How to install optimal on multicore machine?

[summand]

Hi!

I have a quadcore computer with a Radeon 4870 graphics card.

When I started here, I started with the "normal" windows client. And soon I noticed that it used only 25%(aka one core) of my cpu. So I started reading and found the SMP and the GPU clients. So I followed some forum posts and created two instances. Now it uses 50% (2 cores).

What I don't understand is: Can I create a third instance? Can I install the GPU client and the SMP client? The description of the SMP client says it works only with quad core cpus, so I am reluctant to just try it out.

Thanks, Chris

[patonb]

Okay, you've messed up somewhere. Smp isd 100% of a quad, or any cpu.

You should only run 1 smp on a quad technically, and it should run at 100%

You setup 2 normal clients, which is acceptable, and sometimes easier to maintain.

Post the fahlog.txt in

Code: Select all

[/ code] tabs

As you have an ATI, you're going to hAve to setup some variables to dump it to 3-5% cpu usage, but i'm a nvidia person so I'm lost there.

[Christopher N. Lewis]

Is this going to be a dedicated folding machine? Do you intend the machine to be turned on and folding 24/7? If so, having a quad core CPU you can run the Windows SMP client. It isn't difficult to set up, but can be a little sensitive to the environment - particularly if your network connection is not good. (you're not using a dial-up connection are you?). Some people with quad core machines run 2 instances of a virtual macine, and the notfred's Linux client in each. This client runs faster than the Windows client, and so generates more PPD. But if you do this there is not much resource for doing anything else on the machine.
The SMP clients have fairly tight deadlines, so will not be suitable if the machine isn't running 24/7.

If the machine is not running all the time, and you have a quad core CPU, then you can run 4 standard clients. This should be the most trouble free configuration, from a folding point of view, but may make the machine a bit sluggish. If this is not acceptable, then run 3 instances, or fewer. The standard client work units have a long deadline, and are good at backing off when other demands are being made on the machine, so are most suitable for this scenario.

As well as the CPU client, you can also run a GPU client. You'll have to tweak some of the parameters, such as CPU resource, to get a good balance between CPU folding, GPU folding, and other resource demands. The Windows GPU client is more easy to set up in Windows XP rather than Vistsa.

HTH

[jrweiss]

I have a quadcore computer with a Radeon 4870 graphics card.

When I started here, I started with the "normal" windows client. And soon I noticed that it used only 25%(aka one core) of my cpu. So I started reading and found the SMP and the GPU clients. So I followed some forum posts and created two instances. Now it uses 50% (2 cores).

What I don't understand is: Can I create a third instance? Can I install the GPU client and the SMP client? The description of the SMP client says it works only with quad core cpus, so I am reluctant to just try it out.

You can run 1 SMP + 1 GPU client or 4 CPU + 1 GPU client. The Console client works best. The GPU client can be run as a Service in XP, and the CPU client can be run as a Service in all Windows versions. Read the Wiki on multiple installations; the concept and configuration is the same for CPU, SMP, and GPU clients: http://fahwiki.net/index.php/FAH_%26_SM ... le_Clients

Read the thread on Environment Variables to optimize the GPU client for the 4870: viewtopic.php?f=51&t=9162&start=180

While a quad core is recommended for the SMP client, a fast dual-core CPU can run it.

If you decide to replace your CPU clients with the SMP client, first stop the CPU clients and restart them with the -oneunit switch to finish the current WUs.

[MtM]

Hi!

I have a quadcore computer with a Radeon 4870 graphics card.

When I started here, I started with the "normal" windows client. And soon I noticed that it used only 25%(aka one core) of my cpu. So I started reading and found the SMP and the GPU clients. So I followed some forum posts and created two instances. Now it uses 50% (2 cores).

What I don't understand is: Can I create a third instance? Can I install the GPU client and the SMP client? The description of the SMP client says it works only with quad core cpus, so I am reluctant to just try it out.

Thanks, Chris

You can create up to four instances of the uniprocessor ( classic ) client OR one SMP client, and also one gpu client for the Ati.

The official rule is one fahcore per physical processor core, the only exception made is running a gpu client besides a cpu client since gpu clients take very litle processing time away from the cpu. So either running one smp client or four classic clients, and the gpu client is the most optimal setup, depending on your choiche. Smp can be abit more tricky to set up and keep running without intervention, while the classical client ( but use the console clients! ) is set up and forget with the built in service mode. Smp and Gpu are high performance clients, they will get you the most points attributed to your contributions.

Is this going to be a dedicated folding machine? Do you intend the machine to be turned on and folding 24/7? If so, having a quad core CPU you can run the Windows SMP client. It isn't difficult to set up, but can be a little sensitive to the environment - particularly if your network connection is not good. (you're not using a dial-up connection are you?).

Good points, smp is ment for dedicated machines mostly as the deadlines are allot tighter then classical client. The mpich problems ( only applying to 64 bit windows ) are solved by using a static ip address and or installing the ms loopback adapter.

Some people with quad core machines run 2 instances of a virtual macine, and the notfred's Linux client in each. This client runs faster than the Windows client, and so generates more PPD. But if you do this there is not much resource for doing anything else on the machine.
The SMP clients have fairly tight deadlines, so will not be suitable if the machine isn't running 24/7.

I done this to as it gave better ppd and seemed to gve a better cpu utilisation. Also vmware exposes 2 cores max normally so using dual vm's was the only way. This however violates direct Pande Group recommodations and I wouldn't advice doing it. You;ll get more points, but you will be delaying science!

If the machine is not running all the time, and you have a quad core CPU, then you can run 4 standard clients. This should be the most trouble free configuration, from a folding point of view, but may make the machine a bit sluggish. If this is not acceptable, then run 3 instances, or fewer. The standard client work units have a long deadline, and are good at backing off when other demands are being made on the machine, so are most suitable for this scenario.

As well as the CPU client, you can also run a GPU client. You'll have to tweak some of the parameters, such as CPU resource, to get a good balance between CPU folding, GPU folding, and other resource demands. The Windows GPU client is more easy to set up in Windows XP rather than Vistsa.

HTH

[summand]

Thx alot for the useful info! I understand it better now.

You are right, I did not install the SMP client. The tutorial about using multiple cpus directed me to the console client and described how to setup two instances.
I think, I will just try to install another, 3rd, instance and see how it goes.

Anyway, maybe you should consider putting this information into the FAQ? While a lot of the information is already there, I did not understand what client I should install.
Maybe a "Which client should I use?" section, listing the possible choices, would be good.

[bruce]

If the machine is not running all the time, and you have a quad core CPU, then you can run 4 standard clients. This should be the most trouble free configuration, from a folding point of view, but may make the machine a bit sluggish. If this is not acceptable, then run 3 instances, or fewer. The standard client work units have a long deadline, and are good at backing off when other demands are being made on the machine, so are most suitable for this scenario.

This isn't a forum just for Windows, Linux, or MacOS so I'm going to assume that you're running Windows.

Depending mostly on how much RAM you have, one classic client per core might cause a sluggish machine, but that's a rather rare condition. I'd recommend installing all four and see how it goes. If your other applications seem sluggish, then reconfigure one of the clients to run at 85% or 90% CPU and see if that fixes it. If it still isn't acceptable, you can obviously reduce the number on more clients or to smaller numbers, but again, that's rarely required.

[jrweiss]

Also, any sluggishness caused by the CPU clients can usually be completely overcome by reducing the CPU utilization in the F@H configuration to 95% or so (85% in a VERY slow machine like a P3). That frees up some computer cycles so an app starting up (the most common delay) can utilize them while F@H is yielding the rest.

The GPU client can cause some video slowdowns (e.g., gaming; and I see it on browser scrolling), and those cannot be easily controlled except by shutting down the GPU client.

[patonb]

^Areo helps reduce everyday lag on vista win7

[bruce]

Also, any sluggishness caused by the CPU clients can usually be completely overcome by reducing the CPU utilization in the F@H configuration to 95% or so (85% in a VERY slow machine like a P3). That frees up some computer cycles so an app starting up (the most common delay) can utilize them while F@H is yielding the rest.

Actually, it's more an issue of memory that CPU utilization. When a new application starts up, it generally needs a lot of new memory pages to be allocated and often that means waiting both for the new page to be read from disk but some older pages to be written to virtual memory. That doesn't take a lot of CPU time, but it goes a lot quicker if FAH is inactive for 5% / 10% of the time. Once the new app is loaded, the lower priority that FAH uses is sufficient to keep FAH from reclaiming any of those pages that the foreground app is now using.

[MtM]

^Areo helps reduce everyday lag on vista win7

+1 Though not treid on win7 vista definetly runs smoother with Areo enabeld I think it's a driver thing?! But then again, who disabels Areo?

@OP: Btw I forgot mentioning the environment variables which can be used to tweak cpu usage, gpu utilisation and which affect lag of any kind. I could list them but it's better if you would look through the Ati specific section of the forum as you'll find more information there then you could get through me ( as I don't use Ati myself! ).

[BrownBullhead]

I didn't see anyone mention running the .exe with the -smp command line parameter? That was my first thought before I even read the remainder of the thread.

[jrweiss]

That's already covered in the SMP installation instructions. Don't you think a new user of the high-performance clients would read them?

Additional client parameters []?

You can enter the client flags here so you don't have to type the flag each time. I strongly suggest you enter the -smp -verbosity 9 flag here to make things easier to setup. Don't enter -config or -configonly or any single-use flag in this field to avoid problems.

[BrownBullhead]

jrweiss: If you refer to other posts, it has been mentioned before that people have omitted the parameter. So, I didn't want to *assume* anything; it has happened time-and-time again that people have failed to enter that parameter. When he mentioned quad-core CPU and 25% usage, the ¼ usage automatically made me think of the missing SMP parameter.

[summand]

I have installed 4 instances now and it runs smoothly. I didn't notice any sluggishness during regular work, only when playing HD videos I noticed some "jerkyness". But after turning one instance off, that was gone too. I think, I will keep it that way and turn of one instance when I start something with high cpu usage.

About the installation:
I simply got confused because I found several different articles in internet and they all described different setup. Basically smp and setting up multiple instances. I somewhat mixed them up assuming that only with the smp client multiple cores can be used and that I had to install 2 instances and ...

The problem is that everyone basically either describes all options(which is a bit much for a beginner) or the setup for their system.