Should I really be getting so few PPD?

[Acorn]

I built myself my first new computer a week ago and started folding on it a couple of days ago.

The specs are: Core2Duo e8400, ATI HD4850, 4GB RAM.

All I'm getting is about 400 PPD, that's really useless right? Should I be getting more?

I've got it set up as a service running one instance on each core, each one doing 200 PPD.

Help

[toTOW]

That sounds correct with two uniprocessor clients ...

If don't mind taking more risks, you could try an SMP client or a GPU client ... you'll get more PPD, but they're still beta clients.

[Acorn]

Oh okay, well that's reassuring to hear that I'm getting normal results.

I think I might try running GPU2 and one CPU.

When I tried the GPU2 client it seemed really odd. When I went to the display it would start processing some test data and say "core not active". After a bit of that it would carry on doing the folding, and then it would stop. I'm not sure if that was just what the display was showing or if everything was fine behind the scenes. I still had both my CPU processes running, maybe that was causing problems.

How do you even get rid of one of the processes? And how do you tell the GPU2 what core to use?

Also can you not see a log of what GPU2 is doing? And can FAHmon work with it?

[toTOW]

You need to stop or disable your services before trying something else.

If you run an uniprocessor client and the GPU client, you don't have to tell which core to use ... it would work fine automatically.

To get the log on the GPU client, Rick click on the icon > Status > Log file. I you point fahmon to the folder where fahlog is, it will be able to monitor your client.

P.S : don't use the viewer, it's full of bugs

[Acorn]

Great! That's so much help!

What's the best way to get rid of one of the CPU services? Is there some way I can get it to end when it's finished the next WU? Does it matter which core I have the CPU service on?

I was having trouble before trying to get the GPU2 client to work. I eventually worked out that I had to install the hotfix drivers for the 4800.

Which do you think would be faster, dual core SMP or GPU2 and one core?

Oh, and should it really take this long to write checkpoints?

[03:09:58] Completed 100000 out of 500000 steps (20)
[03:09:58] Writing checkpoint files
[03:25:01] Writing checkpoint files
[03:27:16] Writing local files
[03:27:16] Completed 105000 out of 500000 steps (21)

[codysluder]

Each of your services was created from a specific client and when you configured it you told it to create a service. If you reconfigure that same client it will ask you if you wish to unconfigure the service. It will not finish the WU, but it's fairly simple to start the client in non-service mode and allow it to finish.

It doesn't matter which service you stop. They're identical except that they've got different MachineID settings.

It's easier to get the GPU client to run than the SMP client. Neither one works properly as a service yet.

Oh, and should it really take this long to write checkpoints?

[03:09:58] Completed 100000 out of 500000 steps (20)
[03:09:58] Writing checkpoint files
[03:25:01] Writing checkpoint files
[03:27:16] Writing local files
[03:27:16] Completed 105000 out of 500000 steps (21)

A checkpoint is written every 15 minutes between frames. One is at [03:09:58] and the next one is at [03:25:01] which is 15 minutes.

Maybe you really mean should a frame take 17m 18s? Frames (really blocks of 1%) can take almost any amount of time. It depends on the speed of your computer, of course, but it also depends on the complexity of the protein you've been assigned. That's why some projects earn few points and others earn more points.

[Acorn]

Thanks, that's really helpful!

One other thing. How does the GPU client decide which CPU core to use?

well it seems to be using my spare core so that's great

[bruce]

One other thing. How does the GPU client decide which CPU core to use?

The client detects which hardware you have and downloads the appropriate core and .dlls.

[Acorn]

Oops, I think you meant a different sort of core. I meant which core of my dual core processor, 0 or 1. I assume it detects which one already has a folding process running on it or it detects which one is under heavy use. Not that it really matters anyway seeing as it's using the right one.

Oh, and using GPU2 was a BIG success!! I'm getting almost 2000 PPD now! almost 5 times as fast as on my cpu. It's a shame that folding on the processor is so slow though

[bruce]

True. My answer is for the FahCore.

With multiple CPU-cores, the Operating System is responsible for matching up work that needs to be done with resources that can do it. In the simple case, the scheduler looks for the highest priority task that needs processing and assigns it to a processor (CPU-Core), then the next highest priority task is assigned to the next processor, etc. If a task needs to read/write data, it will wait for the I/O and the scheduler will see if there's anything else that can be assigned to that processor. The processor are essentially assigned at random, but it doesn't matter since they're "symmetric" (have the same characteristics).

Normally, processes that use lots of resources (like the FahCores that use the CPU for the heavy work) have to run at low priority so that tasks that use lots of short blocks of processing (like browsing or scanning for viruses) can continue to run at a reasonable speed. Moreover, anything that the human operator is waiting for should have a higher priority than something specifically designed to use ALL of the UNUSED resources. (The GPU core, FahCore_11, is much more like a virus scan than the other FAHcores because it's primary job is moving data, not doing the actual work.)

There are all sorts of things that can make the simple case into a more complicated case but this isn't really a good place to go into the intricate details. I'm sure you can find it on the web if you search.

One further detail, however, since it's a common discussion topic here:
Affinity
You can inform the OS that you do not want a particular task to run on certain processor. e.g.- I do not want FahCore_11 to run on any of my CPU-cores except #3. Making settings like that is an advanced topic. In some cases, it can improve performance and in others it can severely limit your throughput.

[darkcube]

I'm getting a seemingly abnormally low PPD on my computer for my CPU client.

System:
C2D E6600 @ 3.0Ghz (Dual Core)
nVidia 9800 GTX

I'm currently running the GPU2 & the CPU (not SMP) client at the same time. I'm not running the SMP client because 1 core of the CPU is used by the GPU so the CPU client really can only use one core (let me know if this is correct). The problem is while my GPU2 PPD is extremely high, the CPU PPD is extremely low. Both are constantly using 50% cpu (100% total) so the GPU client isn't stealing cpu time from the CPU client.

PPD:
CPU Client 6.20 = ~75 PPD
GPU2 Client = ~6100 PPD

Now i've seen recorded results with people running the SMP client with an E6600 and getting around 1300 PPD so why is mine so low? I'm also running the same CPU client on my laptop which has a Pentium M 2.0Ghz and that is getting ~150 PPD.

The SMP Client is also running at my office on a Q6600 getting 1800 PPD (one core is being used by GPU2 on that computer as well running a nVidia Quadro NVS 290).

The numbers for my E6600 just don't seem to add up. Any suggestions?

[John Naylor]

The standard CPU client has a much lower benchmark score than the SMP client (110PPD rather than 1760PPD). That explains why your Pentium M is getting around 150PPD, but 75 PPD is very low for one core of a Core 2 Duo. What project number is it running?

[7im]

What are you using to measure PPD? Are you doing the math with the time frames, or letting some 3rd party software do it? The readings can be deceiving from fahmon, depending on the setting you are using.

I'm not saying there isn't a problem, just that it may not be the performance. You may want to double check the PPD manually, or change your fahmon setting to use the last 3 frames.

[darkcube]

I can check the project number when I get home from work.

I'm using Fahmon to measure it, and it was set to last 3 frames, and i checked it after a long period of inactivity so computer use wasn't bringing it down.

[darkcube]

The project number was 1737.

That project has since finished and it is now running project 896 and getting about 160PPD, is that about what i should be getting on the single core client?