Simple F@H Questions

[geokilla]

1. Will there be a problem if I fold using just one name but for two or three teams?

2. Does the new client (6.20) use both CPUs to fold or do I have to run two seperate console clients? I'm going to run the GPU client for sure since my 9600GT will be idling there most of the time.

 

[drkicker]

Each instance of the client can only fold for one team at a time, but there is no problem with running the following:

CPUCoreInstance1: Username folding for Team1
CPUCoreInstance2: Username folding for Team2
GPUCoreInstance1: Username folding for Team3

Sorry I can't answer the second question.

 

[Markfw]

What OS are you running ? If its XP, you can only run the GPU client, or the smp client will kill your GPU PPD, and you don't want that.

 

[geokilla]

Thanks for the quick replies.

I have multiple computers that will be using the same username. I'll be folding on at least two of the five computers that I have. It's just that for simplicity's sake, I wanna keep it to one username. Thanks for the answer.

I'm running Windows XP. I don't plan on running SMP, but I was wondering of the new client would take up wahtever CPU "freesage" is left on the second core since I'll be running the GPU client and at least one console client. On the F@H site, for the 6.20 console client, under advanced options there's the following:

The following options require you to restart the client before they take effect
Disable CPU affinity lock (no/yes) [no]?

If you disable the affinity lock, the client will pick any free CPU core. Leave to no if you want the client to be locked to a CPU core.

Also, another question. Since I'm using the GPU client and the console client, do I have to add "-local"?

 

[Insidious]

Originally posted by: geokilla
Thanks for the quick replies.

I have multiple computers that will be using the same username. I'll be folding on at least two of the five computers that I have. It's just that for simplicity's sake, I wanna keep it to one username. Thanks for the answer.

I'm running Windows XP. I don't plan on running SMP, but I was wondering of the new client would take up wahtever CPU "freesage" is left on the second core since I'll be running the GPU client and at least one console client. On the F@H site, for the 6.20 console client, under advanced options there's the following:

The following options require you to restart the client before they take effect
Disable CPU affinity lock (no/yes) [no]?

If you disable the affinity lock, the client will pick any free CPU core. Leave to no if you want the client to be locked to a CPU core.

Also, another question. Since I'm using the GPU client and the console client, do I have to add "-local"?

One of my PCs is an opty 170 (dual core) and I have had no success running any client alongside the GPU client on it. I always end up with a greatly reduced total PPD compared to what the GPU can do on it's own.

I've set the GPU to the slightly higher priority setting, tried with the "Do not lock client to a single core" checked and un-checked. Tried a CLI clent, tried a SMP client (pretty well knew that wouldn't work).

Anyway, for a dual core with windowsXP, I'd love to know how to get the second core working without taking the GPU into the dirt, but I haven't been able to do it.

-Sid

 

[geokilla]

The GPU folding will be done on a Intel E7200 running at 3.7Ghz. So I shouldn't have to worry about greatly reduced total PPD due to the high CPU speeds.

 

[Insidious]

Originally posted by: geokilla
The GPU folding will be done on a Intel E7200 running at 3.7Ghz. So I shouldn't have to worry about greatly reduced total PPD due to the high CPU speeds.

If you say so. :roll:

Please don't forget to post how it goes. I need to know how to do it right on a dual core. (my methods haven't worked so far)

-Sid

 

[geokilla]

Originally posted by: Insidious

Originally posted by: geokilla
The GPU folding will be done on a Intel E7200 running at 3.7Ghz. So I shouldn't have to worry about greatly reduced total PPD due to the high CPU speeds.

If you say so. :roll:

Please don't forget to post how it goes. I need to know how to do it right on a dual core. (my methods haven't worked so far)

-Sid

I may have gotten it solved. Here's what happened.

Before, I was running the CPU and GPU client by itself and everything's fine. If I run them together, the CPU client would GREATLY slow down the GPU client, according to task manager. So then I have F@H do its affinity lock after I reinstalled F@H. Still no go if I ran both the CPU and GPU client together. So I went to Task Manager, and did the following:

F@H CPU (and anything related to CPU)> Set Affinity> Check only Core 0
F@H GPU (and anything related to GPU)> Set Affinity> Check only Core 1

I didn't restart the computer yet and let it run on its own as a service. At the moment, I have the programs running as command prompt.

Oh, and I did a "min-experiment". I had just the GPU client running, and it would complete 1% approximately every 2 to 3 minutes. Now that I changed the affinity and have them both running together, the GPU client takes 2 to 3 min to complete 1% while the CPU client takes about 5 min to complete 1%.

Just tried running the CPU client by itself, and it takes just a bit over 4 min to complete 1%. This is the same as when running both the CPU and GPU client together.

I'm going to restart the computer and let the clients run as a service. I'll use F@H to monitor how long it takes to complete each percent. If anything changes, I'll report it in this thread.

Edit: As soon as I restart the computer, I can't control the Affinity anymore. Guess it's only GPU client till I can find another way to deal with the core affinity problem.

Edit: I think I solved it. This program here lets you change the affinity settings. Just remember that the affinity won't run as a service so you gotta set it up yourself. I'm going to use this program for now. Next up I'll be trying the SMP affinity setter and see if that changes anything.

 

[Insidious]

Sounds like my experiences here.

It seemed like every time I restarted the client or it began a new work unit, the affinities went back to their snaillike selves.

let us know how the affinity setting utility works out.

-Sid

PS: You don't have a prayer of running an SMP client and a GPU client (well, for max total PPD, that is) on a dual core PC.

PPS: The SMP affinity changer only functions if you are running two SMP clients on a quad.

I inow it sounds like I am being the voice of doom.... I don't mean to. It's just that these clients haven't been developed in a way that gets WindowsXP to allocate resources very well.

 

[geokilla]

The program that I linked in the previous post changes the affinity setting for SMP clients. I'm still trying to figure out how to get it to work on the individual CPU clients.

I got the older version of that Affinity Changer program. I'm looking for one that can be used as a service and does NOT run SMP. The latest one, which is SMP, like I said before doesn't work with single core clients.

 

[Markfw]

Well, the one I have works on smp clients, and seems to help gpu also. I can email it to you if you pm me your email address.

 

[Insidious]

I found (at least part of) the answer.

Any time you run a GPU cient alongside one of the CPU clients (GPU or uniprocessor) you can set a wndows system environment variable to force the GPU to use any available core instead of the one that it is assigned by default.

Right-click 'my computer' and select 'properties'
click the 'advanced' tab and you will see two windows for managing windows environment variables.
Use the system window and click on the 'create new' button
enter the name of the variable: NV_FAH_CPU_AFFINITY
enter the value you want for the variable:
0= use all cores
1= use core 0
2= use core 1
3= use core 0 + 1
4= use core 2
5= use core 2 + 0
6= use core 2 + 1
7= use core 2 + 1 + 0
8= use core 3
9= use core 3 + 0
10= use core 3 + 1
11= use core 3 + 2 + 0
12= use core 3 + 2
13= use core 3 + 2 + 0
14= use core 3 + 2 + 1
15= use core 3 + 2 + 1 + 0

I just put in 0 so all cores will always be available.

So far, I know on an Opty (X2) the GPU client and the CLI client both fold at rates very similar to what they would do if operating by themselves.

Also, on my Quads, I did this. They run two GPUs and one SMP. All three of those clients also fold at rates very siimilar to what they would do on their own.

I hope this helps!

-Sid

(edit: don't forget to re-start the clients for this change to take affect.)

 

[geokilla]

Originally posted by: Insidious
I found (at least part of) the answer.

Any time you run a GPU cient alongside one of the CPU clients (GPU or uniprocessor) you can set a wndows system environment variable to force the GPU to use any available core instead of the one that it is assigned by default.

Right-click 'my computer' and select 'properties'
click the 'advanced' tab and you will see two windows for managing windows environment variables.
Use the system window and click on the 'create new' button
enter the name of the variable: NV_FAH_CPU_AFFINITY
enter the value you want for the variable:
0= use all cores
1= use core 0
2= use core 1
3= use core 0 + 1
4= use core 2
5= use core 2 + 0
6= use core 2 + 1
7= use core 2 + 1 + 0
8= use core 3
9= use core 3 + 0
10= use core 3 + 1
11= use core 3 + 2 + 0
12= use core 3 + 2
13= use core 3 + 2 + 0
14= use core 3 + 2 + 1
15= use core 3 + 2 + 1 + 0

I just put in 0 so all cores will always be available.

So far, I know on an Opty (X2) the GPU client and the CLI client both fold at rates very similar to what they would do if operating by themselves.

Also, on my Quads, I did this. They run two GPUs and one SMP. All three of those clients also fold at rates very siimilar to what they would do on their own.

I hope this helps!

-Sid

(edit: don't forget to re-start the clients for this change to take affect.)

Oh wow. Nice find. How'd you manage to get that?

Also, I seem to be getting some EUEs on my CPU folding. Would this be caused by an unstable overclock even though the system is OCCT stable for 2+ hours?

[20:40:47] Completed 1425000 out of 2500000 steps (57%)
[20:42:16] Writing local files
[20:42:16] Completed 1450000 out of 2500000 steps (58%)
[20:42:16] Writing checkpoint files
[20:43:45] Writing local files
[20:43:45] Completed 1475000 out of 2500000 steps (59%)
[20:44:21] Core cannot continue further.
[20:44:21] Going to send back what have done.
[20:44:21] logfile size: 21103
[20:44:21] - Writing 21623 bytes of core data to disk...
[20:44:21] ... Done.
[20:44:21]
[20:44:21] Folding@home Core Shutdown: EARLY_UNIT_END
[20:44:24] CoreStatus = 72 (114)
[20:44:24] Sending work to server
[20:44:24] Project: 4609 (Run 38, Clone 1, Gen 19)

 

[Markfw]

Its been happening to me for weeks. Thats why my 41k pps is really 35. I just killed 2, that 4k points I lost in one day, par for the course. But these are on boxes that run 24/7 for months with no problem until this latest set of bad units.

A bad OC (too much) can also produce the same results. If you can do 2-3 units in a row with no EUE, and then you get one, it could be a bad work unit.

 

[geokilla]

Thanks Mark for the clear up on the EUE and the Affinity setter. I'll see if I continue to get EUEs. If I do, I'll raise my Vcore by one notch.

It seems that running the Affinity Setter that you sent to me, the Folding cores still prefer to be ran on Core 1. GPU folding seems to takes an around 1 min and 55s to complete one percent, while CPU folding stays the same. Before when I wasn't running the Affinity Setter as a service, GPU folding would take 1 min and 51s to complete one percent. That is a difference of 4 seconds.

I just remembered I can't run SMP+GPU since my computer isn't on enough on the weekdays. It's on only 8+ hours a day during the weekdays.

I'm going to restart the computer and see if there's anything different. If there is, I'll post it here.

Oh, and I tested Insidious' method, by changing the Windows System Environment Variable, and it seems to work just as well, if not better since the one that Mark sent was specifically designed for SMP, I think.

Edit: It seems that FAH will take up 100% CPU usage now. Now I just have to figure out why I'm lagging! If worst comes to worst, I'll have to drop the CPU usage down to 95%. This is after I restarted the computer btw.

And I have to raise the vcore by one notch. Got another EUE.

 

[Insidious]

geokilla, The environment variable is only for the GPU clients and yes, it controls what cores they use (sets affinity to the core(s) selected).

I found it while lurking around in the fah forums. It really is a good resource, but you have to wade through an awful lot of BS to find answers in a lot of cases.

I think the "Project: 4609 (Run 38, Clone 1, Gen 19)" indicates this is a uniprocessor client that is giving you the eues. That's not normal at all for the basic client. I would suspect that you might have some instability in your system. (If you OC it, you will probably end the issues if you back it down a little.)

I remember when I was running a lot of the uniprocessor clients, I could have a machine that would do Prime95, Futuremark, memtest all fine, but trash work units when it folded. I finally realized that the clients do not tolerate errors well and ended up using F@H as my stability tester instead of all those other tools. F@H can find instabilities they missed (since it runs for days on end).

Mark, I noticed you have some pretty massive overclocks in your farm. I would make the same suggestion to you about considering backing those down a little bit. You are loosing more WUs than most people seem to be.

You might also want to re-visit your home network setup. If your addresses aren't static or you have any kind of 'flakyness' that causes your router(s) to hiccup, you will loose a lot of SMP work units.

-Sid

(back to your regularly scheduled :beer: )

 

[Markfw]

Sid, I figure (in my case) if I have a box that come go for WEEKS with no eues, then abends 3 times in a row at exactly the same step, its a bad unit. I have reproduced this case more times than I care to do (20 times at least) so I know that bad units are out there. All my boxes are rock solid at their current speed. Rick now its about one in 10 that EUE, and always 2665 units. I am starting to get 2653's again and some 2605's.

 

[Insidious]

I'm glad I've been spared those (so far). I know the frustration of finding a client that has decided to puke and knowing how much gets lost when they do.

Maybe it's that I'm not running any boxes with dual SMPs anymore that is sparing me. I tried the linux route and that was just too high a price to pay for a stable client. (Linux is :evil: )

Then I broke down and started with the GPU stuff so now I can't fit two SMPs on any of my PCs. (I kind of like the GPU stuff so far, but the client and work units are still buggy so I didn't get to retire from babysitting here either. )

Hang in there!

:beer:

 

[Markfw]

Linux is better than windows on dual smp, but I just had a smp client get to 100%, said finished, THEN "client-core communications error" and no credit. A whole day wasted, what a rip-off.

 

[Insidious]

I've read about that one.....

Here's how to get credit when this happens. (also works for partial EUEs)

Good Luck!

 

[Drsignguy]

Originally posted by: Insidious
I've read about that one.....

Here's how to get credit when this happens. (also works for partial EUEs)

Good Luck!

Well this is a great thing to know..To bad I didn't have a couple of weeks ago when I had just about all mine gone EUE. . Thanks for the link sid!

 

[geokilla]

Hey guys. Thanks for all your help on setting up FAH "professionally". I used the "NV_FAH_CPU_AFFINITY" method and now both the GLI+GPU client takes up 100% of both cores, giving me lag. I guess I have to back it down by 5% to take away this lag that I'm experiencing.

Can someone help me set up a Q6600 + GPU for folding? I know the basics, such as SMP+GPU, but how would I change the affinity settings? Does Mark's FAHSMPAffinity setter work just as well for the SMP+GPU folding? I need to be able to help my friend set it up via MSN and phone since I have no clue where he lives.

 

[Insidious]

I was under the impression (from reading their website that the affinity changer will not do anything unless you have more processes spawned by the SMP than you have cores. ie: one smp on a dual core or 2 smps on a quad.

I use the environment variable on my quads (they run 2 GPU + 1 SMP) and the GPU clients PPDs go up to what they would be if the smp wasn't running.

If you ran 2 SMPs and any GPU on a quad, then the affinity changer, in conjuction with the NV_FAH_CPU_AFFINITY environment variable would be the way to go.

If you run 2 SMPs only on a quad, the affinity changer is good

If you run 1 SMP only on a dual, the affinity changer is good

If you run a GPU with anything, the environment variable is good.

As to the rest of the set up, just follow the installation guides at the fah website.

Hope this helps

-Sid

 

[geokilla]

How would dual SMP + GPU work on a Q6600 + GTX 260 though. SMP takes up at least 4 cores doesn't it? And GPU takes up one core, which means that his system will lag and will experience a drop in PPD right?

 

[Insidious]

I don't run two SMPs. I am running two GPU clients and a single SMP client on my quads.

Can't answer that one for ya.

-Sid