Say hello, QMDs...

[Max0r]

I couldn't take it anymore. I stopped and disabled my [email protected] service and decided to do this right. I cleaned up my startup and services to free up AMRAP (as much ram as possible), rebooted, downloaded 1-click for single CPU with -advmethods, edited the client.cfg, rebooted, ran the installer, and waited...

A minute later, seeing no HDD activity or network activity on the router's switch, I checked task manager... FahCore_96.exe using 127 MB of memory... w00t!

Then I checked the log file...

[09:48:15] Writing local files
[09:48:15] Extra SSE2/SSE3 boost OK.
[09:48:15] Entering QMD...
[09:48:52] System: p1912_ALA-dipeptide_umb2

w00t!

Just checked taskman again. FahCore_96 is using 317 MB... good thing I have another 104 free!

Now to wait and see how long it takes electron microscope to give me a PPD calculation... and we'll see how many PPD this 3.4 GHz Celeron badboy can do

 

[Max0r]

460

 

[eystan]

Boooyah!!!!!

 

[pscout]

Don't ya just luv them QMD's

 

[AlabamaCajun]

ALRIGHT!

Glad to see you figured the hold up. Now put some heat on those rigs and join us heading to page 1.

 

[LandShark]

Intel + QMD = RULE!!

 

[jerseytomato]

will cya on page 1!

 

[AlabamaCajun]

Amd + 600 Pt Gromy = PWN(Intel) When we can get them

 

[LandShark]

Amd + 600 Pt Gromy = PWN(Intel) When we can get them

Luckily, my one and only AMD (XP2400+) has been getting nothing but the 600pter lately...

 

[Nexus Realized]

Haven't had a 600pointer in ages, I seem to be getting 2076 all day every day.

 

[Max0r]

Can you guess what that means?

It went down to 150 after somewhere between 3% and 4% and has been that way ever since... GRR! Appearently there's something I didn't know about QMDs... and this is that they become much harder to process after reaching "WF convergence."

Here's some loggage, the fun part is at the bottom....

[09:47:54] - Ask before connecting: No
[09:47:54] - User name: Max0r (Team 32)
[09:47:54] - User ID: 488480086C34F498
[09:47:54] - Machine ID: 1
...
[09:47:55] + Benchmarking ...
[09:47:57] The benchmark result is 6988
...
[09:47:57] + Attempting to get work packet
[09:47:57] - Will indicate memory of 511 MB.
[09:47:57] - Connecting to assignment server
[09:47:57] Connecting to ttp://assign.stanford.edu:8080/
[09:47:57] Posted data.
[09:47:57] Initial: 40AB; - Successful: assigned to (171.64.122.113).
[09:47:57] + News From [email protected]: Welcome to [email protected]
[09:47:57] Loaded queue successfully.
[09:47:57] Connecting to ttp://171.64.122.113:8080/
[09:47:58] Posted data.
[09:47:58] Initial: 0000; - Receiving payload (expected size: 261369)
[09:48:00] - Downloaded at ~127 kB/s
[09:48:00] - Averaged speed for that direction ~127 kB/s
...
[09:48:15] *------------------------------*
[09:48:15] [email protected] QMD Core
[09:48:15] Version 1.04 (Apr 7, 2005)
[09:48:15]
[09:48:15] Preparing to commence simulation
[09:48:15] - Assembly optimizations manually forced on.
[09:48:15] - Not checking prior termination.
[09:48:15] - Expanded 260857 -> 826209 (decompressed 316.7 percent)
[09:48:15]
[09:48:15] Project: 1912 (Run 256, Clone 5, Gen 68)
[09:48:15]
[09:48:15] Writing local files
[09:48:15] Extra SSE2/SSE3 boost OK.
[09:48:15] Entering QMD...
[09:48:52] System: p1912_ALA-dipeptide_umb2
[09:48:52]
[09:48:52] Performing initial WF calculations
[09:48:52] - Number of total steps will change until convergence
[09:50:04] Completed 0 out of 2000 steps (0)
[10:03:46] Timered checkpoint triggered.
[10:08:12] Completed 21 out of 2021 steps (1)
[10:08:12] Writing local files
[10:23:22] Timered checkpoint triggered.
[10:25:00] Completed 41 out of 2041 steps (2)
[10:25:00] Writing local files
[10:40:06] Timered checkpoint triggered.
[10:42:40] Completed 62 out of 2062 steps (3)
[10:42:40] Writing local files
[10:52:07] Timered checkpoint triggered.
[10:52:13] WF converged, jumping to MD
[10:52:14] Verifying checksum
[10:52:15] Finished <<<<<<< ENGAGE UBER SLOWZ THRUSTERZ SULU
[10:53:26] Completed 73 out of 2073 steps (3)
[11:06:21] Completed 83 out of 2073 steps (4)

I really never suspected anything like this, but I guess that's what happens to folding newbs. Anyone got details on this?

Either way, can't undo e-beer... so why try ...

 

[ChasR]

A Celeron D produces around 95ppd/GHz on QMDs. Celeron Ds on a 800 MHz FSB produce around 125ppd/GHz. The preconvergence steps are 25% to 35% faster than the postconvergence steps so you can't look at production until convergence has occurred.

By my calculation from the snip of the log without a normal postconvergence frame, you're producing steps at a rate of 77.5 secs which is around 250ppd, a good bit lower than I'd expect, but better than 150.

 

[Max0r]

Celeron Ds on a 800 MHz FSB produce around 125ppd/GHz.

125 * 3.4 = 425, 800 fsb 200 ram. Preconvergence was 460 ppd which is close to 425, post started in the 200's and sunk to ~150. This is based on total done over the first 24 hours, which was about 34%. While I do use this as my main rig, I know I didn't use up more than 10-15% of the total CPU processing capacity over a 24 hour period. So what explains this lackluster performance?

 

[Joe Camel]

By my calculation from the snip of the log without a normal postconvergence frame, you're producing steps at a rate of 77.5 secs which is around 250ppd, a good bit lower than I'd expect, but better than 150.

ok i NEED to get this figured out once-and-for-all...

ChasR,

you dont go by EMIII's PPD calc (right?)
you take the log and calc out the time per STEP (not frame but step)
and then calc out the PPD. (right?)

how is this different from how EMIII calcs out PPD?


Max0r,

how are the temps?
if your PC is thermal throttling, than your going to see lower # (obviously, the CPU is running slower)

 

[Max0r]

how are the temps?
if your PC is thermal throttling, than your going to see lower # (obviously, the CPU is running slower)

With FAH running all the time, generally hovers around 49-53C. Peaks around 60-62C sometimes. Is this hot enough to engage throttling? How can I see for sure that throttling is occuring?

Just curious, how are you guys CPU's benching in FAH?

 

[ChasR]

ok i NEED to get this figured out once-and-for-all...

ChasR,

you dont go by EMIII's PPD calc (right?)
you take the log and calc out the time per STEP (not frame but step)
and then calc out the PPD. (right?)

how is this different from how EMIII calcs out PPD?


Max0r,

how are the temps?
if your PC is thermal throttling, than your going to see lower # (obviously, the CPU is running slower)

At convergence, the number of steps in a given WU is fixed. The lowest I've seen is 2040 steps and the most is 2237 steps. Sig rig number 1, when it is idle other than folding, produces each step within .1 second of each other. So you can pretty much take the time it takes to fold one post convergence frame in seconds, divide it by the number of steps, and then multiply by the number of steps in the WU to get seconds per WU. Divide by 86400 and you'll get days per WU. 450/days per WU gives you ppd. This will be a conservative estimate because the preconvergence steps are much faster (25-35%) than postconvergence steps.

To be honest, I don't know how EMIII calculates QMD ppd. I do know it varies with each frame, which would indicate it's using the last completed frame to compute ppd. Since the time per frame varies with the number of steps per frame, that could be the cause of EMIII varience.

 

[ChasR]

125 * 3.4 = 425, 800 fsb 200 ram. Preconvergence was 460 ppd which is close to 425, post started in the 200's and sunk to ~150. This is based on total done over the first 24 hours, which was about 34%. While I do use this as my main rig, I know I didn't use up more than 10-15% of the total CPU processing capacity over a 24 hour period. So what explains this lackluster performance?

QMD performance is drastically affected by anything you do with your rig, particularly if you're close to the memory bandwidth threshold for efficient QMD folding. Give me a snip from the log when you know the machine is idle other than folding. Or use the formula in the above post to check it out for yourself.

 

[Max0r]

QMD performance is drastically affected by anything you do with your rig, particularly if you're close to the memory bandwidth threshold for efficient QMD folding. Give me a snip from the log when you know the machine is idle other than folding. Or use the formula in the above post to check it out for yourself.

[19:37:37] Completed 1555 out of 2073 steps  (75)
[19:37:37] Writing local files 
[19:52:52] Timered checkpoint triggered.
[20:04:18] Completed 1576 out of 2073 steps  (76)
[20:04:18] Writing local files 
[20:20:13] Timered checkpoint triggered.
[20:36:04] Timered checkpoint triggered.
[20:51:59] Completed 1597 out of 2073 steps  (77)
[20:51:59] Writing local files 
[21:07:56] Timered checkpoint triggered.
[21:23:53] Timered checkpoint triggered.
[21:37:26] Completed 1617 out of 2073 steps  (78)
[21:37:26] Writing local files 
[21:53:16] Timered checkpoint triggered.
[22:09:05] Timered checkpoint triggered.
[22:24:58] Completed 1638 out of 2073 steps  (79)
[22:24:58] Writing local files 
[22:40:49] Timered checkpoint triggered.
[22:56:43] Timered checkpoint triggered.
[23:12:38] Completed 1659 out of 2073 steps  (80)
[23:12:38] Writing local files 
[23:28:29] Timered checkpoint triggered.
[23:44:18] Timered checkpoint triggered.
[00:00:09] Completed 1680 out of 2073 steps  (81)
[00:00:09] Writing local files 
[00:15:59] Timered checkpoint triggered.
[00:31:50] Timered checkpoint triggered.
[00:45:27] Completed 1700 out of 2073 steps  (82)
[00:45:27] Writing local files 
[00:54:12] - Autosending finished units...
[00:54:12] Trying to send all finished work units
[00:54:12] + No unsent completed units remaining.
[00:54:12] - Autosend completed
[01:01:15] Timered checkpoint triggered.
[01:17:03] Timered checkpoint triggered.
[01:32:52] Completed 1721 out of 2073 steps  (83)

Approx 355 minutes or 5.9 hours to complete 8 percent. Which comes to about 32.5% a day, or 146.25 points. It goes lower when I've been using the comp for a while, anywhere from 88 to 126.

Investigating throttling issue and getting better HSF asap. Even if no throttling is occuring, I don't want to be pushin past 45-50 full load.

Update: CPU has been running at 60C/140F for a while now with ThrottleWatch monitoring things. According to the program, there has been no throttling. Supposedly the throttling temp for a Celeron D is around 65C. I'm still leaning toward better HSF though.

My machine is toward the end of its first QMD and now is running at 200 ppd speed. Not toooooo shabby, but still should be better... Most of the QMD was spent in 150 territory though.

 

[ChasR]

The first frame in the log is run at a rate of 76.2 seconds per step, 246ppd. The last at 135.5 seconds per step, 138 ppd. Something is going on that's either eating cpu cycles or throttling the cpu.

 

[Max0r]

The first frame in the log is run at a rate of 76.2 seconds per step, 246ppd. The last at 135.5 seconds per step, 138 ppd. Something is going on that's either eating cpu cycles or throttling the cpu.

That was "away from desktop" time. Just folding. I suspect different parts of the QMD are faster and slower to process for this system. RAM usage changes from time to time too. Will investigate with current QMD.