Windows SMP frame times....post them here.

3800 x2 @2.5 ghz

Project: 2610 (Run 0, Clone 50, Gen 0)
[12:26:44] Completed 300000 out of 500000 steps (60 percent)
[12:58:55] Writing local files
[12:58:55] Completed 305000 out of 500000 steps (61 percent)
[13:31:09] Writing local files
[13:31:09] Completed 310000 out of 500000 steps (62 percent)
[14:03:16] Writing local files
[14:03:16] Completed 315000 out of 500000 steps (63 percent)
[14:35:24] Writing local files
[14:35:25] Completed 320000 out of 500000 steps (64 percent)
[15:07:35] Writing local files
[15:07:36] Completed 325000 out of 500000 steps (65 percent)
[15:39:42] Writing local files
[15:39:43] Completed 330000 out of 500000 steps (66 percent)
[16:11:52] Writing local files
[16:11:52] Completed 335000 out of 500000 steps (67 percent)
[16:43:59] Writing local files
[16:43:59] Completed 340000 out of 500000 steps (68 percent)
[17:17:47] Writing local files
[17:17:47] Completed 345000 out of 500000 steps (69 percent)
[17:49:57] Writing local files
[17:49:57] Completed 350000 out of 500000 steps (70 percent)
[18:22:57] Writing local files
[18:22:57] Completed 355000 out of 500000 steps (71 percent)
[18:55:02] Writing local files
[18:55:02] Completed 360000 out of 500000 steps (72 percent)
[19:27:04] Writing local files
[19:27:05] Completed 365000 out of 500000 steps (73 percent)
[19:59:10] Writing local files
[19:59:10] Completed 370000 out of 500000 steps (74 percent)
[20:31:13] Writing local files
[20:31:13] Completed 375000 out of 500000 steps (75 percent)

32 minutes a frame.

edit -
ok marty - you're right. So that's 685 PPD. I was getting about 100 more PPD using VMware. But out of the 3 projects out there, 2610 seems to be getting the worst ppd, according to the stanford forums
 
lol, lazy bitches us all for not figuring out the PPD on our boxes...

:D:D:D:D:D:D:D:D

I vote we make BillR do it!
 
marty9876 said:
lol, lazy bitches us all for not figuring out the PPD on our boxes...

:D:D:D:D:D:D:D:D

I vote we make BillR do it!
Click to expand...

That's ok marty...I'm doing it. So I know they are right that way. ;)

Plus, the spreadsheet is almost done.
 
Must....buy....quad....core....argh!


Here's my lappy,

t2500 @ 2.0Ghz

[18:51:10] Project: 2610 (Run 0, Clone 324, Gen 0)
[18:51:10]
[18:51:17] Entering M.D.
[18:51:24] Calling FAH init
[18:51:28] Read topology
[18:51:28] (Starting from checkpoint)
[18:51:28] Read checkpoint
[18:51:29] Protein: Protein
[18:51:29] Writing local files
[18:51:32] Extra SSE boost OK.
[18:51:36] Writing local files
[18:51:36] Completed 0 out of 500000 steps (0 percent)
[19:36:13] Writing local files
[19:36:13] Completed 5000 out of 500000 steps (1 percent)
[20:05:29] Writing local files
[20:05:29] Completed 10000 out of 500000 steps (2 percent)
[20:34:46] Writing local files
[20:34:46] Completed 15000 out of 500000 steps (3 percent)
[21:04:02] Writing local files
[21:04:03] Completed 20000 out of 500000 steps (4 percent)
[21:33:19] Writing local files
[21:33:19] Completed 25000 out of 500000 steps (5 percent)
[22:02:34] Writing local files
[22:02:34] Completed 30000 out of 500000 steps (6 percent)
[22:31:50] Writing local files
[22:31:50] Completed 35000 out of 500000 steps (7 percent)
[23:01:05] Writing local files
[23:01:06] Completed 40000 out of 500000 steps (8 percent)
[23:30:22] Writing local files
[23:30:22] Completed 45000 out of 500000 steps (9 percent)
[23:59:38] Writing local files
[23:59:38] Completed 50000 out of 500000 steps (10 percent)
Click to expand...



 
Another one for my T2500, ignore the first two, it was then I realized I was running the smp client plus two normal clients at the same time.

Code: 
[22:03:01] Project: 2652 (Run 0, Clone 297, Gen 0)
[22:03:01] 
[22:03:01] 297, Gen 0)
[22:03:01] 
[22:03:11] 
[22:03:11] ntering M.D.
[22:03:20] ckpoint
[22:03:42] ing local files
[22:04:03]  Writing local files
[22:04:04] E boost OK.
[22:04:04] ost OK.
[22:04:05] Writing local files
[22:04:05] Completed 0 out of 1000000 steps  (0 percent)
[23:53:59] Writing local files
[23:53:59] Completed 10000 out of 1000000 steps  (1 percent)
[01:14:28] Writing local files
[01:14:28] Completed 20000 out of 1000000 steps  (2 percent)
[01:35:42] Writing local files
[01:35:42] Completed 30000 out of 1000000 steps  (3 percent)
[01:57:00] Writing local files
[01:57:00] Completed 40000 out of 1000000 steps  (4 percent)
[02:18:39] Writing local files
[02:18:39] Completed 50000 out of 1000000 steps  (5 percent)
[02:40:29] Writing local files
[02:40:29] Completed 60000 out of 1000000 steps  (6 percent)
[03:02:25] Writing local files
[03:02:25] Completed 70000 out of 1000000 steps  (7 percent)
[03:24:26] Writing local files
[03:24:27] Completed 80000 out of 1000000 steps  (8 percent)
[03:46:31] Writing local files
[03:46:31] Completed 90000 out of 1000000 steps  (9 percent)
[04:08:34] Writing local files
[04:08:35] Completed 100000 out of 1000000 steps  (10 percent)
 
astyler said:
So from personal experience I'd say the Windows SMP client is better PPD than two single threads, but not as good PPD as running Linux64 SMP in VMWare. Anyone else agree/disagree?
Click to expand...

I only ran vmware for 2 1/2 days (2 WU) but comparing times, the windows smp is only 30 sec/frame slower than in vmware. Not to mention it's only using 230 - 250mb compared to ~900mb plus the ability to actually use my computer. Vmware totally hogged the box. Even simple java games were a slideshow and ripping dvd's took 1 1/2 hrs compared to 15 - 20 minutes.
 
Killer[MoB];1030796664 said:
That's ok marty...I'm doing it. So I know they are right that way. ;)

Plus, the spreadsheet is almost done.
Click to expand...

Is the spreadsheet calculating PpD accurately? I noticed a few math errors on the front page the other day. The other opty 170 had longer time/frame but was producing more points.......no biggie just thought I would point it out.

Thanks for putting everything together so we can all compare notes.

*Edit....Looks like the numbers are fixed already.

 
SpoogeMonkey said:
Is the spreadsheet calculating PpD accurately? I noticed a few math errors on the front page the other day. The other opty 170 had longer time/frame but was producing more points.......no biggie just thought I would point it out.

Thanks for putting everything together so we can all compare notes.

*Edit....Looks like the numbers are fixed already.
Click to expand...

Yes, it's working fine. I have been guilty of copying and pasting and forgetting to change one part or another. I think I have fixed everything I caught. That first post is getting mighty long especially in edit mode with all the tags.
 
And my results from the horrendously slow 2610.

Code: 
[18:44:51] Project: 2610 (Run 1, Clone 323, Gen 0)
[18:44:51] 
[18:44:55] Entering M.D.
[18:45:01] Rejecting checkpoint
[18:45:02] Protein: Protein
[18:45:02] Writing local files
[18:45:03] Extra SSE boost OK.
[18:45:04] Writing local files
[18:45:04] Completed 0 out of 500000 steps  (0 percent)
[19:10:03] Writing local files
[19:10:03] Completed 5000 out of 500000 steps  (1 percent)
[19:35:02] Writing local files
[19:35:02] Completed 10000 out of 500000 steps  (2 percent)
[20:00:00] Writing local files
[20:00:00] Completed 15000 out of 500000 steps  (3 percent)
[20:24:59] Writing local files
[20:24:59] Completed 20000 out of 500000 steps  (4 percent)
[20:49:56] Writing local files
[20:49:56] Completed 25000 out of 500000 steps  (5 percent)
~25 minutes a frame. Ouch.


 
Second set:
P2651 (1760) @ 13:28/f = 1881.9 PPD ( 588 PPDpGHz)
Third set:
P2610 (1523) @ 17:17/f = 1268.9 PPD ( 396.5 PPDpGHz) (Wow, what a point drop!)
 
I'm getting about 18-19 minutes a frame on my E6700 o/ced to 3.16GHz. This is while surfing and watching A Scanner Darkly, so I'm guessing I'd do a bit better if I just let it sit.
 
marty9876 said:
I think he meant the first two frames not the first two donor postings on results.
Click to expand...

Yeah, I eventually noticed what he said. What I meant is I'm not running another client but it still has huge frame times. According to the log, I've been working on this protein since 16:09:47 two days ago until 14:54:33 (right now) and just finished 95%. That's 70 hours for 95% of 1523 points, or 20 points per hour, or 480 PPD. That's about 45 minutes per percent, or twice what Sgraffite is getting.

New ram is getting ordered today.:D
 
Running on a dual Xeon 2.4 HT enabled with 4GB of RAM. Had to reboot it this morning.... Looks to about 45 minutes per frame. I'll see how this runs on the new quad core 2.33 xeon when it comes in next week. ;)

Snipped useless info...
.............
4 cores detected
[11:15:56] Project: 2651 (Run 0, Clone 193, Gen 0)
[11:16:07] Completed 86468 out of 500000 steps (17 percent)
[11:16:10] Extra SSE boost OK.
[11:49:15] Writing local files
[11:49:16] Completed 90000 out of 500000 steps (18 percent)
[12:36:01] Writing local files
[12:36:01] Completed 95000 out of 500000 steps (19 percent)
[13:22:49] Writing local files
[13:22:50] Completed 100000 out of 500000 steps (20 percent)
[14:09:37] Writing local files
[14:09:37] Completed 105000 out of 500000 steps (21 percent)
[14:56:28] Writing local files
[14:56:28] Completed 110000 out of 500000 steps (22 percent)
[15:43:12] Writing local files
[15:43:12] Completed 115000 out of 500000 steps (23 percent)
[16:29:55] Writing local files
[16:29:56] Completed 120000 out of 500000 steps (24 percent)
[17:16:41] Writing local files
[17:16:42] Completed 125000 out of 500000 steps (25 percent)
 
That's screwed... 563 PPD off that box and your going to miss the perfered deadline by a ton-- just barely make the final.

It's a fast box too (IMO), nothign wrong there it's just these SMP deadlines are dumb as moose...:p

This box doing much else?








MixManSC said:
Running on a dual Xeon 2.4 HT enabled with 4GB of RAM. Had to reboot it this morning.... Looks to about 45 minutes per frame. I'll see how this runs on the new quad core 2.33 xeon when it comes in next week. ;)

Snipped useless info...
.............
4 cores detected
[11:15:56] Project: 2651 (Run 0, Clone 193, Gen 0)
[11:16:07] Completed 86468 out of 500000 steps (17 percent)
[11:16:10] Extra SSE boost OK.
[11:49:15] Writing local files
[11:49:16] Completed 90000 out of 500000 steps (18 percent)
[12:36:01] Writing local files
[12:36:01] Completed 95000 out of 500000 steps (19 percent)
[13:22:49] Writing local files
[13:22:50] Completed 100000 out of 500000 steps (20 percent)
[14:09:37] Writing local files
[14:09:37] Completed 105000 out of 500000 steps (21 percent)
[14:56:28] Writing local files
[14:56:28] Completed 110000 out of 500000 steps (22 percent)
[15:43:12] Writing local files
[15:43:12] Completed 115000 out of 500000 steps (23 percent)
[16:29:55] Writing local files
[16:29:56] Completed 120000 out of 500000 steps (24 percent)
[17:16:41] Writing local files
[17:16:42] Completed 125000 out of 500000 steps (25 percent)
Click to expand...
 
marty9876 said:
That's screwed... 563 PPD off that box and your going to miss the perfered deadline by a ton-- just barely make the final.

It's a fast box too (IMO), nothign wrong there it's just these SMP deadlines are dumb as moose...:p

This box doing much else?
Click to expand...

Actually it was shut down for a day so it's far behind but not really doing anything else or very little. It's a domain controller and is only running print services right now and even that's not being utilized. I just looked at task manager and it's showing 4 threads at 25% CPU useage per thread. That seem right? Or should I change something?
 
MixManSC said:
Actually it was shut down for a day so it's far behind but not really doing anything else or very little. It's a domain controller and is only running print services right now and even that's not being utilized. I just looked at task manager and it's showing 4 threads at 25% CPU useage per thread. That seem right? Or should I change something?
Click to expand...

Odds are it's right. I think you need a 3.0 GZh Dullie Xeon to make the perfered dealines.
 
marty9876 said:
Odds are it's right. I think you need a 3.0 GZh Dullie Xeon to make the perfered dealines.
Click to expand...

With Nocona or Irwindale Xeons, yes. Woodcrest or Clovertown or newer run at lower clock speeds but are significantly faster.
 
unhappy_mage said:
With Nocona or Irwindale Xeons, yes. Woodcrest or Clovertown or newer run at lower clock speeds but are significantly faster.
Click to expand...

Nah, I think somethings off on that box. My prestonia's at 2.8 do 25 min a frame on 2610, probably around 30 min a frame for 2651. For only being 400 mhz less, 45 min a frame seems pretty bad.
 
unhappy_mage said:
With Nocona or Irwindale Xeons, yes. Woodcrest or Clovertown or newer run at lower clock speeds but are significantly faster.
Click to expand...

I know a few other boxes, with Nocona/Irwindale which really cut things close on deadlines. I'm just thinking Stanford might end up short changing themselves some by cutting all the dual P3's/1.6 Xeon dullies and such. Obviously SMP is a better way to run those boxes for science part of things, obviously a Woodcrest runs way faster but why not get the most out of every box.

Timeless SMP FTW!!!!! :)

It's in beta, only a few weeks old who knows what the future might bring.
 
spreadsheet will be available tonight. it's done except for the last few entires. :cool:
Well worth the wait IMHO. ;)
 
T5600 (2 MB shared cache) @ 1.83GHz
P2651 (1760) @ 23:58/f = 1057 PPD (578 PPDpGHz)
 
3800+ X2 (2x512 cache) @ 2.45GHz
P2610 (1523) @ 43:50/f = 500 PPD (204 PPDpGHz)



Both of these (this post and previous) based on 1 frame not an average. Better numbers when they are available. I'm getting into the game late.....



 
Got my first WU that wasn't a 1760 pointer.

2610......Opty 165 @ 2.9 2gb ram

Code: 
[11:42:43] Project: 2610 (Run 0, Clone 33, Gen 1)
[11:42:43] 
[11:42:44] Assembly optimizations on if available.
[11:42:44] Entering M.D.
[11:42:51] Protein: Protein
[11:42:51] Writing local files
[11:42:52] Extra SSE boost OK.
[11:42:53] ps  (0 percent)
[12:06:29] Writing local files
[12:06:29] Completed 5000 out of 500000 steps  (1 percent)
[12:30:05] Writing local files
[12:30:05] Completed 10000 out of 500000 steps  (2 percent)
[12:53:42] Writing local files
[12:53:43] Completed 15000 out of 500000 steps  (3 percent)
[13:17:19] Writing local files
[13:17:19] Completed 20000 out of 500000 steps  (4 percent)
[13:40:57] Writing local files
[13:40:57] Completed 25000 out of 500000 steps  (5 percent)

About 23:30 / frame 933 PPD
 
Looks good. Now if you feel like alphabetising, organizing by cpu model, and cross referencing WU..........ahh, J/K :D
 
Good job on the spreadsheet :) If you want to take it a step further, I know of a javascript file that will make all of the headers on the table links. This wil allow you to click any of the column headers and sort the entire table by any column you wish. Very convenient and useful. Since it is javascript, it is all done client side in your browser, so it doesn't affect the server load in any way.
 
You wanna see long frame times?
smp-sad.png

14:04:03 for one percent :( Something made explorer peg the CPU (both cores!) overnight and FAH gave up its timeslice like it's supposed to. So, it took way, way too long for that frame. I hope I make the deadline...
 
^^ and they call me the crapper queen... :D

I've got a dual P3 slot 1 550 box just dieing for some SMP luvin!!!
 
marty9876 said:
^^ and they call me the crapper queen... :D
Click to expand...
This isn't a crapper - board and chip for this sucker were around $170 each when I bought them. It just acts like one :(

A timeless SMP client for all the dual p3 boxen out there would be really nice. Especially if it scaled itself to however many chips you have...
 
Man this thing took forever to get a workunit :\

i am running it on a AM2 3800+ Windsor running at 2 Ghz

Code: 
[01:23:24] Writing local files
[01:23:25] Completed 115000 out of 500000 steps  (23 percent)
[03:01:15] Writing local files
[03:01:15] Completed 120000 out of 500000 steps  (24 percent)
[04:23:02] Writing local files
[04:23:02] Completed 125000 out of 500000 steps  (25 percent)
[05:45:06] Writing local files
[05:45:06] Completed 130000 out of 500000 steps  (26 percent)
[07:06:58] Writing local files
[07:06:59] Completed 135000 out of 500000 steps  (27 percent)
[08:28:58] Writing local files
[08:28:58] Completed 140000 out of 500000 steps  (28 percent)
[09:51:01] Writing local files
[09:51:02] Completed 145000 out of 500000 steps  (29 percent)
[11:12:52] Writing local files
[11:12:52] Completed 150000 out of 500000 steps  (30 percent)
[12:19:20] Writing local files
[12:19:21] Completed 155000 out of 500000 steps  (31 percent)
[13:01:17] Writing local files
[13:01:18] Completed 160000 out of 500000 steps  (32 percent)
 
Hi there.

Looking at your chart and there appear to be some significant inconsistencies:

CPU GHz: Shouldn't that be total GHz? Otherwise you're throwing your PPD/GHz numbers off.

For example, my 3GHz e6600 is effectively 6GHz (2 cores @ 3GHz each). If I had a quad running @ 3GHz, that would equal 12GHz. If you don't break it down that way, there is no way to compare different GHz machines.
 
Project: 2610 (Run 0, Clone 191, Gen 1)

[16:52:41] Completed 45000 out of 500000 steps (9 percent)
[17:11:39] Writing local files
[17:11:39] Completed 50000 out of 500000 steps (10 percent)
[17:30:25] Writing local files
[17:30:26] Completed 55000 out of 500000 steps (11 percent)

So frame time is 18:47

e6400 @ 3.2 ghz
2gb OCZ platinum 4-5-4-15 ddr2 800

Kandor
 
Lupine said:
Hi there.

Looking at your chart and there appear to be some significant inconsistencies:

CPU GHz: Shouldn't that be total GHz? Otherwise you're throwing your PPD/GHz numbers off.

For example, my 3GHz e6600 is effectively 6GHz (2 cores @ 3GHz each). If I had a quad running @ 3GHz, that would equal 12GHz. If you don't break it down that way, there is no way to compare different GHz machines.
Click to expand...

As long as you know GHz per core and the total number of cores for a machine you have all the info you need to do the calculations.

 
Lupine said:
Hi there.

Looking at your chart and there appear to be some significant inconsistencies:

CPU GHz: Shouldn't that be total GHz? Otherwise you're throwing your PPD/GHz numbers off.

For example, my 3GHz e6600 is effectively 6GHz (2 cores @ 3GHz each). If I had a quad running @ 3GHz, that would equal 12GHz. If you don't break it down that way, there is no way to compare different GHz machines.
Click to expand...

By nature, this is a SMP aware application, so there is no need to double the GHz on a dual core system.
 
mwarps said:
By nature, this is a SMP aware application, so there is no need to double the GHz on a dual core system.
Click to expand...

Yeah I agree, especially since it doesn't scale linearly from 2 cores to 4. I think it would really skew that calculation.
 
You must log in or register to reply here.
Back
Top