Posts by mdawson
log in
1) Message boards : Number crunching : a failed WU (Message 14805)
Posted 1938 days ago by mdawson
I seem to be lucky in that I never have failed units with the Collatz project, but I got one recently. Here's what was said about it. The task # 125434088. I've never looked at this info before so I don't know what caused the failure. Anybody got any ideas? There is a line in there regarding AMD Brook+ kernels, but I have an NVidia based card. I definitely don't understand that one.

<core_client_version>7.0.28</core_client_version>
<![CDATA[
<stderr_txt>
Name GeForce GTX 680
Memory 2048 MB
Compute 3.0
Processors 8
Memory Clock 3004000 kHz
Warp Size 32
Shader Clock 705500 kHz
Max Grid 2147483647 x 65535 x 65535
Max Threads 1024 x 1024 x 64
Texture Align 512
Collatz Conjecture v3.11 x86_64 for CUDA 4.2
Based on the AMD Brook+ kernels by Gipsel
Parameters 0
Start 2376873382552907000168
Checking 824633720832 numbers
Numbers/Kernel 262144
Kernels/Reduction 64
Numbers/Reduction 16777216
Reductions/WU 49152
Name GeForce GTX 680
Memory 2048 MB
Compute 3.0
Processors 8
Memory Clock 3004000 kHz
Warp Size 32
Shader Clock 705500 kHz
Max Grid 2147483647 x 65535 x 65535
Max Threads 1024 x 1024 x 64
Texture Align 512
Resuming at 2376873382683081419112

Highest Steps 1679 for 2376873382993361655199
Total Steps 431559660305465
GPU time 1110.28 seconds
CPU time 10.2649 seconds
Total time 1297.96 seconds
05:34:59 (2280): called boinc_finish

</stderr_txt>
]]>
2) Message boards : Number crunching : what is the difference between cuda40 and cuda31 (Message 14756)
Posted 1946 days ago by mdawson
I just let Collatz run on its own, but I see where some files get crunched with cuda40 and others cuda31. I'm curious as to what the difference is between the source files that would cause this. The source files appear to be named (numbered) fairly similarly, at least for the first 6 characters of the file names. Crunch times vary quite dramatically though. Cuda40 crunched files take on avg 22-25 mins and use 90+% of my gpu, while the cuda31 crunched files take 8-9 mins and use only 38% or so of my gpu.

I was thinking of running multiple WU's on my gpu, but if cuda40 is used, I couldn't do that. If restricted to cuda31, then I could run 2, maybe 3 WU's simultaneously, but there could be a tradeoff with the payoff. Not that I really care about the payoff, I'm in this strictly for the science, but it did make me wonder.
3) Message boards : Number crunching : "Aborted by Project" (Message 14294)
Posted 2040 days ago by mdawson
Ok, that makes sense. I forgot to update before I started using the new files. I suppose there will be another since I reverted back to my old executables. The 3.11 file took over 40 mins to crunch one WU, and the newer mini-collatz never did run. The cuda 23 version runs WU's in about 30 mins and the older mini-collatz in about 10 mins. No sense going slow if I don't have to.
4) Message boards : Number crunching : "Aborted by Project" (Message 14288)
Posted 2041 days ago by mdawson
"Aborted by Project" was the status on a WU tonite and I've never seen that before. Did something happen on my end, or the other end? Just curious as to what exactly that means. The application was mini_collatz 2.05 (cuda31).
5) Message boards : Number crunching : high priority???? (Message 14029)
Posted 2071 days ago by mdawson
Thanks for the pointer Claggy. Both of those versions appear to still be in alpha test mode. I don't run alpha. The fact that my gpu is running in high priority mode doesn't really bother me. I was just curious if it had some impact on the game I'm playing. I'm getting some lag and I'm just ruling out any possible causes. On a side note, putting the gpu to sleep doesn't seem to affect the game at all, so it can run in high priority mode all it wants!
6) Message boards : Number crunching : high priority???? (Message 14022)
Posted 2071 days ago by mdawson
What exactly does this mean? Collatz tasks are running in "high priority". There's at least a week before any of these are due, and with my new card, I'm producing better than 50,000 credits per day, so why the rush? Does running in high priority mode cause the gpu to lock out anything else that is using the gpu? For instance, I've been playing an online game at 5940x1080. It requires a lot of processing, and when playing the game, I'm getting a lot of lag. Putting the gpu to sleep for an hour doesn't seem to help the lag at all, so I'm thinking my problem is network related, but I'm not sure. So I thought I'd ask here. Anybody got any thoughts? I also crunch Einstein on 6 of my 8 cpu cores, with the game playing on a non-Einstein core.
7) Message boards : Number crunching : Anyone using a GTX680 yet? (Message 13958)
Posted 2085 days ago by mdawson
I'll get back to this in a day or two. Right now it's Monday, and I've got a lot of other things to do. Thanks for the advice. At some point, I'm sure I'll get this working the way I want.

BTW - I set a personal record yesterday. Almost 100,000 in a single day! That's easily double what I was getting with my old card.
8) Message boards : Number crunching : Anyone using a GTX680 yet? (Message 13953)
Posted 2086 days ago by mdawson
Slicker/Christoph,

Thanks for the tip on using a browser to view this file. I've always used the MS text editor, but you're right, these are xml files afterall, and browsers know XML, so that's what I'll use from now on.

I got the tags straightened out but I have not yet had joy with multiple tasks. I'm assuming that there would be 2 or more entries in the TASKS list for Collatz, but I have only seen the one. Is there a different way to tell if more than one task is running on the gpu? I might have been running more than one but didn't know it.

As far as the CUDA23 and CUDA31 entries go, I seem to remember something about certain files ran better under 23, and others under 31. Maybe that problem has been straightened out. I see 3 different apps running depending on the WU in progress. Some use CUDA23, some use CUDA31 and some use Mini_Collatz.
9) Message boards : Number crunching : Anyone using a GTX680 yet? (Message 13946)
Posted 2086 days ago by mdawson
Slicker,

I spent a good deal of time last night trying to get this to work, and all I seemed to be able to do was to F everything up. I saw how the file names were laid out and I made my changes accordingly, or so I thought. When BOINC MGR loads it erases files and then sends the wu's back. I've since taken the app_info.xml file out of the equation. All I want to do is to run more than one task on my gpu.

Even when I got app_info.xml to be read correctly, and I had changed the gpu value to something like .5 or even .33, it still did not load more than one wu on the gpu.

Here is the app_info.xml file I was editing. I now have 6 files in my collatz folder. Don't worry, it's named properly on my HD, I abbreviate here just because I'm a lazy typist.

The files are:
collatz_2.03_windows_x86_64__cuda23.exe
collatz_2.04_windows_intelx86__cuda23.exe
collatz_2.03_windows_intelx86__cuda31.exe
cudart32_31_8.dll
cudart_CUDA23.dll
mini-collatz_2.05_windows_x86_64__cuda31.exe

I think I used to have a file named:
collatz_2.05_windows_x86_64__cuda31.exe, but it's not there now.

Perhaps you can see something I've missed or I have done wrong. Any help would be greatly appreciated.

Again, all I want to do is to run multiple wu's on my single gpu. Eventually I'll get a new case and be able to use a secondary slot to reinstall my old card for additional processing, but it doesn't fit right now.

Thanks for any help!

p.s. I have nvidia based cards.

<app_info>
<app>
<name>collatz</name>
</app>
<file_info>
<name>collatz_2.05_windows_intelx86__cuda31.exe</name>
<executable/>
</file_info>
<file_info>
<name>cudart32_31_9.dll</name>
<executable/>
</file_info>
<file_info>
<name>collatz_2.04_windows_intelx86__cuda23.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>collatz</app_name>
<version_num>204</version_num>
<plan_class>cuda</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<max_ncpus>2</max_ncpus>
<flops>1.0e11</flops>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
<cmdline>L13 I8 S0</cmdline>
<file_ref>
<file_name>collatz_2.05_windows_intelx86__cuda31.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart32_31_9.dll</file_name>
</file_ref>
<file_name>collatz_2.04_windows_intelx86__cuda23.exe</file_name>
</file_info>
</app_version>
<app_version>
<app_name>collatz</app_name>
<version_num>205</version_num>
<plan_class>cuda</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<max_ncpus>2</max_ncpus>
<flops>1.0e11</flops>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
<cmdline>L13 I8 S0</cmdline>
<file_ref>
<file_name>collatz_2.05_windows_intelx86__cuda31.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart32_31_9.dll</file_name>
</file_ref>
<file_name>collatz_2.04_windows_intelx86__cuda23.exe</file_name>
</file_ref>
<app_name>collatz</app_name>
<version_num>203</version_num>
<plan_class>cuda</plan_class>
<avg_ncpus>0.5</avg_ncpus>
<max_ncpus>1</max_ncpus>
<flops>1.0e11</flops>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
<cmdline>L13 I8 S0</cmdline>
<file_ref>
<file_name>collatz_2.05_windows_intelx86__cuda31.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart32_31_9.dll</file_name>
</file_ref>
<file_name>collatz_2.03_windows_intelx86__cuda23.exe</file_name>
</file_ref>
</app_version>
</app_info>
10) Message boards : Number crunching : Anyone using a GTX680 yet? (Message 13940)
Posted 2087 days ago by mdawson
Christoph,

I checked out the app_info.xml link you supplied. I have a question or two about what I see,. Hopefully you can answer for me.

In looking through the example, I see where collatz_2.05_windows_x86_64_cuda31.exe is listed 3 separate times. I also have a file called collatz_2.03_windows_x86_64_cuda23.exe. I think mini-collatz is using the cuda23 file. My question is, do I have to triple up cuda23 entries like cuda31 is and then just append it to the end of the example (making appropriate changes of course), save it off as app_info.xml and park in the sonntags folder? It seems like a lot of wording but I know there's other stuff in there to optimize optimizations. Did that make sense??? I'll give it a try though. The worst that can happen is that a few wu's might get messed up if I do this wrong. But I'll be watching and be here to fix if necessary.

Thanks for the links!
11) Message boards : Number crunching : Anyone using a GTX680 yet? (Message 13939)
Posted 2087 days ago by mdawson
Thanks again Christoph. I hope I didn't get you into trouble on the Einstein board. I'll be sure to check out your app_info link. I've been all over Nvidia's and EVGA's sites reading anything I could get my hands on regarding the new GTX680. I'll look a little deeper at those sites and see if they have any info that would be useful to me post purchase. I don't remember much about either of them giving out any specifics of using their cards for computing like we do, but I could have missed something.

Cheers! I'm off to the beach for a lobster lunch! Neptune's Net, here I come!
12) Message boards : Number crunching : Anyone using a GTX680 yet? (Message 13916)
Posted 2090 days ago by mdawson
I just dropped one of these into my rig and am curious if anyone else has had some experience with this chip yet.

I'm running Collatz on the gpu, but with this new card, it's only running at 20% or so capacity even when I'm playing a game! I've read where some folks are running more than one task per gpu, and it seems as if I have enough system resources to do that, but I don't know the terminology to use in whichever file it is to make these kinds of modifications. Would someone be kind enough to fill me in on how to accomplish that.

Anyone got any numbers handy so I can get an idea of what this card "should" put out?

Thanks,

mdawson
13) Message boards : Number crunching : why does my mahcine do so poorly? (Message 11670)
Posted 2516 days ago by mdawson
Oh never mind. It seems to be working correctly now.
14) Message boards : Number crunching : why does my mahcine do so poorly? (Message 11669)
Posted 2516 days ago by mdawson
Ok, so I've gone back to the 2.03 version, but now I can only get one gpu to work. There is no app_info.xml file, so how do I restore my 2nd gpu to operation?
15) Message boards : Number crunching : why does my mahcine do so poorly? (Message 11660)
Posted 2516 days ago by mdawson
I saw a post yesterday about using the 2.05 optimized app. I downloaded it and attempted to copy the files into the proper directory. It turns out, the files were already there but 2.05 wasn't being used. I was getting a bunch of mini-Collatz WU's and they were running under 2.03. So then I copied the app_info.xml file into the same place and restarted BOINC. That's when all hell broke loose. My machine was so slow it would take minutes for anything to happen. I modified app_info.xml to lower values (L3 I5 S1), and I now have some control over my machine, but it is a dog. I can't operate this way and I'm going to have to stop Collatz.

I don't know if this is related or not, but over the last 2 months or so, I have experienced a HUGE number of video card crashes. (EVGA GTX260) I enjoy doing this scientific stuff and don't necessarily do it for the numbers, but I need my machine for work as well. Until this gets straightened out, I won't be running Collatz. Sorry fellas.
16) Message boards : Number crunching : new problem (Message 11441)
Posted 2534 days ago by mdawson
Ahh, That helped. I forgot about "All Users". Taskman showed AVP (Kaspersky Anti-virus) as the culprit. I don't know why it was using so much resources. Emailing them for help or with a question may get you an answer, but never in a timely manner.

Meanwhile, I have another problem. My main video card, the GTX260, is crashing like crazy. I got into the machine last night and vacuumed out all of the dust. It wasn't bad. There was more dust on the cpu cooler than the fan structure of the video card. Nonetheless, GPU temp down a little, cruising at 46 degrees centigrade. It still crashes though. Someone had mentioned maybe the power supply was failing. I'm looking into that. I hate to think that is the case as I spent a good deal of money to get one that would comfortably power the GTX260.

I have an Abit IP35 Pro m/b, and Abit's Uguru software indicates these values for power as listed under the OC uru tab:

VCore - 1.3v
DDR2 - 1.95v
MCH 1.25v - 1.25
ICHIO 1.5v - 1.5v
CPU VTT - 1.2v
DDR2 Ref - 0%v
ICH 1.05v - 1.05v

Under the ABIT EQ/Voltage tab, the values listed are all within range and list OK as the status. In fact, all of the 5v and 12v entries are at their marks or just slightly above. The only thing this program doesn't list is amperage. My card required something like 60 amps per rail and I bought a power supply that meets or exceeds that value. I have no way of measuring that. I realize that's for the cpu's, but if there were a failure there, it could possibly ripple into the video card, right?

I'm running both cards at stock values with only the fans being revved up a bit higher.

So I'm at a loss as to why the GTX keeps crashing. It's an EVGA card and they have released a new driver which I installed. That didn't help.

Anybody got any ideas?

17) Message boards : Number crunching : new problem (Message 11335)
Posted 2541 days ago by mdawson
As I mentioned earlier, task manager doesn't show anything using a serious amount of resources. If something evil were on my system, taskman would never indicate it. I use AUTORUNS to determine what is and what is not actually running.

However, I just put BOINC to sleep and opened taskman. This time it shows only 13-17% usage across all 4 cores.

Seems like all is back to normal. I'm keeping my fingers crossed. Thanks for all the responses.
18) Message boards : Number crunching : new problem (Message 11333)
Posted 2541 days ago by mdawson
Both cards are nVidia based. GPU 0 is GTX260, GPU 1 is GT8600.

System seems stable at the moment. Both gpu's crunching madly away. GPU 0 under high under priority.
19) Message boards : Number crunching : new problem (Message 11324)
Posted 2541 days ago by mdawson
I did do a reboot and that seems to have helped. What's particularly concerning to me is I have had quite a number of video card crashes in the last month. I think it's my GTX260, although when it crashes, it seems to take the GT8600 with it. I am not overclocking as I was very concerned with temps. They're both running in the neighborhood of 61-63 degrees celcius. Why doesn't anybody use Farenheit? Both cards are nVidia based.

I had a lot of crashes recently when one particular program was running, which supposedly was their latest and greatest. I still need to use it a bit more, sio I guess I'll just have to grin and bear it. It's a program by a vendor to program their touch panels. It should be harmless.

I'm the only user on this system, so no multiple logins. I've noticed that when I shut down BOINC, my CPU utilization still hovers around 79%. When the system gets into trouble, nothing seems to work in a reasonable amount of time. I had some spyware once that really slowed things down, but I can't find any this time. Spyware Detector and Spybot both indicate a clean system. Kaspersky has been unable to find a virus, so I'm at a loss for the high cpu utilization. Task manager shows me nothing using major resources.

I know I haven't dusted out my case in about 6 months and it's definitely due, but gpu temps seem to be what they have always been. Same with cpu temps. I'm using SIV to monitor temps and whatnot. A very cool program by the way!

I do overclock the cpu a little, it's base is 2.4 ghz, and it is now running 2.72ghz.

Still though, the video crashed 3 times this morning already and it seems to take a fair amount of time to recover. I'm using the latest (afaik) nVidia driver, 260.99
20) Message boards : Number crunching : BAM question (Message 11323)
Posted 2541 days ago by mdawson
Thanks for the info guys. BTW - love that "shocked" smiley face


Next 20

Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.