33 point gromacs

geordie · Post by **geordie** » Sun Feb 15, 2004 4:14 am

My XP2500+ at stock speed has been churning through 32 point gromacs work units in around 9 hours, and FAH_LogStats estimates 650 ppw.

Now I've got a 33 pointer (from p523) and it's estimating 12/wu, or 410 ppw

Obviously I expect some variation in production, especially between tinkers and gromacs, but wiping a third off the weekly production for 1 extra point from the wu

A bit more consistency in how wu's were scored would be nice

TheScarf · Post by **TheScarf** » Sun Feb 15, 2004 4:37 am

Hey geordie,

My two bartons are scoring as follows

1. xp2500@2164 on protein p259_4Na_Mg is looking at 627 PPW

2. xp2500@2000 on protein p253_AAAA Na+ stacked - 480 PPW

Both read from FahLogstats.

I'd like to know as well, might wnader over to F@H forums and see whats to find....I never really thought about it(famous last words), and just assumed it was the way it went....Post any answers if you find them too, it would be nice to find out!

folding as the_crate

haysdb · Post by **haysdb** » Sun Feb 15, 2004 2:57 pm

The "secret" to the mystery is in how the points are assigned. There is a standard benchmark machine, a 500MHz Celeron I believe, which is used to assign the points based on how long it takes to complate a work unit. If you folded with a 500MHz Celeron, your times would always be consistant with the points. For any other processor, how much faster it will be is dependent on the work.

One has to understand that the numbers reported by F@H LogStats apply ONLY to the current WU, and even that number changes from frame to frame. Electron Microscope is more reliable for determining the actual points over a range of work units.

Example: LogStats tells me I'm getting 101.27 PPD from my 2600+ Barton blade on the current WU. My actual production from that blade over the past two weeks, if I don't count a couple of Tinkers, is only 79.15 PPD

David

Code: Select all

Pts  Protein              Time     Date      Blade       CPU       PPD
33   p520_BBA5_pf       11:20:00 2/01/2004 ws4-Asus2 2600+ Barton  69.9
33   p522_BBA5_pf       11:28:20 2/02/2004 ws4-Asus2 2600+ Barton  69.0
33   p522_BBA5_pf       11:05:00 2/12/2004 ws4-Asus2 2600+ Barton  71.5
33   p522_BBA5_pf       11:03:20 2/12/2004 ws4-Asus2 2600+ Barton  71.6
33   p523_BBA5_pf       11:06:40 2/11/2004 ws4-Asus2 2600+ Barton  71.3
33   p523_BBA5_pf       11:11:40 2/14/2004 ws4-Asus2 2600+ Barton  70.7
18   p563_BBA5_ext      5:35:00  2/14/2004 ws4-Asus2 2600+ Barton  77.4
18   p563_BBA5_ext      5:50:00  2/15/2004 ws4-Asus2 2600+ Barton  74.1
13.1 p731_BBA5_md       4:50:00  2/07/2004 ws4-Asus2 2600+ Barton  65.0
11.1 p734_villin_sd_h2o 4:25:00  2/03/2004 ws4-Asus2 2600+ Barton  60.3
11.1 p734_villin_sd_h2o 4:20:00  2/07/2004 ws4-Asus2 2600+ Barton  61.5
32   p823_p53dimer823   7:36:40  2/14/2004 ws4-Asus2 2600+ Barton 100.9
32   p825_p53dimer825   7:36:40  2/14/2004 ws4-Asus2 2600+ Barton 100.9
32   p845_p53dimer845   9:13:20  2/02/2004 ws4-Asus2 2600+ Barton  83.3
32   p845_p53dimer845   9:06:40  2/06/2004 ws4-Asus2 2600+ Barton  84.3
32   p845_p53dimer845   9:15:00  2/10/2004 ws4-Asus2 2600+ Barton  83.0
32   p846_p53dimer846   9:10:00  2/05/2004 ws4-Asus2 2600+ Barton  83.8
32   p846_p53dimer846   9:06:40  2/06/2004 ws4-Asus2 2600+ Barton  84.3
32   p848_p53dimer848   9:00:00  2/13/2004 ws4-Asus2 2600+ Barton  85.3
32   p849_p53dimer849   9:15:00  2/06/2004 ws4-Asus2 2600+ Barton  83.0
32   p849_p53dimer849   8:51:40  2/13/2004 ws4-Asus2 2600+ Barton  86.7
32   p850_p53dimer850   9:08:20  2/01/2004 ws4-Asus2 2600+ Barton  84.0
32   p852_p53dimer852   9:11:40  2/09/2004 ws4-Asus2 2600+ Barton  83.5
32   p852_p53dimer852   9:01:40  2/15/2004 ws4-Asus2 2600+ Barton  85.1

NullObject · Post by **NullObject** » Sun Feb 15, 2004 4:10 pm

http://forum.folding-community.org indicate the following:

The due date is 2.5 * the benchmark time for the above mentioned 500MHz Celeron.
http://forum.folding-community.org/view ... t=due+date

According to Vijay Pande, the assignment servers will only give Gromacs to machines that complete their work with a performance fraction > .8 This does not guarantee that you will recieve a Gromac if your's is > .8, it does guarantee that you will recieve a Tinker if your's is < .8.

What about if it is = .8? Not sure, he isn't clear.

Vijay prefers that gromacs to go the fast machines and the tinkers to slow machines to help insure that the work is done before the deadline. This is all subject to change

The points per wu are based off of benchmarking that they have done. The points for a particular protien can and does change depending on updated benchmark results.

Does anyone use QD ? How does that compare to the other programs listed above.

roym@arm · Post by **roym@arm** » Sun Feb 15, 2004 6:23 pm

NullObject wrote:http://forum.folding-community.org indicate the following:

The due date is 2.5 * the benchmark time for the above mentioned 500MHz Celeron.
http://forum.folding-community.org/view ... t=due+date

According to Vijay Pande, the assignment servers will only give Gromacs to machines that complete their work with a performance fraction > .8 This does not guarantee that you will recieve a Gromac if your's is > .8, it does guarantee that you will recieve a Tinker if your's is < .8.

What about if it is = .8? Not sure, he isn't clear.

Vijay prefers that gromacs to go the fast machines and the tinkers to slow machines to help insure that the work is done before the deadline. This is all subject to change

The points per wu are based off of benchmarking that they have done. The points for a particular protien can and does change depending on updated benchmark results.

Does anyone use QD ? How does that compare to the other programs listed above.

This still does not explain why our fastest folders (i.e. P4 3.2's and Dual Xeons and even our Quad Xeons are getting nothing but sTINKERS...

We put on-line a P3 Celeron 500 and a P3 500 on Friday and both got Gromacks... the other 2 P4 3.2s got sTINKERS. Do you think maybe they got the formula reversed

Roy

haysdb · Post by **haysdb** » Sun Feb 15, 2004 9:23 pm

Roy,

It's definitely not a universal problem. The AS seems to be picking on ARM Systems in particular.

I think you and Stevo should post a serious question over at folding-community.org. You have a legitimate beef with the work you are being assigned. Sure, our cpu time is "donated" to Stanford to use however they please, but you are still within your rights to question why ARM Systems seems to be getting singled out. I know you aren't alone, there have been other complaints, but I haven't read about anyone getting as many Tinkers as you guys have.

Vijay Pande himself monitors those forums closely, and the moderators have the "ear" of people at Stanford. I think it's time to start a thread over there and make your case. I suggest the server forum, since that seems to be the issue.

David

roym@arm · Post by **roym@arm** » Sun Feb 15, 2004 9:48 pm

haysdb wrote:Roy,

It's definitely not a universal problem. The AS seems to be picking on ARM Systems in particular.

I think you and Stevo should post a serious question over at folding-community.org. You have a legitimate beef with the work you are being assigned. Sure, our cpu time is "donated" to Stanford to use however they please, but you are still within your rights to question why ARM Systems seems to be getting singled out. I know you aren't alone, there have been other complaints, but I haven't read about anyone getting as many Tinkers as you guys have.

Vijay Pande himself monitors those forums closely, and the moderators have the "ear" of people at Stanford. I think it's time to start a thread over there and make your case. I suggest the server forum, since that seems to be the issue.

David

Hi David,

You are definitely right. If you look at F@H Stats, you'll notice that we have 77 active processors, compared to around 22-40 when we first started folding 2+ weeks ago but yet our points production only marginally increased while doubling our processors. This started when, first, we could not get any jobs for up to 36 hours. Then, we have gotten only sTINKERS on our fastest processors. By my estimate, we should be producing +1K PPD instead of 600-700 PPD.

We thought that this might be just a storm that will pass after a couple of cylces but its lasted longer than I think normal. Your suggestion to bring this up on the server forum at Stanford is great idea which time has come.

Roy

roym@arm · Post by **roym@arm** » Mon Feb 16, 2004 9:32 am

haysdb wrote:Roy,

It's definitely not a universal problem. The AS seems to be picking on ARM Systems in particular.

David

Hi,

I checked all of our folding clients at ARM this morning and it is worst than before. Every one except one is now churning away at sTINKERS.

This is definitely not normal...

Roy

haysdb · Post by **haysdb** » Mon Feb 16, 2004 10:17 am

DEFINITELY not normal.

I talked to Stevo a couple of days ago about adding -verbosity 9 to the command line and emailing me a FAHlog.txt. I'd like to take a look at one just to see if anything at all looks out of place. The offer still stands.

David

NullObject · Post by **NullObject** » Mon Feb 16, 2004 12:42 pm

David's suggestion is a good one and probably the only sure way of resolving the trouble with tinkers( is that kind of like trouble with tribbles?) without going through endless speculation. When someone else brought up the fact that they were recieving only tinkers, Vijay Pande said he would look into it.

Also there was a post where OC-AMD asked Vijay about a special team flag change. It seemed to me that it was a change requested by OC-AMD. But that is only my impression since I wasn't privy to the rest of the conversation. Maybe he gets special treatment because he is the #1 daily folder but then ARM is *special* too

...and no I didn't mean that ARM takes the short bus to school

roym@arm · Post by **roym@arm** » Mon Feb 16, 2004 7:50 pm

NullObject wrote:David's suggestion is a good one and probably the only sure way of resolving the trouble with tinkers( is that kind of like trouble with tribbles?) without going through endless speculation. When someone else brought up the fact that they were recieving only tinkers, Vijay Pande said he would look into it.

Also there was a post where OC-AMD asked Vijay about a special team flag change. It seemed to me that it was a change requested by OC-AMD. But that is only my impression since I wasn't privy to the rest of the conversation. Maybe he gets special treatment because he is the #1 daily folder but then ARM is *special* too

...and no I didn't mean that ARM takes the short bus to school

Hi Guys!

Well, I just installed a folding client on my laptop (P3 650) at home and guess what! I got a GROMACK. STEVO claims that all his folding units at home are generally getting GROMACKs.

Today, I added 3 more folding clients at the office: 2 - P4 2.0 Celerons, 1 - AMD XP1800 and all 3 of them got sTINKERS. Steve is beginning to think our IP at the office may have been flaged to receive mostly sTINKERS and maybe due to the F@H software is not XEON aware. Since most of our folding clients at the office are XEONs maybe we are now getting doled the sTINKERS because the XEONs maybe registering as PII's

Anyway, STEVO is composing an e-mail to Vijay outlining our difficulties and hope he will have an explanation or relief. If we can get out of sTINKERland, in my estimation, we can be doing closer to 2K PPD.

ALSO, something interesting STEVO pointed out to me. Our average points per WU is 47. SPCR Folding Team's is 33.77 which indicates we are getting unusually high number of sTINKERS compared to the average SPCR folder.

Now check the MacOS teams average points per WU as well as the HardOC's average. It looks like they are getting the cream dela creame of the Gromacks.

Just something to TINK about.

Roy

Stevo@ARM · Post by **Stevo@ARM** » Tue Feb 17, 2004 1:43 am

roym@arm wrote:
...Anyway, STEVO is composing an e-mail to Vijay outlining our difficulties and hope he will have an explanation or relief. If we can get out of sTINKERland, in my estimation, we can be doing closer to 2K PPD....

Roy

See this link to my post for Vijay on the Stanford Folding forums in the a thread he actually started himself called Rebalancing towards more Gros:

Asking Vijay For Help with ARM Tinker problem

David, if you can jump in to help support me on that forum I would appreciate it also. I just know we can put some MONSTER numbers on the board if we can get a proper mix of WU assigned to us.

Stevo
----------------
Asking the Folding Master for guidance and help

haysdb · Post by **haysdb** » Tue Feb 17, 2004 10:18 am

I dare say you need no "support" from me, but thank you for asking. That was a very nicely crafted post, posted to precisely the right thread. The ball is definitely in their court.

Edit: You definitely have their attention.

David