new to folding, what client to use? help please

[jeffrey23]

hi everyone, im new to folding and im having trouble picking the best client for my hardware that will give me the best performance.

i have a machine with an nvidia 8600gt 256mb and a q6600 quad core running at 3.4 ghz with 2 gigs of ram.

now, should i use the gpu2 client for the nvidia card or should i use the smp client to fun on the cores? or should i use some sort of combination of both? which will give me the best performance in folding?

thanks again

jeff

[toTOW]

You can use both : a GPU client for the NV board, and a SMP client for the CPU.

[jrweiss]

For new users, the current beta SMP client may be a bit frustrating. A combination of the GPU client and 3 instances of the CPU client, all running as Services, may be the cleanest, easiest, high-producing combination. Once you figure out the routine (in the Wiki) for running multiple clients, 4 is as easy as 2.

Once you think you have it all figured out, then you can try the SMP client...

[jeffrey23]

ok i went to the wiki and saw how to set up the non graphical client. so i can just install the client as per the wiki instructions and leave the gpu2 client alone? it doesn't need to be configured right? also i am confused about running them as a service, what does that mean? do i have too in order for this to work?

thanks again

jeff

[BrgHW]

Hi Jeff.

Running as a service, meens that the client starts when you starting the PC, and runs in behind as an service.
+ is that if the PC restarts you don't have to think about starting the client(s).
- is that you dont see the progres whitout a monotoring program like Fahmon, and loss of data happens from time to time.
I rekomend running the GPU in systray and the CPU clients as services. Just put link to GPU client in your Startup folder in Windows, so it starts automaticly when Windows starts.
When you are ready, go for SMP as an service if you are running XP. In Vista just run it in consoll mode with link in Startup folder. SMP use your CPU better, but also getting it much hotter. Mine Q6600 is 60-62 degrees celsius 24/7.

[jrweiss]

Yes. Once the GPU client is running, just configure the CPU clients with their individual Machine IDs (2, 3, 4).

As BrgHW noted, running as a Service puts the F@H clients out of sight, and also allows you to log off the computer for better security or to change users, without interfering with the F@H clients. Both the GPU and CPU console clients v6.20 can be run as Services, so you can configure all 4 of them that way. If a client is already running in console mode (in a window), shut it down and restart it using the -configonly switch (create a shortcut, add the switch to the end of the command line, rename it appropriately, e.g., GPU Config or CPU4 Config, and run the shortcut). When you get to the "Start Automatically (Run as a service" prompt, answer yes. Close the window when done. Repeat for all clients. Restart your computer.

FahMon is a popular monitoring tool -- small and versatile. Another guy is working on a new one -- FahProgress -- which is much smaller but not as complete. Take a look at the various monitoring tools available in the Third Party forum.

[jeffrey23]

ok so I set it all up I got the GPU and 3 instances of the console version running, however, do I have to point each instance to run on each core??

thanks again

jeff

[jrweiss]

Core affinity will be handled automatically by Windows. Just make sure you set the Priority of the GPU client to "slightly higher" in its config.

[jeffrey23]

after setting everything up in task manager i notice that 3 of my cores were only at 85-90 % (the other core is at 100% cause of the gpu using it) so i decided to run another instance of the console version , and now every core is maxed out at 100%. so im running the gpu2 and 4 instances of the console version, is this safe or should i cancel one? none of them have "slight higher in the config selected" - you posted as i was typing

jeff

[jrweiss]

The Pande Group does not recommend running more than 1 client per core. While it may physically work on your computer, the likelihood of slowdowns and future problems is much higher. Also, with your CPU and GPU saturated like that, you will likely run into performance problems with your foreground apps, as they try to get al the F@H cores to release th CPU and GPU.

I would recommend closing the 5th instance, configure it to NOT run as a service, and run it with the -oneunit switch (using a new shortcut) to finish the current WU. Then delete its directory after the current WU is finished.

You can compensate somewhat in PPD (points per day) performance by enabling LARGE work units and ADVANCED METHODS in the config for the CPU clients. You can run the -configonly shrotcut while they are running, then restart the computer or the individual Services via the Control Panel Services tab.

[jeffrey23]

thank you very much for your help. i will shut down the extra instance and just run one on each core. on an unrelated note, using fahmon it tells you your ppd but when i go check my actual point status on the web it does not show that many points, why is that? have i not uploaded all my completed WUs?

example: fahmon tells me 4000+ ppd and its been running for 12+ hours and the server only shows 1400 points, 16 completed WUs. when theoretically i should have 2000+ (half a days work according to fahmon) can anyone shed some light on this?

i know it depends on when the server gets updated but if i started yesterday at 6pm and fahmon showed 4500 ppd then today after 6pm when the server gets updated i should have close to 4500 points right?

jeff

[jrweiss]

The server knows nothing until you upload a completed WU. Then the stats get updated about every 3 hours. So, while FahMon reports what you are Folding NOW, the stats will reflect only uploaded WUs.

Since you will be assigned different sized WUs that will take varying lengths of time to complete, you cannot predict more than a day in advance how many more WUs will be completed. After a few weeks, you'll see the trends on your own participation.

[jeffrey23]

thanks for the reply, that explains things more. do completed WUs get stored in queue if they were not able to upload to the server? if yes, will theses ever be over written with new WUs?

sorry this will be my last question on this topic lol

jeff

[jrweiss]

Yes, the queues are in your individual client folders.

[anandhanju]

thanks for the reply, that explains things more. do completed WUs get stored in queue if they were not able to upload to the server? if yes, will theses ever be over written with new WUs?

To add to what jrweiss said, each queue has 10 slots and can hold 9 finished units, if for some reason you are unable to send back work, in addition to the current work unit. The FAH servers are quite decent when it comes to receiving back completed units and you shouldn't be seeing more than a couple of unsent units in your queue in the worst case. Users typically look at the server stats page (to check the server's load) or post here if they see anything unsent.

sorry this will be my last question on this topic lol
jeff

Its good to ask questions. You picked up that multi-client multi-instance setup rather quick, if I may add. Welcome to the Fold!