Page 1 of 1

One for the books-5 hour 11 minute TPF (P10742 R0 C1680 G9)

Posted: Tue Nov 20, 2012 8:16 pm
by GreyWhiskers
Hardware: HP a475c Pent 4/HT 3.2 GHZ (single core - two threads) with GT430GPU / WinXP / FAH 7.2.9

Current progress. HFM is projecting 21 days 16 hours work unit duration, with a Quick Return Bonus (!!??) leading to a total ppd of 133, on a total credit of 2884.2. :roll: :lol:
Project ID: 10742
Core: GRO-A4
Credit: 1813
Frames: 100


Name: HP a475c Slot 00
Path: Bud-s-hpa475c-36330
Number of Frames Observed: 1

Min. Time / Frame : 05:11:31 - 133.3 PPD
Avg. Time / Frame : 05:11:31 - 133.3 PPD
Cur. Time / Frame : 05:11:31 - 133.2 PPD
R3F. Time / Frame : 05:11:31 - 133.2 PPD
All Time / Frame : 05:11:31 - 133.2 PPD
Eff. Time / Frame : 06:32:50 - 94.2 PPD
I've run a Uniprocessor client for about two years on the HP, secondary to the lion's share of points that the GPU produces. Over hundreds of WUs on the uniprocessor, I typically get ~130 to 150 ppd (pretty pathetic - but there it is).

I just picked up one for the books today. P10742 R0 C1680 G9.

I looked in on the rig this morning, checking both the FAH Control GUI and HFM. Both were showing that the client was at about 3 hours into the work unit, but neither showed any progress at all - zero percent. Even the FAH Control GUI, which typically shows incremental progress in hundredths of a percent was showing zero. i checked Windows Task Manager, and the task was humming along at 48-50% (remember this is a single core Hyperthreaded CPU - so each thread is 50% of the CPU). This is full utilization for the uniprocessor core.

A couple of hours later, I see that both HFM and the FAH Control GUI are reporting 1%, and that the Control GUI has started clocking over the incremental hundredths of percents, once it is over 1%. (it takes over 3 minutes to increment this counter by 0.01%!!)

I checked back at other long units this uniprocessor successfully folded, and found others - close but not THIS slow:

Aug 5 to Aug 24 2012. Project 10732 TPF 4 hours 35 minutes for 165.1 ppd
Aug 24 to Sept 12 2012. Project 10732 TPF 4 hours 58 min 57 seconds for 150.7 ppd

The good news is that the V7 client protects the work units across shutdowns and restarts. And that the HP is sufficiently stable it can successfully complete these weeks-long work units. And, now that I am folding with an Nvidia GT430 GPU (getting 3,200 ppd for "ordinary" Core 15 work units), they can coexist very well.

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Wed Nov 21, 2012 2:05 am
by alancabler
Howdy Old Man,
You picked up a dandy- P10742 really is that big.

Here's a link to the project description.

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Wed Nov 21, 2012 2:18 am
by bruce
GreyWhiskers wrote:HFM is projecting 21 days 16 hours work unit duration, with a Quick Return Bonus (!!??) leading to a total ppd of 133, on a total credit of 2884.2. :roll: :lol:
The single-core P4 (with HT) is one of the slowest CPUs that still are used for folding. Nevertheless, P10742 has a timeout (Preferred Deadline) of 34.00 days and a (Final) Deadline of 73.00 days -- both of which should be easily achieved if you fold 24x7.

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Wed Nov 21, 2012 5:17 am
by Joe_H
I remember doing WU's that took 10 days or more, takes me back quite a few years. I recall that working out to about 25-30 PPD on that machine half a dozen years ago.

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Wed Nov 21, 2012 5:35 am
by bruce
For many years, the [email protected] was the official benchmark for the uniprocessor projects. It earned exactly 110 PPD (and still does for non-bonus WUs). Scaling up for 3.2 GHz you should be getting 126 PPD or more with a bonus. Yes, and the GPU works well in this environment.

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Fri Nov 30, 2012 11:18 pm
by GreyWhiskers
Well, its settled down to just under 5 hours TPF - projecting about 20.5 days to complete.

Obviously, the lion's share of the points on this old computer are coming from the GT430 96-CUDA-Core Fermi GPU - close to 3500 ppd on regular Core 15 work units. But, the uniprocessor still has something to add - well within the deadlines.
Project ID: 10742
Core: GRO-A4
Credit: 1813
Frames: 100

Name: HP a475c Slot 00 (Pentium 4/HT @ 3.2 GHz - single core - two HT Threads - Win XP - FAH 7.2.9 - GT430 running in background)
Number of Frames Observed: 44

Min. Time / Frame : 03:59:13 - 198.1 PPD
Avg. Time / Frame : 04:44:38 - 152.7 PPD
Cur. Time / Frame : 04:58:50 - 143.2 PPD
R3F. Time / Frame : 04:56:44 - 144.5 PPD
All Time / Frame : 05:02:29 - 141.1 PPD
Eff. Time / Frame : 04:48:11 - 149.8 PPD

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Sat Dec 01, 2012 12:08 am
by P5-133XL
bruce wrote:For many years, the [email protected] was the official benchmark for the uniprocessor projects. It earned exactly 110 PPD (and still does for non-bonus WUs). Scaling up for 3.2 GHz you should be getting 126 PPD or more with a bonus. Yes, and the GPU works well in this environment.
Actually, I found that if I SMP/UNI fold with a GTX 460 on a HT 3.2GHz P4 that my PPD on the GPU decreases a lot more than the increase from the SMP/UNI. I do not CPU fold on that machine for that reason.

Other people mileage can vary, so if you feel in a similar situation free to test. It can't hurt...

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Tue Dec 11, 2012 9:41 am
by GreyWhiskers
Well, patience has been rewarded. I'm in my last frame and HFM.net data is below. I wonder what it will get next? I'll find out when I wake up in the morning. Three weeks on one WU is really tiring!!

BTW, the benchmark viewer is only showing 84 frames. I am pulling this from my laptop in another room - and there must have been times over the last few weeks when the laptop didn't observe the frame turnover on the old uniprocessor. (HFM.net doesn't parse the logs - it gets the client info in real time via the API - so if it doesn't see an event real time, it apparently doesn't count it in the benchmark viewer stats).

Downloaded: 20 days 20 hours 10 minutes ago
ETA: 4 hours 29 minutes
est completion time: 21 days 0 hours 39 minutes :lol:
est 2925 points
Project ID: 10742
Core: GRO-A4
Credit: 1813

Name: HP a475c Slot 00
Number of Frames Observed: 84

Min. Time / Frame : 03:59:13 - 198.1 PPD
Avg. Time / Frame : 04:56:55 - 143.3 PPD
Cur. Time / Frame : 04:29:44 - 156.2 PPD
R3F. Time / Frame : 04:30:02 - 156.0 PPD
All Time / Frame : 05:13:10 - 134.4 PPD
Eff. Time / Frame : 05:03:06 - 138.9 PPD

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Tue Dec 11, 2012 11:29 am
by compdewd
I think I had one of these on one of my Pentium 4 non-hyper-threaded machines and I think it was going to complete something like 7 days past the final deadline. I couldn't let it do that anymore. It had to be pulled.

Re: One for the books-5 hour 11 minute TPF (P10742 R0 C1680

Posted: Tue Dec 11, 2012 8:22 pm
by bruce
I'm not sure I see a problem. 5h11m works out to 21.6 days. The timeout (preferred deadline) is 34.00 days and the final deadline is 73.00. You'll complete the WU with 12.4 days still remaining before the timeout.

Suppose the PG replaced that WU with WUs that run in one tenth the time and earn one tenth the points and adjusted the deadlines to one tenth (and created ten times as many Gens). Nothing would really change except that you'd add the extra overhead of connecting to the servers nine more times during the 21 days it would take to complete the same work. You'd earn (very slightly) less points for the same work because of the extra overhead.