GPU2 client for nVidia cards

[Samuknow]

How do you possible have time to read all this stuff?

[ThunderRd]

Originally Posted by Samuknow

How do you possible have time to read all this stuff?

Good time management? I read when you guys are sleeping, so I can blast you with the info when you wake up... Finished two books this week, too. Remember, I'm self-employed [when I want to be].

How's the back feeling, BTW? Any better?

[Samuknow]

Actually very good this seek. tight and still sore but sooooo much better. I realize after yesterday that i still need to watch over doing it but all is good.

I finally broke 5000 PPD. Won't last but I have been at it long enough.

[PorPorMe]

Can either of you two think of a reason why both my cores are running at 100% when I'm running one GPU client as well as one CPU client. My Graphics ain't to bad for a little while after a boot but then they turn to crap. I've got the GPU shut down so I can try and find what is going on otherwise I can't even navigate through different parts of the computer. I'm going to find it soon or the cure all is going to come out. Wipe out C:\! One of my spyware programs has an online deep scan of all exe and dll's that are running and gives a report. It found one I could not associate with any program along with a number of hidden ones by Cyberlink and Nero. Removed those but no help.
Virus, spyware, and Trojan scans picked nothing but adware cookies.

Edit:
I can't say for sure but it seemsthis problem devoped at the same time as the GPU client changed. I also could be looking for a place to lay blame and that one looks good. I honesty don't know.

[Samuknow]

The GPU client only uses 25% total. Or 50% of one core. The CPU client uses the other core. Not sure why you are maxed out. Look and the task manager to see what is using the cycles.

I have 2 GPU clients and one CPU client running on one rig. Both cores are maxed out.

[PorPorMe]

Here is what is running on my rig. See if you can spot something.

[ThunderRd]

Originally Posted by Samuknow

The GPU client only uses 25% total. Or 50% of one core.

It uses half of a core, but one GPU client on my machine reports 100% usage of 1 core. I know it isn't constant. That's why 2 GPU clients can occupy 1 core. But 1 client should show 100% use of 1 core in Task manager. [I think.] It needs exclusive access to a core because the CPU has to feed the GPU with the calculations. If it doesn't have exclusive access, it runs much slower.

@PPM: The client isn't new, just the WUs. You were fine before the new WUs, and now you're not. Please delete the work folder and the queue file, and download a new WU. Is it in the "new" group? If it is, delete it again and again until you get an "old" group WU. Run it. Do you have the same problems?

If you do, then it's something else, and we'll go from there. But don't yank stuff off the drive yet. Take it easy.

[PorPorMe]

If I'm understanding you corectly, the objective is to get the GPU client to start a new WU. I have deleted the work dir and the Q 7 0r 8 times. Why don't I just delete the client and reinstall it?

Edit:
Just so you know, I don't have all the dignostic skills in the world and I am not bothered about sticking my curser in where it might not belong. I have learned more that way than any other and I also have learned to have things set up when, not if, the effort to find the problem is not going any where. I'm set up to Format C:\ and be back up to allmost the same point I was in approx. 3 to 4 hours. Not a problem for me. I much prefer to solve the problem. With your help there has been more than one. Just don't get bothered if we hit a wall. Not a big deal.

[ThunderRd]

Originally Posted by PorPorMe

I have learned more that way than any other and I also have learned to have things set up when, not if, the effort to find the problem is not going any where. I'm set up to Format C:\ and be back up to allmost the same point I was in approx. 3 to 4 hours. Not a problem for me.

The objective was to get an "old" style WU [not from 5508-5513 or 5017-5022] and see if the problem persists. When you deleted the work folder 7 or 8 times, did you get an "old" WU any of those times?

Reinstalling the client does nothing if the problems are a result of the WUs themselves. The client was running fine, and all of a sudden, on the same day that the new WUs are released, along with several thousand other users, it begins to misbehave. I just thought I'd save you some effort in diagnosing.

I much prefer to solve the problem. With your help there has been more than one.

My apologies. When I saw your post this morning, I thought you were looking for a hand.

If you're more in your comfort zone tearing down your system instead of figuring out what the problem is by elimination, then feel free. I have a very strong suspicion that the problem will remain even after, until they get the WUs and the drivers sorted out, though. Good luck, then.

[ThunderRd]

Problem updates-@anyone who's interested:
There is a definite problem with these WUs, and PG is on the case:
Folding Forum • View topic - new public NV projects 5508-5513, 5017-5022
According to that post earlier, they discontinued distribution of the new WUs, at least temporarily.

This one is from the Boss:
Folding Forum • View topic - new public NV projects 5508-5513, 5017-5022
This explains a bit about why these WUs have been problematic. A new core version is in the offing-don't be surprised if you see a forced core upgrade very soon.

Bottom line is-if you are having problems, instability, lag, EUEs, Unstable Machine errors, crashes, or heat issues, and the client was running OK before, just sit tight. Help is around the corner. Hundreds, if not thousands, of others are having the same trouble.

[PorPorMe]

Originally Posted by ThunderRd

The objective was to get an "old" style WU [not from 5508-5513 or 5017-5022] and see if the problem persists. When you deleted the work folder 7 or 8 times, did you get an "old" WU any of those times?

Reinstalling the client does nothing if the problems are a result of the WUs themselves. The client was running fine, and all of a sudden, on the same day that the new WUs are released, along with several thousand other users, it begins to misbehave. I just thought I'd save you some effort in diagnosing.





My apologies. When I saw your post this morning, I thought you were looking for a hand.

If you're more in your comfort zone tearing down your system instead of figuring out what the problem is by elimination, then feel free. I have a very strong suspicion that the problem will remain even after, until they get the WUs and the drivers sorted out, though. Good luck, then.

Back up the trolley! We have miscommunicated again. I thought I was clear that I would prefer to work through the problem with your help but if we hit a wall I could go another way as well.
I could not get a different WU and thought reinstalling the client would solve that problem. I did ask, I didn't do it. I tried again to delete the files and it finialy worked. I got a different WU. a 5510 but, as you guessed, it didn't solve the problem.
So, you want to just sit tight for a few days and see what shakes out?

[ThunderRd]

OK then, what I suggest is to get yourself a WU that is of the old style: 480 points. Those WUs are not in the group 5508-5513 or 5017-5022[these are all 430 points]. As you see, the new unit you finally got was a new one, and will exhibit the same problems, if I'm correct.

It may take a number of tries to get an old one, but I think that the chances are better right now with some of the new WUs taken off the shelf by PG.

When you get an old one, run it and see if you get the same problems.

Incidentally, I know it's probably a silly question, but after a reboot, with NO CLIENTS running, do you still get funky video? [That means that you have to turn off the autostarting services]

[Samuknow]

I have some 5508s right now and no issues. two folding just fine. At least it seams so.

[PorPorMe]

Originally Posted by ThunderRd

OK then, what I suggest is to get yourself a WU that is of the old style: 480 points. Those WUs are not in the group 5508-5513 or 5017-5022[these are all 430 points]. As you see, the new unit you finally got was a new one, and will exhibit the same problems, if I'm correct.

It may take a number of tries to get an old one, but I think that the chances are better right now with some of the new WUs taken off the shelf by PG.

When you get an old one, run it and see if you get the same problems.

Incidentally, I know it's probably a silly question, but after a reboot, with NO CLIENTS running, do you still get funky video? [That means that you have to turn off the autostarting services]

Both serices are turned off. I have spent the last 45min deleting the GPU files and starting the client agian. Approx. 35 times. It won't drop the 5510. I'll keep at it if you want.
With both services turned off and butted up graphics are fine, no hang ups.

[ThunderRd]

I've got no more ideas for you right now. Are you deleting the queue file as well as the complete contents of the work folder? If the client won't pull an old WU to check what's at fault I can't give you any more information. I can't understand why it won't pull an old one eventually. My box has been getting some of those between the new ones.

The system has the download of all WUs weighted by some probability formula I'm not 100% clear on. Today [last night for you] the newer WUs were weighted differently so they don't go out as often as before. This is temporary until the new core comes, I would think. I think the fact that your system is OK without the clients running mens that the WUs are at the bottom of the problem-not the clients; I mean, everything was good til the new WUs came and then BAM. Problems. If I were you I'd run the clients overnight when you don't need the machine, until the developers settle the problems down.

I'm having problems with EUEs and this Unstable Machine message on WUs sandwiched between successful completions, and I can't get it stable either. I may just shut the GPU clients down indefinitely and wait til I can run stably again. This box is pulling 352 watts at the wall with 3 clients running. That's a lot of juice.

[PorPorMe]

I delete the work folder itself.
This will throw a curve ball at you. After trying another 10 or 15 time, I decided that I'd geive a new client a try. Deleted the GPU client and installed a new one but it picked up a 5506 WU.
Now for the curve ball. The graphics are fine. It's been running 20min or so and I'm good to go. I am waiting for the shoe to drop but so far it's OK.

Edit:
Reading in F@H forum that haveing the vid card OC'd to much will be problimatic, mainly with EUE rate. I had stopped all OC on mine early yesterday. The timeing is wrong-still had flaky graphics after OC-but I thought I'd bring it to your attention anyway.

[PorPorMe]

thunder,
I was looking at overclockers and how they were dealing with the change in the GPU change and two things caught my attention. This is what they were saying.
1: Shader increase = PPD increase
2: Having a CPU client running at the same time drops PPD.

I had gone default earlier so I increased shaders from 625 to 675. Thats the only change.
Timing was perfect. GPU was about to finish a WU. I shut down the CPU client.
When the GPU finished and then picked up another WU it was another 5506 but the PPD was 3602 where the last one was 25** . Bye the way, the credit is 480, they were saying that for some it had dropped to 435.

Edit:
Graphics have started to lag a little. Not anything I can't live with but just noticed it.

[ThunderRd]

This new update from the Boss:

Originally Posted by VijayPande

Here's an update. Mark has a new core with several mods for these issues. [EUE, UM, etc.] It's in QA right now. ETA is at least a few days to go through QA though.

Originally Posted by PorPorMe

...it picked up a 5506 WU. Now for the curve ball. The graphics are fine.

That's not a curve ball, that is good news, and it's what I expected to find. You finally got a 5506, which is a 480 point WU from the old set. [It's 480 points, no matter what anyone tells you-PG never changes the value of an existing WU, only the value of new ones.]

The fact that your graphics are OK now tells us that the WUs are the problem-I seriously doubt that reinstalling the client was the problem. It might have had an effect on the ability of the client to download an old WU, though, although I'm not sure why-maybe a config file of some sort that was replaced.

In any case it's better than it was. I would avoid all overclocking of any sort for the time being, until the new core and driver is available. YMMV, though.

Incidentally, I'm getting new WUs only about 50% of the time, and about 1 in 3 have a problem as of this post.

Originally Posted by PorPorMe

Having a CPU client running at the same time drops PPD.

I have seen nor heard no evidence of this. There isn't any reason for this to happen if the GPU client has exclusive access to its own core. Ask the poster if he set the core affinities of the GPU executable properly.

Originally Posted by PorPorMe

When the GPU finished and then picked up another WU it was another 5506 but the PPD was 3602 where the last one was 25**

Of course. The 480 point old WUs run at least 30% faster and make more points. The last WU you got was 25 and change because it was in the new series. Normal.

Originally Posted by PorPorMe

Graphics have started to lag a little. Not anything I can't live with but just noticed it.

The GPU client stresses the card more than any stress test or game can. On the main rig in my signature, which is no slouch, I get some laggy scrolling and other stuff with the client running. How much depends on the WU, and it's normal. I think you have to just get used to it.

[PorPorMe]

Bear in mind that tonight I'm a wee bit busy. Now, I'm not sure you are correct about the CPU client having no effect on the PPD of the GPU. I'm not sure because I got busy and wasn't paying attn. You may be right, but something is going on here. Bear with me.
I shut down the CPU client. I also increased shaders. that was a mistake-confuses the issue. PPD that had dropped 1k went back up 1k. While I was busy and paying no attn, it finish the 5506 and started a 5022 and the CPU started as well. the PPD for the GPU again dropped 1k. Graphics are still good. The GPU is at 25% and should be done before I go to bed so I'm shutting down CPU and all OC on the card. I want to see if the PPD goes back up 1k. I think it will.

[ThunderRd]

Originally Posted by PorPorMe

Now, I'm not sure you are correct about the CPU client having no effect on the PPD of the GPU.

It certainly wouldn't be the first time I was wrong. But I can't find anyone else on the forum with the issue, and I can't reproduce the problem on any of my 4 GPUs. This guy observes quite the opposite, in fact: Folding Forum • View topic - new public NV projects 5508-5513, 5017-5022

Originally Posted by PorPorMe

I shut down the CPU client. I also increased shaders. that was a mistake-confuses the issue. PPD that had dropped 1k went back up 1k.

Yes , an extra variable can obscure the truth. What WU was this?

Originally Posted by PorPorMe

... it finish the 5506 and started a 5022 and the CPU started as well. the PPD for the GPU again dropped 1k.

I don't understand. Are you saying that the CPU client started by itself?

Also, be careful of comparing apples to oranges. 5506=old=fast. 5022=new=slow. It would seem perfectly natural to me that a 5022 would run much slower than a 5506.What makes you believe that the CPU client caused the slowdown? What I'm hearing you say is this:
1. You were running a 5506 with no CPU client running, and it ran OK, with normal frame times.
2. The client snagged a 5022 and began running it, and somehow the CPU client started at the same time.
3. The frame times dropped dramatically. The question is, was it because of the new WU or the CPU client running?

Again, I'll say that I have not been able to reproduce the alleged behavior - with the core affinities set to place the GPU client on its own core, and the CPU client running, the GPU runs at exactly the same rate as it does with the CPU client shut down.