F@H Optimisation Question

[Able-Contention]

I was wondering if anybody had any current information about running the F@H beta clients? I am running an E8400 (stock speeds) and I installed the DEINO beta SMP client and fahmon 2.3.4 is reporting 547.41ppd for project 2484. I have a 9600GT that I have OC to 727(clk)/1817(shd)/900(mem) and I have been getting 2555 ppd on project 5903. Are these results pretty typically or am I over or under where I should be? I did some googling and set affinity as such; and the DEINO (fahcore_78.exe as reported in the client files) core is running on two cores, while the GPU (FAhcore_14.exe) is running only on the second cpu core. My daily totals have been from 2900-3300 ppd.

I am founding for 2630 and I have been slowly moving up the ladder, just wondering if there was some tweaking I could do before I trying and installing the linux client or outright buying a better video card. Yeah I have that bug now I guess...

A bit unrelated. Anybody use the new ATI drivers that have a performance boost for F@H?

[Flying Fox]

Welcome to TR and the Folding team!

I installed the DEINO beta SMP client and fahmon 2.3.4 is reporting 547.41ppd for project 2484.

Project 2484 seems to be a single core WU. Did you remember to put "-smp" in your command line? Check the FAHlog.txt file and the command line should be in there somewhere. If you forgot the -smp switch, Ctrl-C to stop work and relaunch the folding client with the "-oneunit" switch to resume work (but don't put the -smp switch in yet) and finish the current unit. Then you relaunch with the -smp switch (or you can then use the -configonly option to add -smp in the "additional client parameters" question).

[Ragnar Dan]

The only thing I'd add to Flying Fox's comments is that nVidia GPUs increase output when the shader clock is sped up. It's probably the same with ATi GPUs, though your reports about the matter would help our team learn for certain.

FF: I'm not sure if you have to wait until the current WU is ended to re-configure the folding client to add the -SMP switch. I've never tried it as far as I can remember, but what would happen is an interesting question.

[Flying Fox]

FF: I'm not sure if you have to wait until the current WU is ended to re-configure the folding client to add the -SMP switch. I've never tried it as far as I can remember, but what would happen is an interesting question.

The Folding client does not quite like it when a 2nd instance is run even with just the -configonly option (they detect files being locked/accessed or something?). I ended up needing to relaunch anyways I think. I did that a few days ago actually. So may as well just Ctrl-C out of it, get the current WU to finish with -oneunit, reconfigure and then relaunch.

[Ragnar Dan]

I didn't mean to reconfig while the client was running. Just to leave the old WU in there and change the command line parameters before it's completed. I expect it would just let it finish and then only download SMP WU's, though as mentioned I've never tried it.

[Flying Fox]

I didn't mean to reconfig while the client was running. Just to leave the old WU in there and change the command line parameters before it's completed. I expect it would just let it finish and then only download SMP WU's, though as mentioned I've never tried it.

I tried the reverse with the 6.02 LinuxSMP client with disastrous results. I was working on an SMP WU, and I needed to stop it for a while. So I Ctrl-C and do whatever I needed to do. After coming back I ran the client forgetting the -smp switch, and it complained about wrong SMP settings and such (it stopped though). I immediately Ctrl-C and relaunch with -smp but it was already too late. The previous SMP WU was trashed and it grabbed a new one and began anew.

I would have to say the new beta client with the "additional client parameters" config option is a godsend for this. I can put -smp in there and then I can vary between -oneunit or not.

[Ragnar Dan]

Hm. Nasty bugger. Well, maybe it would trash the WU, then. If so, it's not that big a loss, though it's possible to automate everything and do it the right way, I suppose, if one really wanted to be careful about things.

[HurgyMcGurgyGurg]

The only thing I'd add to Flying Fox's comments is that nVidia GPUs increase output when the shader clock is sped up. It's probably the same with ATi GPUs, though your reports about the matter would help our team learn for certain.

I've written about the ATI side of GPU optimization a few times here.

I'll confirm that ATI clients only boost when the Core clock speed is increased (No shader setting on the ATI side). In fact, I've downclocked my memory settings to the lowest CCC allows for the 4870, 475 Mhz instead of 950 Mhz, and not a single drop in output. Of course that's probably GDDR5 working its magic and I haven't gotten anyone else to confirm this for cards not using GDDR5.

Actually, decreasing the memory speed usually allows you to hit higher stable core clock speeds. I squeezed an extra 15 MHz out of the GPU (Around 100 ppd) by lowering memory speeds, which would normally give artifacts and unstable machine errors if I tried to run at that speed with the memory speeds at stock.

A bit unrelated. Anybody use the new ATI drivers that have a performance boost for F@H?

The new drivers will not actually give a performance boost yet. According the mhouston (The AMD representative that moderates the ATI Specific Folding GPU forum on the official folding forums) the improvements will only happen after a new version of the client core is released. The core is still in beta for some time.

I'll probably make a topic here to alert everyone when its released.

[Ragnar Dan]

The only thing I'd add to Flying Fox's comments is that nVidia GPUs increase output when the shader clock is sped up. It's probably the same with ATi GPUs, though your reports about the matter would help our team learn for certain.

I've written about the ATI side of GPU optimization a few times here.

I'll confirm that ATI clients only boost when the Core clock speed is increased (No shader setting on the ATI side). In fact, I've downclocked my memory settings to the lowest CCC allows for the 4870, 475 Mhz instead of 950 Mhz, and not a single drop in output. Of course that's probably GDDR5 working its magic and I haven't gotten anyone else to confirm this for cards not using GDDR5.

Actually, decreasing the memory speed usually allows you to hit higher stable core clock speeds. I squeezed an extra 15 MHz out of the GPU (Around 100 ppd) by lowering memory speeds, which would normally give artifacts and unstable machine errors if I tried to run at that speed with the memory speeds at stock.

[...]

Now that you mention it, I recall having read at least a couple of your posts. We need more people to join the team, and now that ATi GPUs are faster, and Stanford seems to be producing WUs for them at the expense of some nVidia cards' performance, the more people with ATi cards the better off we'll be.

[Able-Contention]

Thanks for the input guys. I just got back to my pc after a little vacation. Anyway. I am messing with the -smp and -deino flags but now it only loads the FahCore82.exe. It says:

Launch directory: C:\FAH
Executable: Folding@home-Win32-x86.exe
Arguments: -smp -deino -verbosity 9 -forceasm

Warning:
By using the -forceasm flag, you are overriding
safeguards in the program. If you did not intend to
do this, please restart the program without -forceasm.
If work units are not completing fully (and particularly
if your machine is overclocked), then please discontinue
use of the flag.

[01:19:27] - Ask before connecting: No
[01:19:27] - User name: Able-Contention (Team 2630)
[01:19:27] - User ID: xxxxxxxxxxxxxxx
[01:19:27] - Machine ID: 1
[01:19:27]
[01:19:27] Loaded queue successfully.
[01:19:27]
[01:19:27] - Autosending finished units... [April 2 01:19:27 UTC]
[01:19:27] + Processing work unit
[01:19:27] Trying to send all finished work units
[01:19:27] Work type a1 not eligible for variable processors
[01:19:27] + No unsent completed units remaining.
[01:19:27] Core required: FahCore_a1.exe
[01:19:27] - Autosend completed
[01:19:27] Core found.
[01:19:27] Working on queue slot 03 [April 2 01:19:27 UTC]
[01:19:27] + Working ...
[01:19:27] - Calling 'mpiexec -np 4 -channel shm -env MPICH_USE_SMP_OPTIMIZATIONS 1 -host 127.0.0.1 FahCore_a1.exe -dir work/ -suffix 03 -priority 96 -checkpoint 15 -forceasm -verbose -lifeline 4120 -version 623'

[01:19:31] CoreStatus = 63 (99)
[01:19:31] + Error starting Folding@Home core.
[01:19:36]
[01:19:36] + Processing work unit
[01:19:36] Work type a1 not eligible for variable processors
[01:19:36] Core required: FahCore_a1.exe
[01:19:36] Core found.
[01:19:36] Working on queue slot 03 [April 2 01:19:36 UTC]
[01:19:36] + Working ...
[01:19:36] - Calling 'mpiexec -np 4 -channel shm -env MPICH_USE_SMP_OPTIMIZATIONS 1 -host 127.0.0.1 FahCore_a1.exe -dir work/ -suffix 03 -priority 96 -checkpoint 15 -forceasm -verbose -lifeline 4120 -version 623'

[01:19:41] CoreStatus = 63 (99)
[01:19:41] + Error starting Folding@Home core.
[01:19:46]

Any ideas? Then it just keeps downloaded the core A1 again, and again.

[Gerbil Jedidiah]

Thanks for the input guys. I just got back to my pc after a little vacation. Anyway. I am messing with the -smp and -deino flags but now it only loads the FahCore82.exe. It says:

Launch directory: C:\FAH
Executable: Folding@home-Win32-x86.exe
Arguments: -smp -deino -verbosity 9 -forceasm

Warning:
By using the -forceasm flag, you are overriding
safeguards in the program. If you did not intend to
do this, please restart the program without -forceasm.
If work units are not completing fully (and particularly
if your machine is overclocked), then please discontinue
use of the flag.

[01:19:27] - Ask before connecting: No
[01:19:27] - User name: Able-Contention (Team 2630)
[01:19:27] - User ID: xxxxxxxxxxxxxxx
[01:19:27] - Machine ID: 1
[01:19:27]
[01:19:27] Loaded queue successfully.
[01:19:27]
[01:19:27] - Autosending finished units... [April 2 01:19:27 UTC]
[01:19:27] + Processing work unit
[01:19:27] Trying to send all finished work units
[01:19:27] Work type a1 not eligible for variable processors
[01:19:27] + No unsent completed units remaining.
[01:19:27] Core required: FahCore_a1.exe
[01:19:27] - Autosend completed
[01:19:27] Core found.
[01:19:27] Working on queue slot 03 [April 2 01:19:27 UTC]
[01:19:27] + Working ...
[01:19:27] - Calling 'mpiexec -np 4 -channel shm -env MPICH_USE_SMP_OPTIMIZATIONS 1 -host 127.0.0.1 FahCore_a1.exe -dir work/ -suffix 03 -priority 96 -checkpoint 15 -forceasm -verbose -lifeline 4120 -version 623'

[01:19:31] CoreStatus = 63 (99)
[01:19:31] + Error starting Folding@Home core.
[01:19:36]
[01:19:36] + Processing work unit
[01:19:36] Work type a1 not eligible for variable processors
[01:19:36] Core required: FahCore_a1.exe
[01:19:36] Core found.
[01:19:36] Working on queue slot 03 [April 2 01:19:36 UTC]
[01:19:36] + Working ...
[01:19:36] - Calling 'mpiexec -np 4 -channel shm -env MPICH_USE_SMP_OPTIMIZATIONS 1 -host 127.0.0.1 FahCore_a1.exe -dir work/ -suffix 03 -priority 96 -checkpoint 15 -forceasm -verbose -lifeline 4120 -version 623'

[01:19:41] CoreStatus = 63 (99)
[01:19:41] + Error starting Folding@Home core.
[01:19:46]

Any ideas? Then it just keeps downloaded the core A1 again, and again.

I've never seen the -deino flag. Are you sure that's a flag? Try running with -smp and take -deino out.

[Able-Contention]

It is apart of the new guide from stanford's WinSMPGuideDEINO. I did get it to finally work though, something with the credential store. That only took 3 hours! Now folding Project: 2653 (Run 6, Clone 150, Gen 100)

[Flying Fox]

I've never seen the -deino flag. Are you sure that's a flag? Try running with -smp and take -deino out.

Stanford uses 2 multiprocessing libraries now on the 32-bit WinSMP client: deino and mpich. The -deino is to use this mode. They are different downloads too so they are quite different.

[Able-Contention]

I just wanted to post that my PPD has jumped up as a result of this thread. Fahmon is now reporting my PPD as averaging between 3800-4300! The Deino SMP client being around 1505 PPD itself! Thanks for the help guys!

[Gerbil Jedidiah]

I just wanted to post that my PPD has jumped up as a result of this thread. Fahmon is now reporting my PPD as averaging between 3800-4300! The Deino SMP client being around 1505 PPD itself! Thanks for the help guys!

Awesome!

[lou15214]

Do you have the newest nvidia driver? I'm using 182.46(beta) for my 2 8800GT cards. seem OK so far. was using 182.20. and before that 182.08

[jeffry55]

Do you have the newest nvidia driver? I'm using 182.46(beta) for my 2 8800GT cards. seem OK so far. was using 182.20. and before that 182.08

Good point Lou! The newest NVidia drivers reduce CPU usage down to near zero which would free up more CPU cycles for your SMP client which means more points!! It's all good.

[Ragnar Dan]

Precision in such discussions is of value.

Which OS? The current WHQL driver available from nVidia is 182.50. I'm downloading it for my XP Pro boxes, but I doubt it will make much difference one way or another. I assume you are talking about a Vista OS, since that's generally been the case.

[heyal256]

I think he was referring to older versions of the nvidia driver (pre 180.xx) that dropped the cpu usage in winxp. It was discussed in this thread ( viewtopic.php?f=9&t=63642 ). From that thread it looks like you already upgraded to a version that has the reduced cpu usage for gpu folding.

[lou15214]

Yes its Vista x64 drivers.Thanks heyal, I didn't realize that there was an even newer driver update since I updated several days ago . Now I'm running 182.50 whql, will see how it works. May dl the 182.65(beta) to try later.

[Able-Contention]

Those numbers are with Nvidia driver 182.08. I checked today and its at 182.50, so I will post updated numbers later.

[mmmmmdonuts21]

I figured since we are talking about F@H optimization, I would post my question here. Here is what I am currently running:

Windows 7 Beta x64
2 Radeon 4850's
E6750 O.C. @ 3.2Ghz
CCC 9.3
F@H Core 1.24

Is there any benefit to running the console client as opposed to the system tray client (x2) (what I am currently running) for the GPU's. WIth the system tray client running both GPU's I have maxed out both of my CPU cores. I read places where you can set different intervals to make this much more optimized. Correct me if I am wrong but I thought you can only do this for the console GPU client. I guess I basically want to know if there is anything I can do to fold more (maybe getting my CPU involved) by switching to console clients instead. Mind you I have no stability problems running two system tray problems in 6 months.

Thanks

[Able-Contention]

I also run the system tray for the video card and I know that it slows down the processing while the visualization is open (which I don't do often). I do not know if the console runs faster then the system tray in the minimized state.

[heyal256]

I've never run the system tray version, but I would think that the system tray version should run at the same speed as the console version while it is minimized. The specific advantage (to me) of running the console version of the gpu client is that I have less things in my system tray, of course the disadvantage I have is that I have to stop the service if I'm going to do any gaming (instead of just closing it from the system tray). I believe that the system tray vs console issue is more for regular cpu folding because of the extra advantages (starts automatically without needing to be logged into a pc, showing the protein while folding doesn't take extra cycles away from points production, etc).