New extra large -bigadv on Beta

EngrChris said:
Not over yet. They will be useful for years to come for SMP.
Click to expand...

Yeah, but the ppd hit will be pretty severe going back to regular SMP from bigadv. I probably could add a GPU, but i'm really not into GPU folding one bit due to heat and power consumption. I'm just hoping that my Gulftown will be up to the challenge with the upcoming superbigadv as that 10 day bonus deadline scares the hell out of me.
 
The old bigadv had a short deadline to ensure fast WU turn around and now they're extending it to 10 days which I assume is still a pretty tight deadline. I'm still amazed at how extremely complex the stuff being modeled is, even when they're striping the problems down into basic downloadable WUs. A gulftown CPU can do 50 billion double precision flops every second and if you do that for a week or more for one unit that is an insane amount of complexity for what is only one tiny drop in that proteins research project. (30,000,000,000,000,000 calculations in a week on a gulftown?) I just like how technology keeps progressing. If next gen GPU's have even better double precision performance then I think it would be really powerful to have some type of "bigadv" gpu unit where 2-4 identical GPU's could work together on the same project, maybe by using each GPU to run a different "thread" in a similar way to how the SMP client is multi-threaded. The direction multi CPU bards is allowing bigadv to go is fine as well :)
 
15/20 of my machines can do the new units...
 
I just got one...
What does "Run 0, Clone 0, Gen 0" means? Am I the first one to get it?


p6903.jpg
 
These things have 2,533,797 atoms, reckon on a tpf of at least 2.5 times your usual (very rough guestimate)
 
and now I have one.... 33:48 tpf on dual hex 2.8ghz ... its not that mean of a unit...
ppd 184k

unless they meant for it to be way high.... its way out of balance...
 
Patriot said:
and now I have one.... 33:48 tpf on dual hex 2.8ghz ... its not that mean of a unit...
ppd 184k

unless they meant for it to be way high.... its way out of balance...
Click to expand...

That would put my twin 5670 @ nearly 195k PPD, pity i do not run linux yet...

Oh, Grandpa over at FF would life some tpfs....http://foldingforum.org/viewtopic.php?f=55&t=18788

On a serious note my guess is the hardware is still outpacing the software. Remember the systems that we run are not your average F@H system
 
Patriot said:
and now I have one.... 33:48 tpf on dual hex 2.8ghz ... its not that mean of a unit...
ppd 184k

unless they meant for it to be way high.... its way out of balance...
Click to expand...
dual hex at 2.8 gets 184k ppd?

Those of us with SR2s at 3.6ghz and up would see some crazy ppd. Doing some extrapolating, say a dual 3.6ghz hex gets TPF of 25 mins, that means 290k ppd.

Don't let apollo see this thread.....
 
Just had a quick thought, if these tpf's and PPD calculations are correct, we are about to start dishing out some serious pernts as a team.:cool::D
 
Nathan_P said:
Just had a quick thought, if these tpf's and PPD calculations are correct, we are about to start dishing out some serious pernts as a team.:cool::D
Click to expand...

any team that is mostly gpu would be hurt...
 
Haven't gotten one yet on the SR-2's....

Did get a "84 of all things

/shrug
 
All 5 of mine loaded up with regular bigadv - 2686s and 6901s - I guess I really can't complain about that. Serves me right for not running the bigbeta flag.
 
What an excellent weekend to bring my new SR-2 online! :D
 
So just run -bigadv or do I needsomething extra/new?
 
I just got a 6903 on my SR-2 first frame was 27:06 tfp
Dual Hex at 3.5GHz U11.04 2.6.39-0-generic

You have to have -bigbeta
 
Last edited:
may i be worthy said:
What are your 6901 frame times to compare?
Click to expand...

6901:
Code: 
[07:21:07] Project: 6901 (Run 3, Clone 12, Gen 37)
[07:21:07] 
[07:21:07] Assembly optimizations on if available.
[07:21:07] Entering M.D.
[07:21:13] Mapping NT from 24 to 24 
[07:21:15] Completed 0 out of 250000 steps  (0%)
[07:32:48] Completed 2500 out of 250000 steps  (1%)
[07:44:22] Completed 5000 out of 250000 steps  (2%)
[07:55:58] Completed 7500 out of 250000 steps  (3%)
[08:07:33] Completed 10000 out of 250000 steps  (4%)
[08:19:06] Completed 12500 out of 250000 steps  (5%)
[08:30:41] Completed 15000 out of 250000 steps  (6%)
[08:42:17] Completed 17500 out of 250000 steps  (7%)
[08:53:53] Completed 20000 out of 250000 steps  (8%)
[09:05:26] Completed 22500 out of 250000 steps  (9%)
[09:17:01] Completed 25000 out of 250000 steps  (10%)
~11:35tpf for first 4%

6903:
Code: 
[23:35:23] Project: 6903 (Run 9, Clone 0, Gen 0)
[23:35:23] 
[23:35:23] Assembly optimizations on if available.
[23:35:23] Entering M.D.
[23:35:30] Mapping NT from 24 to 24 
[23:35:34] Completed 0 out of 250000 steps  (0%)
[00:02:40] Completed 2500 out of 250000 steps  (1%)
[00:29:43] Completed 5000 out of 250000 steps  (2%)
[00:56:50] Completed 7500 out of 250000 steps  (3%)
[01:23:56] Completed 10000 out of 250000 steps  (4%)
~27:06tpf for first 4%

Edited my above post, I calculated that time poorly in my head.
 
First frame below, 48 cores @ 2.7ghz. Memory is running at ddr3-1066 (1150) 6-6-5-1T speeds currently due to troubleshooting experiments. Down from 6-7-5-1T @ 1333/1440.

So, 818K PPD (!)

For comparison, this machine does a TPF of 5:18 at 2.7ghz on 6901 units.


[02:41:12] Project: 6903 (Run 3, Clone 1, Gen 0)
[02:41:12]
[02:41:12] Entering M.D.
<snip>
:) G R O M A C S (-:
Reading file work/wudata_06.tpr, VERSION 4.5.4-dev-20110530-cc815 (single precision)
[02:41:21] Mapping NT from 48 to 48
Starting 48 threads
Making 2D domain decomposition 8 x 6 x 1
starting mdrun 'Overlay'
250000 steps, 1000.0 ps.
[02:41:26] Completed 0 out of 250000 steps (0%)
[02:53:55] Completed 2500 out of 250000 steps (1%)


Zink said:
48 core G34 frame times please
*popcorn*
Click to expand...
 
Last edited:
You must log in or register to reply here.
Back
Top