Run as fast as you can......and don't look back!
Run as fast as you can......and don't look back!
That almost sounded angry...
Looking at PG site, the top rated CPU only crunchers are Dual AMD Opteron(tm) Processor 6172 [Family 16 Model 9 Stepping 1] (48 processors). That's 4 CPUs @ 12 cores each.
RAC of just under 150K for CPUs alone (no GPUs installed).
2nd Best CPU only machine?
Dual AMD Opteron(tm) Processor 6172 [Family 16 Model 9 Stepping 1] (24 processors)
RAC of just under 112K for CPU alone (no GPUs installed).
Granted, there are over 650 machines with higher RACs, but all of those have one or more GPUs as well.
Spring 2008 Race: (1st Place)
The problem is that credit is based on RAC and Total credit granted by the project they are running. This has nothing to do with GPU vs CPU, just credit in general. The projects can dole out the credit anyway they want to. As much as they want, or as little.
What I see, is a fundamental problem with this thinking because CPU credit payout is so far below GPU payout that it's ridiculous in comparison. Yet, many many CPUs are crunching hard to do the work also.
Personally, I think it should be put into a pool where credit is based on the output of a particular "Box", which includes the GPUs AND CPUs as a whole of a persons individual clients. This brings problems however, because the computing world as we know it now, is a separated world between the projects. Some dole out CPU and GPU WUs, some give out only GPU WUs, so it's a mix of types of work all done on the same box.
Another problem is that there is NO standard for giving out credit. Rosetta for instance, wants their work done just as any other project. Yet if they REALLY wanted to have people crunching their work, they should be smart enough to realize that, if you want the work done quickly, dole out massive amounts of credit and the people will flock to their project in hordes.
This brings around another issue. If the projects can get the work done faster by handing out the credit, then that project gets more attention than others do. We see this all the time.
Now....If there were a standard for credit, based on the amount of crunching power a person's boxes can produce and ALL the projects gave out exactly the same credit for the work done, then we'd have a level playing field for everyone. No project would be getting ahead of another based on granted credit as it is now. It wouldn't matter what project a person had a preference for, as they would get the the amount of credit they earned from having done the work on any project in the Boinc community.
If, a person wanted to make a difference to a particular project, which we also do here at SETI, then having more boxes running a project will give that project the edge over the others. The point being however, is that the distribution of credit is an anarchy as it is right now.
Maybe someday it'll change, but for now it seems somewhat hopeless simply because of people that can't seem to get their heads outta their.......
My two cents
Yo-
Heh. Cross-project equality is impossible.
http://www.boinc-wiki.info/User:Nico...edit_scenarios
Unfortunately, you'll have to go through all 650 of those and sort out the CPU credits. There most likely are several Intel based CPU comps out-performing the AMDs you listed, they just had the sense to add GPUs, where the AMD guys didn't. Otherwise, this comparison isn't fair either
You know, I'm beginning to think that the only "fair" comparison is $/GFLOP. In other words, someone wanting to buy some hardware just needs to throw out their "budget $ figure" and then the hardware junkies on the team could go to work crunching numbers to figure out the BEST BANG for that member's upgrade BUCK.
Does that sound like the best solution, JPM? My whole point with the thread I stickied was I'd like to see the team's $$$ stretch as far as possible...and get the most credits for the team while doing so, of course.
I'm afraid of that comparison too, and I'll tell you why: It would be easy to go through every part for sale and build a system that costs $xxx and have it perform as fast as it could for that amount of money. But then, let's say that team member comes into an additional $200 and wants to upgrade the system he just built based on our recommendations. Come to find out, it will actually cost more to upgrade from what this member has to anything better because the parts used are all out of production and have been for a couple years. In that case, would it not be cheaper to get a slow system for the same money, but with newer tech so that future upgrades are possible?
Of course if a member never wanted to upgrade, a $/GFLOP comparison works perfectly fine. Based on the cost of a CPU only, INTEL (NOT AMD) wins that with the Core Duo T2400 @1.83GHz. Who would want that crap? Can't be upgraded to anything useful. But the #1 rule of crunching (which was in my sig on the old forums) is EVERY WORKING COMPUTER MUST BE UPGRADED!
Sacrosanct or alien2
If you wish to play (chicken) and begin a troll attack at AssociatedContent.com and Epinions.com against me then I will do your team a favor and leave.
Real simple man, I do this for fun. When it is no longer fun, I roll.
Your activity:
http://www.associatedcontent.com/art...html?#comments
http://www.epinions.com/review/INTEL...c_~ORS_details
Grown men don't have to put up with high school B.S. like this. We report in the place where it all began and let everyone know what is happening.
Just another little goldfish... steamrollin the competition one project at a time!
Staff Hardware Reviewer - BayReviews.com
Top Reviewer - Computer Hardware - Epinions.com