What Client should I choose?

animebuster

Specs
Ram- 4gb -2X2gb Kingston HyperX 1066Ghz DDR2
Processor- Intel Core 2 Quad Q6600 2.4Ghz
Graphics card- Galaxy Geforce GTX 260+ 896MB DDR3

I usually leave my computer on 24/7. I use it during the day with torrents downloading during the night. I was looking through the folding@home website but I wasn't sure what version to get. Also I have a few questions.

1. If I were to play a game lets say a resource heavy game like Crysis, would Folding@Home decrease its needs to make way for the game or would I need to Pause folding@home If I wanted to play the game lag free?

2. Can I use my computer Normally when folding?

3. How much internet would this take up and is there a way to set it to download and upload at a given time? I have free internet between 4am-8am so, if it takes up enough internet it would be nice having it downloading/uploading at that time.

Thanks, I hope to start folding soon!

Snarkasm

1) Depends on the client, but that's the idea, yes. SMP clients will bow out when other resources are requested; GPU clients do as well, but in some cases, they can still introduce some measure of stutter into the graphics if you don't pause them beforehand.

2) Yes, absolutely.

3) The package sizes are miniscule - looks like mine are averaging about 600kb. You shouldn't even come close to impacting your normal network usage with the package transfers. I don't think there's any built-in scheduling, though.

With that setup, you could run an SMP with MPICH (not Deino) on the processor, and a GPU2 client on the graphics card. You could leave the SMP folding at all times, and just pause the GPU client (it'll install a toggle in the system tray so you can easily do so) when you're going to put a hurt on the GPU through gaming or whatnot.

Hope that helps.

_k

*Darn you Snark you got me while I was typing, how do you do a mod move so I am before you. Snark's post is easier to read and much more to the point.

You should run the SMP and the GPU2 client, here is the download page for both. You should set the SMP client set-up to use the -advmethods and -smp flag, plus a passkey.

Whenever you want to play a video game the GPU2 client should be paused since it uses the GPU however for most games the SMP does not need to be closed, SMP is a console client so it can not be paused only closed and you should use ctrl+c to do that(cmd window close). However, for a game like Crysis I close my SMP even though all CPU based f@h clients use only spare/idle CPU cycles. I close SMP for Shattered Horizon and Bad Company 2 as well because I have seen a performance difference in FPS when it is running verses when it is not, though in games like Red Faction, TF2, L4D2, Assassins Creed, etc it does not make a difference. I run with v-sync on and settings to max.

Yes you can use your computer normally while folding. The only thing you will need to do is pause the GPU2 client if you play real video games, flash based games you don't need to worry about, or launch apps like photoshop or video editors. But using office apps, watching movies, and surfing the net are things that you can be doing while the folding clients are up.

Not much, a couple of MB a day at max. Most Work Units you download or upload are under a MB. The last WU I finished wrote 20,552,373 bytes to the HDD on completion but you don't transmit that much.

If you are having trouble with something and need explicit instructions let us know. And welcome to f@h, hopefully Welcome to Team 93, Team Icrontic.com.

animebuster

Thanks a lot guys I'll do that. I have the newest drivers installed for my graphics card will those include the CUDA support?

_k

Yup, the CUDA toolkit has been included in the driver set back before the 180s.

animebuster

Awesome thanks guys and I joined Team 93, Icrontic.com I wonder how far my rank will climb. Rank 805 atm.

_k

You are off to a good start, your SMP2 client production will most likely triple or quadruple as soon as you get your 10 WUs that are marked with the passkey.

animebuster

Ah is that why the SMP Client does 500,000 steps when the GPU Client only does 15,000 steps? The SMP Client is at 68% atm

_k

No, this kind of falls into the realm of how fast certain parts of your computer are at doing certain types of math and how often the returned values are wrong.

First off, computers produce wrong answers to their math...granted it is a fairly small percentage. The math that computers compute is really just an approximation of the values, like most everything anyone does in math(just look in a calculus book). The GPU can do a huge number of fairly accurate computations extremely fast the major issue is that the accuracy and precision suffers slightly, thus you have industrial grade GPUs that actually calculate to a higher accuracy(more decimals) and have a much higher precision than most desktop versions. The people that run f@h understand this but they also know getting their data back quickly is really important so people's projects can meet deadlines and such, the accuracy of results is of course checked because they send thousands out of the work units.

The CPU cannot do functions as quickly as the GPU but is better at actually doing it correctly. Also the fact that you a SMP WU means you have hardware that can do larger amounts of work quick enough for it to be used. The single core CPU client WUs are much smaller than the GPU2 WUs.

Each client has an understand hardware level required to run its specific set of Work Units and return them in a certain time frame. These requirements are built around how important is the data, when is it needed by, and what is the return rate needed. Essentially the small protein strings have fewer steps than the big ones and it simply takes more time to do more steps. The size and time required to do the WUs is taken in to account when points are given for them, SMP2 units have a time bonus for how quickly you return them though. This is to catch the CPU client back up with the numbers the GPU2 can produce, points wise, and to encourage a speedy turn around instead of just looking at how to produce the highest PPD with no concern with time frames.

I think I kind of got across what I was meaning to convey but the issue is to really understand the whole system you need to know what every client does and how much work it does plus the different flags that give you different WUs. This would take several pages. And after you turn in 10 SMP2 WUs with your passkey attached you start to get the time bonus but otherwise you get regular credit so really low, 500ish points per WU.

animebuster

_k_ wrote:

No, this kind of falls into the realm of how fast certain parts of your computer are at doing certain types of math and how often the returned values are wrong.

First off, computers produce wrong answers to their math...granted it is a fairly small percentage. The math that computers compute is really just an approximation of the values, like most everything anyone does in math(just look in a calculus book). The GPU can do a huge number of fairly accurate computations extremely fast the major issue is that the accuracy and precision suffers slightly, thus you have industrial grade GPUs that actually calculate to a higher accuracy(more decimals) and have a much higher precision than most desktop versions. The people that run f@h understand this but they also know getting their data back quickly is really important so people's projects can meet deadlines and such, the accuracy of results is of course checked because they send thousands out of the work units.

The CPU cannot do functions as quickly as the GPU but is better at actually doing it correctly. Also the fact that you a SMP WU means you have hardware that can do larger amounts of work quick enough for it to be used. The single core CPU client WUs are much smaller than the GPU2 WUs.

Each client has an understand hardware level required to run its specific set of Work Units and return them in a certain time frame. These requirements are built around how important is the data, when is it needed by, and what is the return rate needed. Essentially the small protein strings have fewer steps than the big ones and it simply takes more time to do more steps. The size and time required to do the WUs is taken in to account when points are given for them, SMP2 units have a time bonus for how quickly you return them though. This is to catch the CPU client back up with the numbers the GPU2 can produce, points wise, and to encourage a speedy turn around instead of just looking at how to produce the highest PPD with no concern with time frames.

I think I kind of got across what I was meaning to convey but the issue is to really understand the whole system you need to know what every client does and how much work it does plus the different flags that give you different WUs. This would take several pages. And after you turn in 10 SMP2 WUs with your passkey attached you start to get the time bonus but otherwise you get regular credit so really low, 500ish points per WU.

Ah ok. Thank you so much for all the time/help. You've been a great help. Is there anyway to +Rep/Karma on this site?

_k

Nope, but the best thing you could do is refer a friend to the forums or just talk about f@h to other people and see if they are interested. Getting a new member for Team 93 isn't the first step or most important getting them involved with f@h is, teams are secondary.

And you are welcome, we are here to help

animebuster

two more things. Is there a way to reset the settings for My SMP Client? When I configured it I entered the wrong passkey. Can I still claim the points if I know what passkey I entered? Also how do you unlock the points with the passkey?

_k

create a shortcut for it and at the end of the text on the Target line put -configonly after the ". You can kill the client and do this to change the config file then launch normally without forcing a restart of the WU. You can also open the client.cfg file in the folding dir and change it there but the client still has to be restarted for it to take effect.

You just have to turn in 10 SMP2 units with your correct passkey attached, then you get the time bonus on the 11th. It is really a passive thing.

animebuster

Thanks. Sorry I forgot that before you already told me that you have to do 10 before you can get the bonus. I've set the passkey in both the SMP Client and the GPU Client.

animebuster

Hmm While running CoreTemp to monitor my cpu temperature I realized that only core 0 is being used. Sometimes the work seems to get split 50-50 on core 0 and 1 but other than that, Even though I've set the - smp flag only core 0 is being used. Task manager is only showing 30-40% CPU Usage. When I open the SMP Client it says 4 cores detected its just not using them.

It takes 10 Minutes per 1 %/5000 steps Is this normal for a Quad processor at 2.4Ghz?

_k

In the Windows Task manager what process is running for FahCore_XX.exe, 78 or a1 or a3 or something else?

animebuster

_k_ wrote:

In the Windows Task manager what process is running for FahCore_XX.exe, 78 or a1 or a3 or something else?

its Fahcore_b4.exe

_k

Most likely for whatever reason it queued up a single core client WU instead of a variable core one. This has happened to me several times recently when installing SMP clients and occasionally you will be given one even though you have the -smp and -advmethods flags, generally because they want to check something quickly and the people running the advanced clients get the work done faster.

There are two ways to deal with this; let it run through the WU or find the folder with the Work folder in it and delete a lot of stuff. I would say just let the b4 core process through because it is no big deal.

If you want to kick it out check in your programs dir. for the "Folding@home-Windows SMP client v1.01"(something close to this it should be apparent) folder and look in it for a Work folder and queue files. If it is not there check in your users folder>hidden folder Appdata>Roaming(if win7)>then a folding folder. Once you locate where the work folder is, with the f@h client shut down, delete that folder, all queue and unit info files, all FahLog files, and the FahCore_b4.exe. Then launch the folding client again and make sure that it is downloading an a1 or a3 core. If you are having trouble finding the app data folder create a desktop shortcut to launch client from the folder installed in your Program Files then check the properties and use the file location from the start in line to get to it.

animebuster

_k_ wrote:

Most likely for whatever reason it queued up a single core client WU instead of a variable core one. This has happened to me several times recently when installing SMP clients and occasionally you will be given one even though you have the -smp and -advmethods flags, generally because they want to check something quickly and the people running the advanced clients get the work done faster.

There are two ways to deal with this; let it run through the WU or find the folder with the Work folder in it and delete a lot of stuff. I would say just let the b4 core process through because it is no big deal.

If you want to kick it out check in your programs dir. for the "Folding@home-Windows SMP client v1.01"(something close to this it should be apparent) folder and look in it for a Work folder and queue files. If it is not there check in your users folder>hidden folder Appdata>Roaming(if win7)>then a folding folder. Once you locate where the work folder is, with the f@h client shut down, delete that folder, all queue and unit info files, all FahLog files, and the FahCore_b4.exe. Then launch the folding client again and make sure that it is downloading an a1 or a3 core. If you are having trouble finding the app data folder create a desktop shortcut to launch client from the folder installed in your Program Files then check the properties and use the file location from the start in line to get to it.

Thanks, good to see its not a big problem. I'll just be nice and let it run and hopefully next time I'll get one of those a1 or a3 cores.

_k

And on the performance side, forgot to address that. It taking that long to do a single set of steps is fine. My q9450 @3.55 does a single percentage in approx. 5 minutes.

once you get everything running don't forget to download HFM.NET to monitor all of your folding clients for ppd, completion and failure rate, and you can even compare points per MHz through it. The program just gives you a lot of useful information about everything concerning the clients if you fill out all the fields in it. I just started a PPD thread and contributions are much appreciated.

animebuster

Ah ok thanks. About HFM.NET, I've tried to install it But I get this error when I try to run it after installation.

HFM.NET v0.5.1

Logging Failed to Initialize.

Exception Thrown:
Castle.MicroKernel.ComponentActivator.ComponentActivatorException: ComponentActivator: could not instantiate HFM.Forms.frmMessages ---> System.Reflection.TargetInvocationException: Exception has been thrown by the target of an invocation. ---> System.ArgumentException: Font 'Courier New' does not support style 'Regular'.
   at System.Drawing.Font.CreateNativeFont()
   at System.Drawing.Font.Initialize(FontFamily family, Single emSize, FontStyle style, GraphicsUnit unit, Byte gdiCharSet, Boolean gdiVerticalFont)
   at System.Drawing.Font..ctor(String familyName, Single emSize, FontStyle style, GraphicsUnit unit, Byte gdiCharSet)
   at HFM.Forms.frmMessages.InitializeComponent()
   --- End of inner exception stack trace ---
   at System.RuntimeMethodHandle._InvokeMethodFast(Object target, Object[] arguments, SignatureStruct& sig, MethodAttributes methodAttributes, RuntimeTypeHandle typeOwner)
   at System.Reflection.RuntimeConstructorInfo.Invoke(Object obj, BindingFlags invokeAttr, Binder binder, Object[] parameters, CultureInfo culture)
   at Castle.MicroKernel.ComponentActivator.DefaultComponentActivator.FastCreateInstance(Type implType, Object[] arguments, Type[] signature)
   at Castle.MicroKernel.ComponentActivator.DefaultComponentActivator.CreateInstance(CreationContext context, Object[] arguments, Type[] signature)
   --- End of inner exception stack trace ---
   at Castle.MicroKernel.ComponentActivator.DefaultComponentActivator.CreateInstance(CreationContext context, Object[] arguments, Type[] signature)
   at Castle.MicroKernel.ComponentActivator.DefaultComponentActivator.Instantiate(CreationContext context)
   at Castle.MicroKernel.ComponentActivator.DefaultComponentActivator.InternalCreate(CreationContext context)
   at Castle.MicroKernel.ComponentActivator.AbstractComponentActivator.Create(CreationContext context)
   at Castle.MicroKernel.Lifestyle.SingletonLifestyleManager.Resolve(CreationContext context)
   at Castle.MicroKernel.Handlers.DefaultHandler.Resolve(CreationContext context)
   at Castle.MicroKernel.DefaultKernel.ResolveComponent(IHandler handler, Type service, IDictionary additionalArguments)
   at Castle.Windsor.WindsorContainer.Resolve(Type service)
   at HFM.Framework.InstanceProvider.GetInstance[T]()
   at HFM.Program.Main(String[] args)

I have some time free tomorrow So I'll look into that error and see if I can find out whats wrong

EDIT: after I actually read the error I know whats wrong with it. Its one of those stupid font errors... I have had so many problems with windows fonts for some reason they seem to keep getting corrupt. I'll download that font and hope that it fixes it.

EDIT2:

Yep problem fixed was indeed the fonts fault.

_k

Try posting about the font issue in the general software or OS forum. Pretty sure someone can help you with it.

animebuster

Ah ok good idea. I'll do that tomorrow.

In HFM.NET, I go Clients--> add Client. For My cpu, I just put the speed of a single core right? So 2400Mhz for me. When I add the GPU do I put the Clock, Memory or shader Mhz?

Thanks

_k

Yes just a single cores frequency. GPU has always been a debate in my head since all three affect production and ATI cannot change the shader clock the GPU core clock is universal between companies but with Nvidia most people just care about the shader because changing the clock on it has the most effect on production. So shader, really your choice.

animebuster

Wow I had to take the side of my case as the cpu cores were getting 70C+! I really need to get a after market heatsink. The B4 core finished but then started again, it said something like "doing same thing with different settings"... So I stopped it and deleted the work files and now I have Finally gotten a a3 core! Seems like a biggie compared to the B4 one. This a3 has 2,000,000 steps! Seems to be getting 1% every 20 Minutes which is 20,000 Steps so its going double the speed of the single core B4.

_k

was that ppd you posted from the b4 core or a3 because you should be around 4k with an a3 WU.

animebuster

_k_ wrote:

was that ppd you posted from the b4 core or a3 because you should be around 4k with an a3 WU.

That PPD was from just now. My first multicore WU is still going, so maybe the PPD will update after my current WU is done?

_k

let it get to about 30% and then check on it, make sure that when it is calcing ppd it is only using the previous 3 frames. In preferences under options tab, top right side of that pane.

animebuster

_k_ wrote:

let it get to about 30% and then check on it, make sure that when it is calcing ppd it is only using the previous 3 frames. In preferences under options tab, top right side of that pane.

Ok its at 6% atm. in the options it shows that it is calculating PPD only using the last 3 frames. My PPD is now @2040, so Hopefully It will get better.

animebuster

animebuster wrote:

Ok its at 6% atm. in the options it shows that it is calculating PPD only using the last 3 frames. My PPD is now @2040, so Hopefully It will get better.

Its now 27% and the PPD is 2297, so the problem doesnt seem to of resolved itself.

_k

Make sure it is using 100% of your proc, then close the client and launch it again.