Joining GPUGRID, a problem

[PoppaGeek]

Got tired of Folding @ Home problems between 4850 and GTX 260. They seem to have enough anyway. Time for something new.

I downloaded the 6.4.5 64 client as per instructions, lost all my current work which I thought Boinc did not do that? Same 64 bit client and I installed over old one 6.2.28. Oh well.

I started Boinc and had to attached to WCG and get new WUs. It started 4 jobs, quad core. Attached to GPUGRID, it downloaded some work and started a job.

One of the WCG jobs went "Waiting to Run" and CPU usage shows GPUGrid using 4% or less CPU and 3 WCG WUs using 25%. The 4 th one still shows in Task Manager at 0%. I have 100% in preferences for CPUs and usage.

Tried restarting it and got same. Any ideas?

[PoppaGeek]

AND another thing! BoincView shows the GPUGrid yellow which means not running 100%. 336.8 MFlops. WCG WUs run 2 GFLops up. Video is GTX 260 192.

:toast:

[MikeB12]

I'm still using the same client I was using when I started a month or so ago.. cpu util is about 2% per gpu.
The 2 gts 250's turn about 19-20k boinc gpugrid rac, and the cpu turns about 2k boinc wcg rac.
*Q6600@3330mhz 1.35v - True120 | Abit IP35pro | 2x2GB Gskil 1000@925mhz | 2 x EVGA GTS 250- shader 1890 | Corsair HX520 | OCZ Vertex 60GB/Seagate 7200.11 500GB
http://img248.imageshack.us/img248/5083/45909169.jpg

[MikeB12]

btw: back up your data directory before uninstalling the client, then copy it to the new fresh data directoory the new client installs. this saves your work. hindsight 20/20. LOL!

[Naja002]

Got tired of Folding @ Home problems between 4850 and GTX 260. They seem to have enough anyway. Time for something new.

I downloaded the 6.4.5 64 client as per instructions, lost all my current work which I thought Boinc did not do that? Same 64 bit client and I installed over old one 6.2.28. Oh well.

I started Boinc and had to attached to WCG and get new WUs. It started 4 jobs, quad core. Attached to GPUGRID, it downloaded some work and started a job.

One of the WCG jobs went "Waiting to Run" and CPU usage shows GPUGrid using 4% or less CPU and 3 WCG WUs using 25%. The 4 th one still shows in Task Manager at 0%. I have 100% in preferences for CPUs and usage.

Tried restarting it and got same. Any ideas?

Very rarely will you lose the WUs when upgrading versions. I lost them all going into and then again out of 6.4.5. Did that to some people, but not others. 6.6.36 is running very well for me and there have been a lot of fixes between 6.4.5 and 6.6.36 for GpuGrid. You may want to go ahead and upgrade.

Check Advanced>Preferences>Processor tab and see if "Use GPU while computer is in use" is checked. If not, check it. If that does not resolve the 3 + 1 issues then:

Make a file in notepad
Name it: cc_config.xml

Copy and paste the below into it:
------------------------------

<cc_config>
<options>
<ncpus>5</ncpus>
<report_results_immediately>1</report_results_immediately>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>

------------------------------

Save it to: C:\Programdata\Boinc\

"ncpus" = whatever works to get things running and may not be needed. I had to add it back when I started (9-08), then removed it, then added it back recently. Originally the number was the total cores + gpus. Now it the same except minus 1. Example: i7 running 8 threads and 2 gpus=9. Same for my other rigs....go figure.

"use_all_gpus" stays at 1 regardless of how many you have.....the number is irrelevent from what I understand. :up:



Unless there's something I unaware of....the 4850 will not run on GpuGrid atm. It will run on "Milkyway" (http://milkyway.cs.rpi.edu/milkyway/) which is also a boinc project. I think MW and F@H are the only 2 that currently use ATI cards. The Lattice Project might...


AND another thing! BoincView shows the GPUGrid yellow which means not running 100%. 336.8 MFlops. WCG WUs run 2 GFLops up. Video is GTX 260 192.

:toast:

That's a GOOD thing! :up: :p: I use boincview also. It reads cpu usage, so the lower the number the less cpu that is being used and the more gpu. Don't know why, but also the faster you overclock the card--the lower the number will go, so lower is better! Sadly, work on boincview has been discontinued. I believe it's open source now, so hopefully someone will pick up the ball and run with it. The version at Boinc is old. I have the last version: 1.4.2 if anybody wants a copy.

HTH

[PoppaGeek]

I upgraded to 6.6.36 and when I finally realized there was a new option, "Use GPU when computer in use" and checked it it was running. I tried the <ncpus>5</ncpus> and got 5 WCG WUs crunching. lol yeah that is how it was going.

Anyway after it finally ran a few hours I went to web site and got this:http://img.techpowerup.org/090704/gpugrid.jpg

I just detached from project. When I feel like reading more forums i'll see about trying again sometime but I saw alot of people have problems with different drivers and some not. Sounds as spotty as F@H.

I may get a 4870 for gaming and put the GTX 260 on one of the Opteron boards with Win7 and 182.50 and leave it at that. I have run Fur and CUDA Memtest several times and no errors so I do not think it is the GPU.

Thanks for replies. :up:

[PoppaGeek]

btw: back up your data directory before uninstalling the client, then copy it to the new fresh data directoory the new client installs. this saves your work. hindsight 20/20. LOL!

I thought I was already running the 64 bit client and it was the 32 bit so the 64 bit ignored it during install. When I upgraded from 6.4.5 to 6.6.36 I backed up the data dir and then replaced it after upgrade. Live and learn. :up:

[Naja002]

I upgraded to 6.6.36 and when I finally realized there was a new option, "Use GPU when computer in use" and checked it it was running. I tried the <ncpus>5</ncpus> and got 5 WCG WUs crunching. lol yeah that is how it was going.

The "ncpus" is now minus 1 for some reason (for me anyway)--I don't know why. Currently it's: However many cpus cores (4) + the number of nvidia gpus (1) minus 1. So, I should have put 4 inside the <ncpus>?</ncpus>. Regardless though, if just checking the "Use GPU whill computer is in use" works...then that's all that's needed. You can delete the "ncpus" and leave the other 2 options.....or just delete the entire file--it doesn't matter. :up:


"ncpus" = whatever works to get things running and may not be needed. I had to add it back when I started (9-08), then removed it, then added it back recently. Originally the number was the total cores + gpus. Now it the same except minus 1. Example: i7 running 8 threads and 2 gpus=9. Same for my other rigs....go figure.





Anyway after it finally ran a few hours I went to web site and got this:

I just detached from project. When I feel like reading more forums i'll see about trying again sometime but I saw alot of people have problems with different drivers and some not. Sounds as spotty as F@H.

Right. Atm I'm not sure what's wrong. Drivers? Yeah, I didn't even go there in my previous post. The forum over at GpuGrid has been pretty slow lately, because things are actually running pretty well. I'm currently not having any issues other than self-induced issues! :D GpuGrid just crawled back out of H3ll about 1 month ago. Real nightmare for people running lower end cards like myself (8800s). A lot of the driver info you are seeing is dated. I am currently using 185.85 and 185.26. The problem was actually on either GpuGrid, Boinc, Nvidia--take your pick. I think is was GpuGrid, but whoever got it sorted out. :up: The screenshot that you posted looks familiar.....new folks show up with that type of scenario. I saw your post over at GpuGrid--someone should step up before long and help you out.

The ATI card is not going to run on GpuGrid. Will it cause your error issues across-the-board if it's in the PC--I really don't know. Is the ATI card installed in the PC? If it is, you may want to remove/disable it and see if it's causing the issues....:shakes:




I may get a 4870 for gaming and put the GTX 260 on one of the Opteron boards with Win7 and 182.50 and leave it at that. I have run Fur and CUDA Memtest several times and no errors so I do not think it is the GPU.

Thanks for replies. :up:

I too doubt it's the Gpu......there's something else going on. You may just want to detach from both projects and reattach--If I understand your post above....you copied and pasted 32-bit back into 64-bit? Seems like you've realized it and moved beyond that.....?

[PoppaGeek]

Thanks for the replies!

At the moment I took CPU to stock clocks, set everything back to [AUTO] in BIOS and dropped drivers to 182.50. It has run one WU 1 hour 42 minutes so far. Never went past 40 minutes before. So we may be getting there.

Ya know after awhile all these driver and software version numbers get so confusing I can't remember what's where or when it was or how it got there! I have to keep looking to see what is what every time I post.:ROTF:

I dunno I guess it may have had to do with the OC on the CPU although F@H ran error free at that clock as did WCG. Guess GPUGrid does something different. No matter as long as it all works.

Hope this project works out ok. Kinda tired of F@H and it's many issues. GPUGrid seems the underdog compared to Stanford, number of crunchers wise. I like helping the underdog. Also like Boinc.

Thanks again!

:toast:

[Naja002]

Thanks for the replies!

At the moment I took CPU to stock clocks, set everything back to [AUTO] in BIOS and dropped drivers to 182.50. It has run one WU 1 hour 42 minutes so far. Never went past 40 minutes before. So we may be getting there.

Your OC on the gpu crossed my mind, but I wasn't certain whether it was even OCed or not. As long as the rig is stable....the cpu OC shouldn't matter much, but the gpu will, of course. :up: Personally, I would OC the cpu where I want it and then start on the gpu. Once you find your OC (if you OC hard) you may find that it will still throw a random error--could be the OC, the WU or a number of other things....:rolleyes:




Ya know after awhile all these driver and software version numbers get so confusing I can't remember what's where or when it was or how it got there! I have to keep looking to see what is what every time I post.:ROTF:

My Friend, You are definitely not alone. :up: My best advice and the advice that I use is: Upgrade cautiously! :up: If the setup you have is working...don't be eager to fix what isn't broken. :up: I usually start twitching to upgrade every ~2 months if things are running well--Only because of version improvements, driver improvements, etc. But I have gotten into the habit of checking the GpuGrid forums for info on what's working and what's not before I pull the trigger. :up:




I dunno I guess it may have had to do with the OC on the CPU although F@H ran error free at that clock as did WCG. Guess GPUGrid does something different. No matter as long as it all works.

My guess is the Gpu OC if it is OCed.





Hope this project works out ok. Kinda tired of F@H and it's many issues. GPUGrid seems the underdog compared to Stanford, number of crunchers wise. I like helping the underdog. Also like Boinc.


Being the underdog is one of a list of reasons why I have stuck with GpuGrid. Once you get it going well--I think you'll like it. But keep this in mind: It goes through the same roller coaster ride as F@H. Everything runs itself and then turns to shat until they get it sorted out, straightened out then it goes back to running itself. One big difference is that the GpuGrid folks listen and accept responsibility if it's on their end. :up: They work diligently and constantly with Boinc and Nvidia. Problems have arisen because of Seti bringing their Gpu client online, Milkyway, etc. Other gpu clients do not always play well with Boinc and can screw up things for other projects. So, they all have to roll up their sleeves and get it all sorted out. I think GpuGrid is still technically Beta....but I think F@H is also.....:shrug: This is all just FYI. It's nice having everything in one client and it all counts towards Boinc-points. So, whatever you do in the future that scores Boinc-points--this will pile up with that. I've got 1 mil F@H, but they don't do squat for my Boinc-combined total! :up:

[PoppaGeek]

The GPU was not OCd at all. The CPU was OCd from 2.6 -3gz. At any rate it is still crunching that WU so looking good.

I like that they give you credit for errors. At least they recognize you are trying. I had so many F@H WUs go 6 - 8 hours and then error out. Wasted elec and heat. I do not mind if they cause an error as long as they admit and fix it. I just hate waste. Not checking closely and being in a hurry I lost alot of WUs, some under way and others in queue when I upgraded the Boinc client. I hate that. Even the ones in queue. Be nice if you could go to web site and mark them as lost so they could resend them. I like to do a good job and be responsible and that ain't doing a good job.

I agree with the "If it ain't broke don't fix it" but I have had other issues with a game I play online, LOTRO. It is a game with lots of bugs though so your never quit sure if it is the game or a driver.

The CPU overclock has not been the issue all the time either as I had problems with LOTRO and F@H when it was at 2.6 too. I am guessing GPUGrid just does something differently and it showed a weakness in the OC the others did not. 2.6gz is ok was at 3gz for a little more umph for WCG. I hope to get a 955 or 965 for the machine first of the month or Sept anyway.

How long does an average WU take?

:toast:

[Naja002]

How long does an average WU take?

:toast:


I'm running the same 8800s that I've been running all along. IIRC, on a 260 they should take 6-8 hours--not sure if that's with any OC at all or not. But the WU length has increased, so it may be a bit longer than that now. :up:

[123bob]

I'm running very smoothly on GPUGrid with three OC 260-216s, on three different machines. These are factory OCed BFG cards at 590 MHz core. All run Vista Ultimate 64 bit. I'm running BOINC 6.6.20 64bit, my driver details under device mgr says nVidia version 7.15.11.8209 dated 2-9-09. I don't know what that translates to in the "18X.XX" numbering.

Below is the run times of a page of my validated WUs.

http://i149.photobucket.com/albums/s48/123bob_bucket/680i%20install%20pics/gpugridwus.jpg

Note that I seem to get two flavors of WUs. Some take around the 2,900 second timeframe, and the majority are around 6,200 seconds.

I get about 15k BOINC PPD on these cards.

Hope this is helpful,
Bob

[PoppaGeek]

Finished a WU yesterday.:woot:Mom will be so proud! :party:


Machine was locked up when I got up today. That has never happened with this machine. :brick:

Rebooted we'll see what happens next. :shrug:


:toast:

[PoppaGeek]

At a loss here. GPUGrid finished a WU under 182.50 but as soon as I went into LOTRO game crashed/OS froze. I installed 185.85 and GPUGrid worked about 40 minutes and errored. Win7 has never felt right on this machine. My 9850 with Vista 64 has same model HD and it is snappy, apps load faster. HD Tach and other benchmarks show the 9850 HD as faster. I guess I can install Vista on the 810, I think I have 30 days to get a license for it. See if Vista does better. I think there is a Northbridge driver problem. I have the latest installed and it has been better with it.

I do not know what else to do.

[road-runner]

Video card getting hot after awhile?

[PoppaGeek]

No never saw it over 65 usually under 60. Fan on 65% in a HAF932 with 230mm fan blowing on it.

I have never been happy with the Win7 install on it. I guess I'll try Vista64 on it see if things are any better. I have tried so many Boinc versions and driver versions I deserve a rebate or something. I am real happy with Vista on my 9850 it just means I gotta buy a license in 30 days and of course had other plans. lol Win7 gave me a little longer. It's my intended game machine else I would stick Linux on it.

[PoppaGeek]

Installed Vista 64 and a big difference in the way the machine runs. Snappy, responsive, apps pop open. Tired 186.18 and problems all over. Now on 182.50 and so far looks ok. Do not know what Win7s problems were but neither installs ever felt good. Guess I'll buy another license. It is worth $100 to see it run so nicely.

[OldChap]

Sounds like I'll be trying out vista 64 on the qx rig then :up:

[PoppaGeek]

Have completed 3 WUs now. Do not know what the problems were with Win7 but Vista 64 runs so nicely and I could see a immediate difference in LOTRO. GPUgrid got happy with 182.150 drivers and LOTRO seems happy with some tweaks to it's *.ini file.

Reading so many people saying Win7 works great for them I was surprised in the difference with Vista but I do not care. Just happy things seem ok now.

:toast: