at least 24 cores
~4% of active FAH machines with SMP>2 are at least 32 cores
Adjusting deadlines to match new hardware requirements will be a necessity...
If you look at the OS Stats page, there are typically about 5000 active Linux clients. Incorrectly assuming ALL of those systems are BA systems, because we know they are NOT all BA systems, that makes the number of active Linux systems at most 2% of all the active Windows, OSX and Linux SMP clients combined (5000 out of 225,000).
Clearly the numbers from Kasson either contain all OS types with those core counts, or his numbers are wrong, or the OS Stats page is wrong. Make your own assumptions here, as expected.
Grandpa_01 posted:
I hope you aren't basing suggestions on these faulty assumptions. The suggestions would then also be faulty.Bigadv = Greatest amount of science done much larger strings and quick returns are valued
GPU = A large amount of science done and much shorter time frame than smp may do smp type work in future
SMP = slowest form of folding lesser value but still needed
First, SMP and BA work units, when processed on the identical BA system do the same amount of science. Fahcore_a5 for BA work is based on the SMP Fahcore_a4. There is no speed difference on an identical system, simply a points difference. The QRB and BA bonus points formula also bears this out. BA points are simply the SMP points value, multiplied by a speed factor. And the speed factor difference between the two is not the fahcore speed, but the assumed faster speeds of BA systems with much higher core counts than a typical SMP system.
And if BA systems are only 1% of the total folding masses, then they certainly do not do the most science, they simply do the quickest science, and only because of their increased core counts. And the 1% does not set policy for the other 99%.
As for GPUs, it has been proven to be able to fold SMP work, hence the SMP points and QRB given to Fahcore_17. Those new projects are weeks if not days away from being released. There is no MAY about it. Fahcore_17 has done and WILL be doing SMP work going forward. (Hint, P7810 and P7811 WUs were GPU WUs on Core_17, the follow-on P7812 was released as SMP WUs on Core_a4.)
I also have a suggestion along the lines of equal pay for equal work. Folding@home has always tried to be hardware and software agnostic, however various operating systems have been more difficult on which to develop software than others. For example, they currently only have BA work for the Linux OS. But because the SMP fahcore_a4 completes scientific work at the same speed as the BA fahcore_a5 on identical hardware, they should also add a BA style bonus to ALL SMP work units processed on systems with 24 cores or more starting with the February 17th deadline. There is no reason that any donor with BA level hardware should get less points for the same amount of work they do just because of the choice of their operating system. This would also help bring some non-Linux big iron back to Folding, and help with the backlog of SMP work units. And as a result, the actual BA program needs to be adjusted less often.