GPU folding is here!

[Handruin]

Nice! Glad to add one more to the team.

I agree with Mark, it doesn't look like a slot was created for the GPU. I see the CPU Core_a4 was configured and it appears to have started running from what the log shows. Are you running this all through command line or are you using the F@H GUI? If you plan to run both CPU and GPU, you'll want to free one of your cores to communicate with the GPU for better results once you get to the point where GPU folding is working on your machine.

 

[P5-133XL]

You are also going to want to turn off client-type = beta and switch to client-type = advanced. There are no Core_17 beta WU's being offered right now.

 

[timwhit]

9 seconds of log isn't much to go on but I have questions:

Are you running a 64 or 32 bit Linux kernel? X64 Kernal is required for GPU folding in Linux and you also have to have the x64 bit Linux folding client.
Next, what happens if you try to add a GPU slot (There isn't one defined in your config) using the Advanced control (FAHControl)?

I am running an x86_64 kernel. I am also using the 64 bit client.

Yes, I tried adding a GPU slot several times. I just tried again and the only thing that showed in the log is this:

05:11:36:Saving configuration to config.xml
05:11:36:<config>
05:11:36:  <!-- Folding Slot Configuration -->
05:11:36:  <power v='full'/>
05:11:36:
05:11:36:  <!-- Network -->
05:11:36:  <proxy v=':8080'/>
05:11:36:
05:11:36:  <!-- User Information -->
05:11:36:  <passkey v='********************************'/>
05:11:36:  <team v='10047'/>
05:11:36:  <user v='timwhit'/>
05:11:36:
05:11:36:  <!-- Folding Slots -->
05:11:36:  <slot id='0' type='CPU'/>
05:11:36:</config>

I left the gpu-index at -1 and didn't change any other options.

 

[timwhit]

You are also going to want to turn off client-type = beta and switch to client-type = advanced. There are no Core_17 beta WU's being offered right now.

I will change this after a WU finishes.

 

[P5-133XL]

You shouldn't wait. The setting only matters when you get your next WU and waiting till after is too late.

The system config doesn't see your video card so my next question is are you using the Nvidia proprietary drivers? They are also required for GPU folding

 

[timwhit]

Yes, I'm using the proprietary drivers. I just manually changed my config.xml and added a GPU slot. It doesn't appear to be a completely valid configuration, but it started GPU folding. Though, now CPU folding is broken.

View attachment fah.txt

 

[P5-133XL]

Manually editing the config is not normally recommended but it did get you your GPU slot.

Now you need to add the CPU slot back in. I assume you made a syntax error...

 

[timwhit]

I got the CPU slot added back in, but now the GPU slot just sits there with the message: "Paused:waiting for idle".

 

[P5-133XL]

Move the power slider from medium to full (In FAHControl).

By default (Medium power level) the GPU only folds when the computer is idle (i.e. screensaver is running). This is so you do not experience Lag when using the computer.

 

[timwhit]

Move the power slider from medium to full (In FAHControl).

By default (Medium power level) the GPU only folds when the computer is idle (i.e. screensaver is running). This is so you do not experience Lag when using the computer.

OK, good to know. This is certainly a very complicated piece of software.

 

[Handruin]

It is, but it's a lot better than it was years ago. I know that doesn't seem like much but every little bit helps especially when getting less-technical people to contribute to the project.

 

[timwhit]

I think my issue is that I couldn't find any documentation on how to manually configure config.xml. If I could simply change the configuration manually and know what the various options are it would be much easier.

Having a human-readable configuration file is pointless if I am not supposed to edit it by hand. This also goes against everything *nix.

 

[P5-133XL]

If you need/want all the possible options, you can set the verbosity = 5 and restart the client. After the client is restarted it will produce a new log and at the start it will list your current config with very possible option with its current value.

The real issue is why doesn't FAHControl modify the config properly so you have to manually edit it.

P.S. once you have all the options set verbosity back to the default 3. It is much easier to debug and diagnose with the config only showing options that have changed from their default values.

 

[timwhit]

FAHControl now seems able to modify the config. I have no idea why it works now and it didn't before. I checked earlier today and it was doing CPU and GPU folding.

Thanks for the help.

 

[P5-133XL]

I do not know if you've done it but if not, you are still likely to want to drop the number of threads used by the CPU slot from (-1 = automatic) to one less than the number of CPU cores the processor has so that you are not over-subscribing the CPU.

 

[timwhit]

I do not know if you've done it but if not, you are still likely to want to drop the number of threads used by the CPU slot from (-1 = automatic) to one less than the number of CPU cores the processor has so that you are not over-subscribing the CPU.

I set it to 3 last night before going to bed.

I did notice that my system felt pretty laggy when it was doing GPU folding with Folding Power set to Full. What setting do you guys use for that?

 

[Handruin]

I set it to 3 last night before going to bed.

I did notice that my system felt pretty laggy when it was doing GPU folding with Folding Power set to Full. What setting do you guys use for that?

I usually leave it at full/100% all the time unless I'm playing a game. My GTX 560 also had slight desktop lag when running at full. I don't notice much lag with the GTX 780 when running at full.

 

[P5-133XL]

I too leave it at full 100% of the time. However, I maintain advanced control on my screen and if I have any need I simply right-click on the gpu slot and pause it and the right-click to unpause it when the need no longer exists.

You can probably do better by moving the power-slider between full and medium and back. That way if you forget to restore it back to full, the computer will still restart GPU folding when it thinks it is idle. You will have to activate the fahscreensaver though and set the idle timer on it for that to work properly.

 

[CougTek]

With the arrival of the Radeon R9 295X2, which can consume up to 500W, I started thinking about a way to properly use all that heat somewhere at home. Then, I thought about my water heater...the heating element can need up to 4500W. Someone needs to design a Folding@Home GPU-based water heater. Sure, it would cost something like 15000$, but damn, what an awesome HPC system it would be!

 

[mubs]

The tail wags the dog.

 

[P5-133XL]

I don't think a water heater would work well for it is not always heating water, unlike the GPU's folding.

 

[ddrueding]

With the arrival of the Radeon R9 295X2, which can consume up to 500W, I started thinking about a way to properly use all that heat somewhere at home. Then, I thought about my water heater...the heating element can need up to 4500W. Someone needs to design a Folding@Home GPU-based water heater. Sure, it would cost something like 15000$, but damn, what an awesome HPC system it would be!

At home the computer rack is water cooled. During the winter the radiator is in the house, during the summer it is in the garage.

Designing a water-cooling system where the water heats to a sufficient temp to act as a water heater might lead to some system instability.

 

[timwhit]

Is 120 degrees F too hot for GPUs or CPUs?

 

[P5-133XL]

Typically neither but it is a judgement call as to what temperatures you are comfortable running at. All chips have their own limits and when they get close, they throttle the clocks down, for self protection, till the temperatures are more reasonable. GPU's typically do not start throttling till 95C-110C and CPU's 80C-100C (notebooks being the high-end of those ranges). Of course, the lower the temps the longer chips last but low temps typically require fans to run fast and loud. The problem is that fans that run fast for extended time periods have short lives and chips do not like it at all when their fan's die. It is very common that video cards die when their fan(s) fail.

There really isn't a hard and fast rule and as I said it is a judgement call. I run my CPU's so they do not exceed 65C and my GPU's 85C while continuous folding and I try not to exceed 80% max fan speed.

 

[Handruin]

P5-133XL, have you configured your F@H with any specific options? Reason I ask is it seems your PPD hasn't taken a hit like many others. It seems 60K PPD is my new norm vs 200K+ PPD in the past. Is there anything I can do differently to get a higher PPD with the same hardware?

 

[P5-133XL]

Those p8018's (Core_15) are PPD murder, so I try to optimize client type to ensure Core_17 to chase the good projects. The easiest way is to run Linux because only Core_17's are offered. Second choice, is that several internal Windows Core_17 projects just graduated to client-type = beta. So that would be my recommendations right now.

 

[timwhit]

P5-133XL, have you configured your F@H with any specific options? Reason I ask is it seems your PPD hasn't taken a hit like many others. It seems 60K PPD is my new norm vs 200K+ PPD in the past. Is there anything I can do differently to get a higher PPD with the same hardware?

You could also try the NaCl client for CPU folding. Gives me pretty decent PPD.

 

[P5-133XL]

Handruin is dealing with GPU folding issues, while NaCl is CPU only

 

[timwhit]

Handruin is dealing with GPU folding issues, while NaCl is CPU only

Yeah, I know. I was giving him a suggestion that might help his overall PPD.

I run the Linux client and I believe I get all Core_17 WUs.

 

[Handruin]

Linux isn't really an option for me for both my desktops. Sounds like client-type=beta is really my only option for GPU folding improvements. Is there any way to fake the OS type as Linux?

 

[Handruin]

I'm getting mostly p8018 with an estimated 30K PPD on my GTX 780. It used to estimate closer to 200K. :-(

 

[P5-133XL]

Linux isn't really an option for me for both my desktops. Sounds like client-type=beta is really my only option for GPU folding improvements. Is there any way to fake the OS type as Linux?

dual booting?

 

[Handruin]

dual booting?

It's something to consider. I changed to the client-type=beta so I'll see how that goes for now. Are you dual-booting your setups or are you strictly running Linux?

 

[P5-133XL]

Just Windows. Most of my folding machines won't run X64 (They're P4's) and for Linux that is required for GPU folding. Stanford never produced an x32 Core_17 variant. The X64 machines I have need to stay in Windows for they are not dedicated folding machines but have other primary functions.

 

[Handruin]

Just Windows. Most of my folding machines won't run X64 (They're P4's) and for Linux that is required for GPU folding. Stanford never produced an x32 Core_17 variant. The X64 machines I have need to stay in Windows for they are not dedicated folding machines but have other primary functions.

So...if you're not running Linux, how are you keeping your PPD as high and avoiding the p8018?

 

[P5-133XL]

I wasn't, but I am now that I switched to beta. I was operating at 112K PPD where my normal is 210K PPD. I know that my drop is less than yours but I would attribute that to me having much lower-end video cards (GTX 460's and 580's) that don't normally get nearly as much QRB as your GTX 780's. They don't get as high, so when the Core_17's leave, they don't drop nearly as much.

 

[P5-133XL]

The cause of the current problem with not getting Core_17: https://foldingforum.org/viewtopic.php?f=80&t=25887&start=105#p262342. PG tried to fix an issue with the Maxwell cards failing with Core_17 by limiting who got them. It didn't work...

PG knows about the issue and is working on a fix...

 

[Handruin]

Switching to beta seemed to have made a huge difference so far. I'm at 142K+ vs 63K. If only it will last like this.

 

[P5-133XL]

unfortunately, projects don't typically stay in beta for long so it is probably just a temporary fix but hopefully by then PG has a more permanent fix to its current problem.

 

[P5-133XL]

Currently, with client-type general release, PG is releasing p7620 and p7621 (Core_15) WU's with QRB. I'm sure it is an anomaly, but my clients are reporting 150K PPD on GTX 580 and 47K PPD for my GTX 460's where normal is 45K and 20K. Another person reported 320K PPD for a 780Ti.

I have no idea what is going to happen with these or even if the QRB points will be given, but I'm setting my clients to general release to get them while they still exist.