Folding Forum

Here's a new fact : since two days, Fahmon is overestimating my frame time on my two GPU2 clients. I get twice the real time per frame, hence half the usual PPD ...

Well, mine is underestimating the PPD on my GPU3 Client (GTX 260) and it is showing the Client to be paused even though it is running: (happened after the upgrade to 2.3.99.4)

On this GPU Project, I get ~8K PPD. I use the last 3 frames to calculate the PPD.

Like PantherX, I'm also getting strage PPD on a networked machine.
The machine is running both the SMP and the GPU3 clients.
The problem only occurs on the GPU client, and only on that machine.
Unlike PatherX, the bad PPD calculation doesn't seem to go away.

That GPU is currently working on P5746 (384 pts) at 2 minutes per 1% and is showing 110,592 PPD.

FYI, the bad PPD doesn't go away for me It is like that no matter what I do. Sometimes, i also get 0 PPD on certain Projects and this is only for GPU Client. The Classic and SMP2 are fine The ~8K PPD is from HFM.NET which I also use.

I guess that now the SMP2 Client is going nuts. All I did was a simple system restart and now this is what it is showing:


EDIT -> The above image is before my Client finished 1 Frame. Now my Client has finished 1 Frame and this is what it shows:

I have set it to calculate to the last 3 frames. I really hope that for a week, I have the above PPD, that would be awesome

UPDATE -> After the 2nd Frame, the PPD Calculation is back to normal for the SMP2 Client However, the issue with the GPU Client is still there

When you get a strange PPD like that, gather all of the information. Post the log showing the last three frames (if that's the method you're using) and calculate the PPD manually. What do you get. (Be sure to refresh the client so that you're sure the log information is the same as what HFM is using.)

This is what I show with the GPU2 client:
Project: 5739 (Run 3, Clone 561, Gen 1) pts=511
PPD=220,752 ETA=2 Min
Last 3 frames is selected.

Time Completed Min/% Frames/Day PPD
21:57:31 35%
22:01:29 36% 3.97 363.03 1855
22:05:46 37% 4.28 336.19 1718
22:10:28 38% 4.70 306.38 1566
22:15:16 39% 4.80 300.00 1533

Machine is networked.
Running WIN/XP.
Client 6.23 (service)
Core 11 version 1.24

I think that I have found the issue, it is the FahMon benchmark: Since the Average time is incorrect, it messes up the initial calculation (when 0 Frames are done) but when it gets data for 1 Frame, it goes back to normal. Does anyone know how to "reset" the benchmarks i.e. can I delete a particular file and FahMon will rebuild the benchmark database?

EDIT -> This is for SMP2 Client. I don't know why the GPU Clients are messed up PPD.

You can delete %APPDATA%\FahMon\benchmarks.xml and that will completely wipe the database. Alternatively you can open the xml file and delete the offending entry if you wish.

I'll take a look at this later to see what causing the PPD craziness.

I tried to deleted that file, and it's worse ... the clients are still showing as Inactive, but now, the PPD stays at 0 and no new data is entered to the benchmarks database

For me, it's also GPU clients, but I'm also seeing this on a local GPU client ...

I am seeing similar things:

Linux SMP Console Edition Folding@Home Client Version 6.29
Launch directory: /home/folding/fah
Executable: ./fah6
Arguments: -verbosity 9 -smp



Current WU:
Project : 6701
Core : SMP2 Gromacs
Frames : 100
Credit : 921
Min. Time / Frame : 2mn 42s - 56194.65 ppd
Avg. Time / Frame : 2mn 59s - 48382.46 ppd
Cur. Time / Frame : 50mn 54s - 686.54 ppd
R3F. Time / Frame : 50mn 54s - 686.54 ppd
Eff. Time / Frame : 52mn 36s - 653.52 ppd

Previous WU:
Project : 6702
Core : SMP2 Gromacs
Frames : 100
Credit : 921
Min. Time / Frame : 33s - 611218.67 ppd
Avg. Time / Frame : 33s - 611218.67 ppd
No Cur. Time / Frame
No R3F. Time / Frame
No Eff. Time / Frame
[/code]

Any updates on the status of this bug?

Is there a reason why Fahmon refuses to give me an estimate for PPD?

Check your preferences and make sure that the Web Setting tab->Project Download URL is correctly assigned to "http://fah-web.stanford.edu/psummary.html" (exclude the quotes) and then go to tools->download projects from Stanford.

I have read that some donors were having issues with GPU PPD on version 2.3.99.4 but version 2.3.399.3 worked fine for them. Maybe you can give it a try.