New reference machine for points ...
csimon
Acadiana Icrontian
Vijay Pande:
New reference machine for FAH points determination
In the coming week, we will roll out our new points determination system. The advantages are:
1) There will be much more standardization for points and deadlines.
2) The points and deadlines will be determined on a more modern machine (2.8GHz P4), although they will be scaled to match current Gromacs projects.
What is the upshot of all of this? Put simply, Tinker WU points are about to go way up, Gromacs will be mostly the same, and there will be less fluctuation between projects. We take the points in FAH very seriously and try not to make such signifiacnt changes without much consideration. Indeed, this is the first change in the history of the project. However, this change is critical to account for changes in FAH donator computer hardware over the last 4 years.
We will post more specific info in the FAQ as we roll out the new points.
what this means to me is more consitency throughout the pointing system.
Check out the thread here
New reference machine for FAH points determination
In the coming week, we will roll out our new points determination system. The advantages are:
1) There will be much more standardization for points and deadlines.
2) The points and deadlines will be determined on a more modern machine (2.8GHz P4), although they will be scaled to match current Gromacs projects.
What is the upshot of all of this? Put simply, Tinker WU points are about to go way up, Gromacs will be mostly the same, and there will be less fluctuation between projects. We take the points in FAH very seriously and try not to make such signifiacnt changes without much consideration. Indeed, this is the first change in the history of the project. However, this change is critical to account for changes in FAH donator computer hardware over the last 4 years.
We will post more specific info in the FAQ as we roll out the new points.
what this means to me is more consitency throughout the pointing system.
Check out the thread here
0
Comments
KingFish
Jon
Keep this in mind for the future because as it stands now we still do benefit from the athlon/gromacs/-advmethods
As far as worries that the Double Gromacs might replace Gromacs quickly, though, I think we will see a phase-in that matches the hardware put online for the Folding effort. It is NOT suddenly that most of the boxes are P4s, rather that bunches of the boxes are faster.
By taking and using a P4 2.4 GHz Northwood bench box at folding (I doubt this will be a Prescott box), we have a bencher that matches the speeds of calcing of the majority of faster boxes. To a degree, note that the preferred days might get more tight, but the deadline for points credit is likely to get not too greatly changed-- doing so would put the folks with slower boxes at a disadvantage, so it is not likely. Instead, the stats change already done and accounted for here allows for a wider range of calcs and for compensation for faster benching-- keep deadline for getting points same, preferred tighter.
John D.
I knew this was going to happen a long time ago.
Tinkers will get a huge boost for AMDs and less for P4s but a boost all around. Having extra optimizations wont help as much anymore.
Really, this is good news
But I could use some explanation, though. I though the -advmethods flag was to get Gromac cores. I didn't realise it had something to do with Tinkers.
Will it hurt to have the -advmethods flag in, once the new system starts?
If it does, I'm counting on your guys to tell me it's time to take that flag out :bigggrin:
They already have been. My Intel box has been outproducing my AMD box by a landslide. Even with -advmethods turned on for the Athlon, it just doesn't receive many Gromacs; when it does, it just can't keep pace at all with the P4 at hyperthreading. Tinkers are much more efficient with my Barton. Unfortunately the rewards for the Tinkers just hasn't been good.
"As announced in the previous thread, the new points are in. You can see the details of how we determine points in the FAQ."
"the new points will be awarded to WUs that come in after the switch, even if they were assigned before the switch. Points are awarded when they are received and the assign date is not used for credit (other than determining deadlines, etc)."
-advmethods is the flag used to "request" gromacs however it doesn't mean you will always get them ...sorta like setting a preference.
Watch this thread ...as soon as the tinker points rise then you can do away with the -advmethods flag. We'll let you know!
Yep Tinkers may no longer be sTinkers anymore. AMDs will certainly see an edge in this department. As a result -advmethods may end up hurting the Athlons since that allows for more gromacs when Tinkers may be worth me.
Stay tuned to this thread for more details.
Edit: Post from frontpage.
GAH tinkers should get a boost as well.
KingFish
I am removing -advmethods from my athlons only as we speek but I am leaving -forcesee in the event I get gromacs (tinkers are not affected by -forcesse).
I think things are going to get interesting!
KingFish
Nice but will that make them take longer time. Doesn't really matter BTW. It's working, i'm getting points, and thegr81 doesn't stand a change
p1110_L939_K12M_nat_min1 = 249 points
as far as I know this protein has 393 atoms and a return date of 50 days and has the highest points value along with the p1111_L939_wt_ext_444K.
Hope you get one or two soon!!!
actually I got the info from EM and double checked it on the project summary page at stanford. Maybe you need to update the EM 251 and/or the latest EM protein data file?
the latest EM protein data file was released two days ago!
Though, I'm crunching away at a pretty tasty Tinker right now...
My baraton at 2.3G is working frames in 4:38, that is 200ppd.
I do like this.
I just got my second p212 in a row. It's a double gromacs, so even for my P4 it takes longer than usual. I guess it's not too bad, it's worth 135 points.
Do you guys think I should take out the -advmethods on my P4 box?
Looks like Stanford's done their homework thoroughly. My P4C box and AMD box (with the P1110 processing) are now running fairly evenly now for points production, according to Electron Microscope.
Current activity:
P4C@3500, Hyperthreading, two instances of Folding simultaneous --
P936_fkfe2_all @255.384/day
plus
P259_4Na-Mg @87.48/day
XP 2800+@2200, one instance of Folding--
P1110-L939_K12M_nat_min1 @188.712/day
342.864 : 188.712 = EM calculated comparison
I believe I'm going to drop the -advmethods flag off a few AMD machines and experiment myself, as the Tinkers are no longer (s)Tinkers when it comes to points production.