New bigadv project 6904

may i be worthy

[H]ard|DCer of the Month - December 2010
Joined
Aug 17, 2010
Messages
1,186
Nice !!...time to fire up my second rig ;)

looks like they know what upcoming Bulldozer can handle :D
 
Last edited:
They need to tighten up the deadlines on these WU.

~14 days is just crazy.
 
interesting to see what "a little slower" means
 
The 6903 is:
Code:
Base points 28384
Final deadline: 17 days
Preferred:  10.20 days

Comparing to 6904:
Code:
Base points 39426
Final deadline: 23 days
Preferred: 13.8 days

base on my limited Math knowledge, .. it would be 1.35 time longer ... :D ...
 
That was quick. I just hope it comes to Windows soon since i'm just not in the mood to fiddle with my folding rigs right now. Do you guys think that these bigassadv work units will ever be allowed to run on anything less than 12 cores? With those kinds of bonus deadlines, even our lowly Bloomfields wouldn't have any problems getting done in time.
 
That was quick. I just hope it comes to Windows soon since i'm just not in the mood to fiddle with my folding rigs right now. Do you guys think that these bigassadv work units will ever be allowed to run on anything less than 12 cores? With those kinds of bonus deadlines, even our lowly Bloomfields wouldn't have any problems getting done in time.
It's a good question. Pande's contradictions confuse me.

With these huge bonuses, they're basically telling us that faster work = better science. So, they incentivise us to turn in work as quickly as possible.

But then, they also leave the door open for the "slow" machines by leaving these really long deadlines.
 
It's a good question. Pande's contradictions confuse me.

With these huge bonuses, they're basically telling us that faster work = better science. So, they incentivise us to turn in work as quickly as possible.

But then, they also leave the door open for the "slow" machines by leaving these really long deadlines.

Yeah, I wish their logic was either more consistent or more apparant in why they do this. This effectively shuts out the Zambezi 8 cores from folding these despite the fact that they have 8 cores while the those of us with Gulftowns only have 6 cores. If you remember, we know that the Thubans will have no problem with bigadv with a decent overclock, and it has 6 cores while the Bloomfields only have 4. The Bloomfields also require an overclock as well in order to clear the 2684's on time. This logic is really perplexing to me. If I was still one of those AMD fanboys like those in the AMD cpu subforum spreading a bunch of conspiracy theories, i'd start claiming that the PG are Intel fanboys.
 
Yeah, I wish their logic was either more consistent or more apparant in why they do this. This effectively shuts out the Zambezi 8 cores from folding these despite the fact that they have 8 cores while the those of us with Gulftowns only have 6 cores. If you remember, we know that the Thubans will have no problem with bigadv with a decent overclock, and it has 6 cores while the Bloomfields only have 4. The Bloomfields also require an overclock as well in order to clear the 2684's on time. This logic is really perplexing to me. If I was still one of those AMD fanboys like those in the AMD cpu subforum spreading a bunch of conspiracy theories, i'd start claiming that the PG are Intel fanboys.

I might be prone to use other solutions for my amd x8 system ...
 
You can just tweak linux and/or your VM and tell it that it's running on 12 cores if you want to run these units on slower hardware. With a preferred deadline of 14 days I don't really see how it would negatively affect PG to have SB setups etc. folding these units. One downside would be less WU for the multi CPU rigs:p but other than that if a i7 970 can run them why not a 2600k?
 
I think that's giving PG too much credit. I think they're just confused and illogical.

:D

Keep in mind the deadline is part of the bonus formula. If they shorten the deadline, they have to adjust the base points and/or k factor so that the total credit works out to what they want.
 
Assuming 6904 is really 1.35x vs. 6903, here's the ppd difference for one example quad AMD system:

822,882ppd 6903 11:45TPF 0.82day completion
858,536ppd 6904 15:52TPF 1.10day completion

One thing about these bigger work units is lost units are more painful. Though over the long term, that's a greater incentive for people to have stable systems.


The 6903 is:
Code:
Base points 28384
Final deadline: 17 days
Preferred:  10.20 days

Comparing to 6904:
Code:
Base points 39426
Final deadline: 23 days
Preferred: 13.8 days

base on my limited Math knowledge, .. it would be 1.35 time longer ... :D ...
 
Extrapolating -- I haven't seen a 6904 yet. In fact, my 4p AMD 6180 system has only been getting 26xx bigadv units the last couple days.

unless sfield is extrapolating it appears he did... I have not yet that I know of... been quite busy to check...
 
Yet to see one... had a pair of 6903s just after the 6904 announcement, and now a 6903 and 2689, can't complain...
 
2x 2689s here.

I would like to see the 6904s so we can keep up the teams PPD.
 
Wow. That 970 was not. What they had in mind, I think :)

AFAIK, that's ghetto first one we've seen. How is that ppd?
 
my i7 970 @4.5ghz just picked up one 6904 .. darn . it is SLOWWWWWWWWWWWWWWW ...

WU 6903: TFP = 40m 26s (from HFM) ---> PPD: 128909 (from here http://www.linuxforge.net/bonuscalc2.php)

WU 6904: TFP = 58m 26s (from HFM) >>>> would take over 4 DAYS to complete 1 Unit ..:eek::eek::eek::eek::eek: ---- > PPD: 121477 (from here: http://www.linuxforge.net/bonuscalc2.php)

So, regarding PPD, 6903 is better .. :D

I'm closing in on your 970 performance with my dodeca, once I slap some better cooling on i can hit 3.2ghz which should put me within 5 seconds of your 6903 tpf. It will be interesting if the true 12 core chip handles the 6904 with less variation...
 
6904 is being done at 20 min 40s TPF @ 8 steps in... also seeing lower PPD than 6903 on the quad 6174 :-/
 
Just picked one up on the SR-2 after a bunch of regular big-adv units. We'll see how it goes.
 
6904 is being done at 20 min 40s TPF @ 8 steps in... also seeing lower PPD than 6903 on the quad 6174 :-/

Same here on a quad 6166. It is close, but it is still 10K-15K lower.
 
I'm getting 37 min TPF for the first 4 frames, for 240k PPD and a 2.5 day run. This is down about 20k PPD from the 6903s. Still, lots better than just regular bigadv units.
 
I'm getting 37 min TPF for the first 4 frames, for 240k PPD and a 2.5 day run. This is down about 20k PPD from the 6903s. Still, lots better than just regular bigadv units.
I'm crunching a 2684 on my Sr2 so I don't wanna hear any complaining out of you!
 
I also just finished a string of 2684's on the SR-2 and snagged a 6904, we'll see how it does.
 
Actual difference for 6903 vs 6904 on 4P AMD @ 2.7ghz (6-7-5-20-1T)

822,882ppd 6903 11:45TPF 0.82day completion
847,827ppd 6904 16:00TPF 1.11day completion

Check if dynamic load balancing is engaging on your systems. This appears to make a 1:00 TPF improvement on this system (~75K PPD) for 6904.

Code:
[15:15:08] Completed 55000 out of 250000 steps  (22%)
[15:31:08] Completed 57500 out of 250000 steps  (23%)
[15:47:09] Completed 60000 out of 250000 steps  (24%)
[16:03:09] Completed 62500 out of 250000 steps  (25%)
[16:19:08] Completed 65000 out of 250000 steps  (26%)
[16:35:08] Completed 67500 out of 250000 steps  (27%)
[16:51:07] Completed 70000 out of 250000 steps  (28%)
 
Last edited:
hmm. interesting info sfield. These 6904s have potential to score higher than 6903, which is the way it should be if you ask me, since we're taking on more risk with a longer WU that we will lose our progress sometime during the WU.
 
6904 finished without incident. Results file is basically same size as 6903.

6904 conclusion:

Code:
        Parallel run - timing based on wallclock.

               NODE (s)   Real (s)      (%)
       Time:  96416.794  96416.794    100.0
                       1d02h46:56
               (Mnbf/s)   (GFlops)   (ns/day)  (hour/ns)
Performance:   2261.327    118.990      0.896     26.782

Thanx for Using GROMACS - Have a Nice Day

[12:08:05] DynamicWrapper: Finished Work Unit: sleep=10000
[12:08:15]
[12:08:15] Finished Work Unit:
[12:08:15] - Reading up to 121544064 from "work/wudata_08.trr": Read 121544064
[12:08:16] trr file hash check passed.
[12:08:16] - Reading up to 108833664 from "work/wudata_08.xtc": Read 108833664
[12:08:19] xtc file hash check passed.
[12:08:19] edr file hash check passed.
[12:08:19] logfile size: 208308
[12:08:19] Leaving Run
[12:08:20] - Writing 230759028 bytes of core data to disk...
[12:09:20] Done: 230758516 -> 221934015 (compressed to 3.1 percent)
[12:09:20]   ... Done.
[12:09:48] - Shutting down core
[12:09:48]
[12:09:48] Folding@home Core Shutdown: FINISHED_UNIT
[12:09:50] CoreStatus = 64 (100)
[12:09:50] Sending work to server
[12:09:50] Project: 6904 (Run 1, Clone 7, Gen 0)

Results from a 6903:
Code:
        Parallel run - timing based on wallclock.

               NODE (s)   Real (s)      (%)
       Time:  71747.717  71747.717    100.0
                       19h55:47
               (Mnbf/s)   (GFlops)   (ns/day)  (hour/ns)
Performance:   3041.837    160.001      1.204     19.930

Thanx for Using GROMACS - Have a Nice Day

[13:26:21] DynamicWrapper: Finished Work Unit: sleep=10000
[13:26:31]
[13:26:31] Finished Work Unit:
[13:26:31] - Reading up to 121622496 from "work/wudata_06.trr": Read 121622496
[13:26:32] trr file hash check passed.
[13:26:32] - Reading up to 108763764 from "work/wudata_06.xtc": Read 108763764
[13:26:35] xtc file hash check passed.
[13:26:35] edr file hash check passed.
[13:26:35] logfile size: 206562
[13:26:35] Leaving Run
[13:26:36] - Writing 230765814 bytes of core data to disk...
[13:27:38] Done: 230765302 -> 222418816 (compressed to 3.3 percent)
[13:27:38]   ... Done.
[13:28:08] - Shutting down core
[13:28:08]
[13:28:08] Folding@home Core Shutdown: FINISHED_UNIT
[13:28:11] CoreStatus = 64 (100)
[13:28:11] Sending work to server
[13:28:11] Project: 6903 (Run 7, Clone 18, Gen 3)
 
Back
Top