Newbie Reality Check

[Tobit]

Ok, so I have been folding for two weeks as of tomorrow. As of right now, Stanford is showing me having completing 164 units for a total score of 15,418. I will be adding a quad next week and trying to determine if I am running the right combination of clients to maximize PPD.

Here is what I am running right now, anyone see any issues with the client combinations I am running at the moment?

• Jupiter - Celeron 331 clocked @ 3.4GHz - 1GB RAM - 1x Linux 5.04 console instance with normal WU sizes and -advmethods
• Io - AMD X2 5000+ @ 2.9GHz and HD3870 - 2GB RAM - GPU2 client and 1 x 6.10beta3 console instance with normal WU sizes and -advmethods
• Metis - P4 3.0 GHz - 512MB RAM - 1 x XP 5.04 console instance with normal WU sizes and -advmethods

I was thinking about trying to run another instance on the P4 because it has Hyperthreading but that machine only has 512MB of RAM so I'm not sure if it'd actually help or only make matters worse.

Anything I should change or is this about as good as it gets right now? On Io, I was running the SMP client but found I could get slightly more points by switching to the GPU2 client and the classic console. It's hard to turn away an easy 98 points every 75 minutes.

 

[digital_exhaust]

Don't run 2 clients on the P4...Stanford want's results back as fast as possible, and you won't really see much as far as a ppd increase doing so anyhow, so keep it at one.

I would think you should be running the SMP client on your 5000+ boxen.... I have it crunching away no problems on my 4400+ and it's been very reliable using notfred's headless cd for quite a while.....but if your getting higher ppd out of it the way it is... leave it be...

Other than that, I don't see any problems, but there are many here that are much more knowledgeable than I... so take it for what it's worth...

 

[Tobit]

I would think you should be running the SMP client on your 5000+ boxen....

I was thinking this too. In fact, the first client I tried on this boxen was the SMP client. I completed a 1400p unit in 24 hours and then switched to the GPU2 client. I am now getting 2000 PPD just from the GPU2 client alone (according to FahMon). Unfortunately, I need to dedicate a CPU core to it so I am apprehensive about trying to run another instance of the classic console. However, I am tempted to try it.

When the quad arrives, I am thinking of moving the 5000+ boxen to the Linux SMP client and, on the quad, running 2 x SMP clients in vmware sessions and have the HD3870 running alone on the host.

 

[digital_exhaust]

I am now getting 2000 PPD just from the GPU2 client alone (according to FahMon)

WHAT?!?!?!

Oh man.... this really burns my ass...... nothing against you of course...

I have been faithfully running the GPU1 client for several months and have yet to see much over 660ppd, and now they're dropping the damn thing.... and the pisser is I was stupid enough to buy two X1950's specifically for folding....... and now they're basically useless to me as I am not a gamer....

Oh well.... if your cranking out those type of points on the GPU client, I would definitely stick with it.....

Damn.....

 

[capreppy]

When the quad arrives, I am thinking of moving the 5000+ boxen to the Linux SMP client and, on the quad, running 2 x SMP clients in vmware sessions and have the HD3870 running alone on the host.

If running 2 SMP clients in VMWare, that'll take up all 4 cores. The GPU client really needs a FULL dedicated core to feed it data.

From a cost perspective and best bang for the buck, you would be better off leaving the GPU Client running with either Jupiter or Metis. If the 3870 is your main graphics card and is going with the Q6600, then you'll probably have to give up the GPU client.

 

[Tobit]

Oh well.... if your cranking out those type of points on the GPU client, I would definitely stick with it.....

Yeah, I am cranking out 97 to 98 points every 75 to 85 minutes. It won't last though once chewier WUs come out and they weigh these more appropriately.

 

[Tobit]

From a cost perspective and best bang for the buck, you would be better off leaving the GPU Client running with either Jupiter or Metis. If the 3870 is your main graphics card and is going with the Q6600, then you'll probably have to give up the GPU client.

Right now the 3870 is in Io - my main rig running the AMD 5000+. Jupiter is my Linux server running DHCP, tftp, and NFS. I can't really move it into Jupiter. Metis is an older Dell at my parents house and I donut think it has the power supply capability to run the 3870. 3870 needs to either stay in my 5000+ rig or move to the new quad. Probably just leave it as is in Io.

 

[APOLLO]

WHAT?!?!?!

Oh man.... this really burns my ass...... nothing against you of course...

I have been faithfully running the GPU1 client for several months and have yet to see much over 660ppd, and now they're dropping the damn thing.... and the pisser is I was stupid enough to buy two X1950's specifically for folding....... and now they're basically useless to me as I am not a gamer....

Oh well.... if your cranking out those type of points on the GPU client, I would definitely stick with it.....

Damn.....

Well, if it's any consolation I feel your pain. 2000PPD from a single GPU client is good reason to be green with envy and upset at the same time after running the older client for many months, which has been slated for obsolescence in the near future.

I'm basically in the same boat as you but only purchased a single X1950 (thank the gods). After all is said and done, I will have seen at least a half year of solid folding on it, probably more. The principal reason I wanted to do GPU folding was for the science, not the points. Second reason was to get my WU count up since this system is able to turn in 2 WUPD and still be in the 600+PPD range. Not many old systems can do that.

BTW, has Stanford announced the exact date of the project's termination?

 

[Shambler]

Well speaking of termination... will there be another release of a GPU client?

 

[Tobit]

Well speaking of termination... will there be another release of a GPU client?

Umm, yeah. It's called GPU2 and the beta was just started a couple weeks ago. That is what I am running and these guys are upset about because their cards are no longer supported.