Posts by BeemerBiker
log in
1) Message boards : Windows : collatz cannot find MSVCP110.dll (Message 24529)
Posted 50 days ago by Profile BeemerBiker
Back again, been 9 months, thought I would try collatz again.

New system build, Windows 10 clean from scratch, 2015 VC runtime both 64 and 32 bit with 3 old nVidia gtx 670, 650ti, 650ti. Got same missing 110 dll. Detach and reattach, same problem. Installed 2013 and 2010 VC runtime, both 32 and 64. This time got different problem
    -1073741515 (0xffffffffc0000135) Unknown error number



about 50 or so work units error'ed instantly with 0 cpu / gpu time at least no "missing 110 dll" but that was because the instant error did not let the app get to the end of the crunch and then find out the 110 dll was not to its liking.

Upgraded from 388.00 to 388.31 nvidia, removed the 2010 and 2013 VC runtimes and repaired the 2015 runtimes and re-attaached. Now collatz crunches just fine for couple of hours then near the end of the crunch realize that 110 dll is not to its liking and hangs.

I used to crunch ths project on all my win7 machines but since upgrading to 10 have not had any success at all. If you check my task list most are aborts because there is no sense letting them finish if the first one cant fine the 110dll then none will. Some tasks show up as abandoned, not sure why because I definitely terminated them on purpose.

2) Message boards : Windows : collatz cannot find MSVCP110.dll (Message 23836)
Posted 344 days ago by Profile BeemerBiker
I started up collatz after a long hiatus on win7x64 and it crashed with the following error


I first tried copying the "missing" file from windows\syswow64 down to
ProgramData\Boinc\Projects\Boinc.thesonntags.com_collatz but it still could not find it. I then reset the project which caused a master download but that did not help either. Something is wrong. There are 2 of those dll the other one is named MSVCP110_clr0400.dll and both of them are in that syswow64 directory.

if these are 64bit dll's maybe I am missing the 32bit one??

TIA
3) Message boards : Number crunching : How long should 3.13 with cuda 4.2 take? (Message 15967)
Posted 1811 days ago by Profile BeemerBiker
I switched to primegrid for a while and observed the 9800gtx took 5000 seconds and the gtx570 took 991 for the same credit of 3371. This is normal according to my other gtx570 systems. However, I noticed a Primegrid task that took 3880 seconds and looking at the task details I observe that it started on the gtx9800 and finished on the gtx570. I didnt think that was possible, but it was validated as correct.

I also discovered that I had left speedstep enabled, a mistake. Instead of multiplier of 12, the multiplier was 10 and probably a lot less when the monitor was off and the system left alone. Possibly colatz got starved for data as there were 3 instances of Poem running. However, I did have 1 cpu unassigned so collatz should have had that to itself.

I have disabled the speedstep and the multipler is were it is supposed to be. So far, I have not seen a big jump in % progress after correcting the speedstep problem.
4) Message boards : Number crunching : How long should 3.13 with cuda 4.2 take? (Message 15959)
Posted 1812 days ago by Profile BeemerBiker
I am seeing 8 hours for either a gtx570 or a 9800gtx+

Other systems with gtx570 run 2.03 and cuda23 and take anywhere from 24 minutes to just over 2 hours.

OK, I can see that 9800gtx+ takeing a while, but how can gtx570 take the same 8 hours?




    13 1/29/2013 9:45:04 PM NVIDIA GPU 0: GeForce GTX 570 (driver version 310.70, CUDA version 5.0, compute capability 2.0, 1280MB, 8381568MB available, 1405 GFLOPS peak)
    14 1/29/2013 9:45:04 PM NVIDIA GPU 1: GeForce 9800 GTX/9800 GTX+ (driver version 310.70, CUDA version 5.0, compute capability 1.1, 512MB, 464MB available, 705 GFLOPS peak)
    15 1/29/2013 9:45:04 PM OpenCL: NVIDIA GPU 0: GeForce GTX 570 (driver version 310.70, device version OpenCL 1.1 CUDA, 1280MB, 8381568MB available)
    16 1/29/2013 9:45:04 PM OpenCL: NVIDIA GPU 1: GeForce 9800 GTX/9800 GTX+ (driver version 310.70, device version OpenCL 1.0 CUDA, 512MB, 464MB available)
    17 1/29/2013 9:45:04 PM Config: use all coprocessors

5) Message boards : Number crunching : CUDA42 WU's "Marked as Invalid" (Message 14624)
Posted 1976 days ago by Profile BeemerBiker
I have the same problem: The cuda42 are all invalidated but the cuda23 (mini collatz) all work fine.


Compared to wingman, my gtx570s (a pair but not in sli mode) take 2x as much time time as, for example, a slower gtx440

Collatz runs fine on my HD5850 systems so I am suspending collatz on all my gtx570s till the problem gets fixed.

I do see a lot of waiting for validation on my gtx460 systems but there are no invalid results for cuda40 unlike cuda42.
6) Message boards : Windows : Lot of ATI errors (Message 13574)
Posted 2147 days ago by Profile BeemerBiker
I found 11.6 and that seemed to solve the problem. FWIW, my HD5850 handles the 12.1 (1.4.1664) just fine but the older HD4890 clearly had problems with 12.1


    CAL ATI RV770 (1024MB) driver: 1.4.1417
    Microsoft Windows Vista Home Premium x64 Edition, Service Pack 2, (06.00.6002.00)


    NVIDIA GeForce GTX 460 (1024MB) driver: 28562, CAL Cypress (1024MB) driver: 1.4.1664
    Microsoft Windows 7 Home Premium x64 Edition, Service Pack 1, (06.01.7601.00)



I am running the 4890 at lowest settings as I have not gotten around to replaceing the fan. It was a real PITA on my 5850 as I could not find an original OEM.

7) Message boards : Windows : Lot of ATI errors (Message 13565)
Posted 2148 days ago by Profile BeemerBiker
I looked at the event logs and the system has been entering sleep every hour. Maybe this is the problem. I just set the system to high performance to keep it from sleeping. If this does not work then I will find 11.whatever. I cannot roll back to 11.x, as this was a rebuild after a motherboard swsap and I put in the latest 12.1 after booting the original vista.
8) Message boards : Windows : Lot of ATI errors (Message 13561)
Posted 2149 days ago by Profile BeemerBiker
Not sure what is going on. I have two systems running ATI 12.1. One is HD5850 and runs fine. 27 valid results since I put 12.1 in. My other ATI system has the HD4890 and has 22 errors and only 3 valid results during the same time.

The error is access violation reading from location 0x00000190

http://boinc.thesonntags.com/collatz/result.php?resultid=109223673

I only discovered this problem after I realized I had a problem with almost ALL of my primegrid wu's erroring out on both ATI systems. Staff at primegrid states a bad bunch of WUs were recently released. I assume this is just a coincidence, but the errors from my two systems (over 100 each) show access violation reading though a different location: 0x10 (in addition to a lot of timeout errors).

http://www.primegrid.com/result.php?resultid=353228294

tthrottle indicates temps have been under 77c for 24 hours and I run the 4890 at the lowest gpu and memory clock (underclock to minimum).
9) Message boards : Number crunching : Problem with dissimilar GPU's ? (Message 13404)
Posted 2178 days ago by Profile BeemerBiker
I recently added a gtx 570 to a system that had a gtx460. I was wondering if the following error indicates a problem when starting with the 460 and finishing with the 570. Could there also be a problem where the "No checkpoint data found" is given? Something resumed or restarted but there is no statement to that effect. Later on, there is a "Resuming from a checkpoint" and the program then got an error.

from messages:

    GPU 0: GeForce GTX 570
    GPU 1: GeForce GTX 460



from the task information after the upload:


    <core_client_version>7.0.3</core_client_version>
    <![CDATA[
    <message>
    - exit code -1 (0xffffffff)
    </message>
    <stderr_txt>
    Running Collatz Conjecture (3x+1) CUDA GPU application v2.01
    based on version 1.2 by Gipsel
    instructed by BOINC client to use device 1
    Reading input file ... done.
    Checking 824633720832 numbers starting with 2373659816691222948200
    No checkpoint data found.
    Running Collatz Conjecture (3x+1) CUDA GPU application v2.01
    based on version 1.2 by Gipsel
    instructed by BOINC client to use device 0
    Reading input file ... done.
    Checking 824633720832 numbers starting with 2373659816691222948200
    Resuming from checkpoint ... done
    cudaSafeCall() Runtime API error in file <c:/Source/Collatz/CollatzCudaKernel3.cu>, line 513 : unknown error.

10) Message boards : Number crunching : Bad WU's? (Message 7995)
Posted 2797 days ago by Profile BeemerBiker
This just happened to me. I had two wu's in a dual 9800gtx+ box running xp-32 pro that take normally 30 minutes showing that almost 24 hours had elapsed. I rebooted and they finished within minutes.

There are no gpu nor cpu restrictions. Running 6.10.55. Temps were down in the low 50's, a sure sign that neither gpu was crunching.

What is strange is the following:

Before rebooting I very clearly saw 23:xx:xx(00:00:xx) for one of them. I dont remember the value of the xx, just the 23 hours and the low temps. This indicates 23 hours elapsed time and less than 1 minute cpu time (Elapsed time, boingtask column)

After rebooting, the tasks finished within minutes and are "ready to report" with the time for each one showing

00:35:41 (00:00:17) and
00:35:46 (00:00:18)

so what happened to the actual elapsed time of 23 or so hours? I think collatz was hung in the gpu and the time counter was not incrementing. It would seem to me that elapsed time should have been reported correctly and not the 35 minutes. Something is wrong here

I just did an update that release the two to the project. I went to the project and the run time is about the 30 minutes, Not anywhere near the almost 24 hours that had actually occured.
11) Message boards : Number crunching : CUDA tasks erorr out after suspend / resume (Message 7647)
Posted 2819 days ago by Profile BeemerBiker
I lost 300+ collatz tasks as shown here when I resumed collatz. I had been cherry picking some projects using 6.10.43. This is what happened.

Milkyway ran out of work late yesterday. I then removed the block for no new tasks for collatz but nothing happened even after an update. All I saw was "no ati gpu" and this system has a pair of nvidias. On a hunch I suspended aqua which then got me some collatz tasks. Aqua was using 3 cpus of the the 3 allocated for boinc. I had one cpu set aside for feeding the gpu but I recall that does not seem to work all the time when starting up with no gpu tasks. I also suspended collatz while I made some other changes.

Anyway, I quit watching after the first collatz task started running when I did the resume, and I did not bother to check for another hour or so and that was when I found that the 2nd gpu was malfunctioning (?) and all the collatz tasks were being dumped. I rebooted the computer to fix the problem. This windows 7 system has been running just fine w/o error till now. The only thing i did recently was to add a dummy vga adapter as occassionally I boot headless and one of the gpu's stops working when I do that.
12) Message boards : Unix/Linux : cannot run collatz except at nice=19 in karmic (Message 7310)
Posted 2830 days ago by Profile BeemerBiker
SOLVED - I finally got collatz to run at nice=0 in Karmic which solved my GPU feeding problem. I also learned something (possibly)

I installed the recent (2.3 actually) linux cuda toolkit from nvidia. After doing ldconfig, the tool ldd ...collatz... showed that I was now using the 2.3 version and that seems to have got me nice=0. So - I am assuming the problem was using an older library, 2.2, that came with the collatz linux package. That one was showing nice=19.

13) Message boards : Unix/Linux : cannot run collatz except at nice=19 in karmic (Message 7300)
Posted 2830 days ago by Profile BeemerBiker
I have two almost identical systems, Opteron 270's, GTS250, 6.10.44. The only difference seems to be that collatz runs at nice=10 in Dotsch_UX (ubuntu 8.1) and at nice=19 in Karmic (9.1 ubuntu)

That makes a big difference as I have just discovered. BTW, I tried re-niceing to 10 but the next collatz task picks back up at 19. I do now know how to keep it at 10.

Dotsch_UX


Karmic





This runs Dotsch_UX (ubuntu 8.1) and nice is set to 10 for collatz. Other tasks are at 19. Note that all of the runtime average about 2100 seconds (same as gpu time) and the cpu time is about 30 seconds.

This identical system runs Ubuntu 9.1 and nice is stuck at 19. If I renice collatz the next task goes back to 19. If you look at the statistics you will see two somewhat good results, the rest are way up at 20,000. The reason that I got two fairly good results is because I set %cpu to 75 to free up one of the 4 cores for collatz. I would not have to do that if I could set nice to 10.

thanks for looking
14) Message boards : Number crunching : Running multiple GPUs (Message 7233)
Posted 2832 days ago by Profile BeemerBiker
A couple questions concerning multiple GPUs on AMD 790GX and 790FX MBs. I can run 2 ATI cards on these boards but of course can't run 2 NVidia cards at once. Is it possible to run 1 ATI card and 1 NVidia card at the same time? If so is there any special configuration or considerations to keep in mind?

Also have a 790FX board that has 3 double spaced PCIe slots. Using all 3 leaves the cards very little clearance between and the temps skyrocket. Any tricks to getting this to work other than liquid cooling?

Thanks in advance.


I cut a huge hole for a 200mm fan in the side on three cases, added forced air 120mm fans in the front (1 or 2 depending on the height of the case), alternate 2/3 length card with full length and installed Evga Precision to allow the fans to be run at 100% (not an option with linux unfortunately) ATI overdrive worked fine, unlike nvidia's performance.





I do not recommend this one

15) Message boards : Number crunching : ati13ati is burning up a full cpu, way more than cuda (Message 7214)
Posted 2832 days ago by Profile BeemerBiker
However, what I meant to report was the difference between Collatz 2.02 and Collatz 2.09 (the CUDA -vs- the ATI). I am guessing that the ATI version has the CPU polling to see if the ATI is finished while the CUDA interrupts the CPU to tell it is done.

No, the ATI app doesn't use a full CPU and it doesn't busy waits for the GPU (unless you tell it to do so by some command line parameters).
What you see in the task manager is real, I just had a look on your tasks and you are using roughly 2 seconds CPU time per WU. The difference between the CUDA and the ATI app is really that the CUDA app reports the CPU time, but the ATI app reports the GPU time. So in fact it is roughly the GPU utilization you see there. Look at the task details (or the task manager) to see the real CPU load for the ATI app. It is very low with the standard options. Kashi is completely right that it is a cosmetic issue.


Agreed, the CPU is not anywhere near %99. However, I disagree with the reasoning behind substituting GPU time for CPU time. I do not see any problem with milkyway and they return CPU time in the CPU field which you can see in the following statistics.



Substituting GPU time for CPU time is misleading. I do not see that being done on your cuda tasks as shown here
16) Message boards : Number crunching : ati13ati is burning up a full cpu, way more than cuda (Message 7212)
Posted 2832 days ago by Profile BeemerBiker
It's just a cosmetic issue.

The difference is that the latest version of the MilkyWay ATI application is reporting actual CPU time as BOINC CPU time. This has been known to cause problems for BOINC scheduling when the CPU time reported is much lower than the GPU runtime. It is more likely to happen with faster ATI applications that use a small amount of CPU time to process tasks rather than with CUDA applications. Therefore the Collatz Conjecture ATI application reports the GPU runtime as CPU runtime to BOINC. This means the CPU % calculated and shown by BoincTasks for the Collatz Conjecture ATI tasks is not correct.

"That is done because BOINC gets severly confused with the task duration factor and stops asking for work on some client versions if the returned CPU time is very low." from here


Yea - I have seen that TDF problem where I cant get any data unless I suspend all other projects and reboot.

However, what I meant to report was the difference between Collatz 2.02 and Collatz 2.09 (the CUDA -vs- the ATI). I am guessing that the ATI version has the CPU polling to see if the ATI is finished while the CUDA interrupts the CPU to tell it is done. That would account (if true) for what would appear to be a larger %cpu usage during ATI processing as compared to CUDA processing. That is just a guess based on what I read in a thread earlier.

I set the boinc %cpu available to 75% which freed up one of the core in my quad for collatz. That had no appreciable effect on wu's production for collatz and tells me that the 99% I am seeing is not real. That plus the task manager showing no cpu usage indicates something is amiss in the computation of percent cpu busy.

I took another snapshot, this shows collatz 2.02 (linux cuda), 2.09 (ati) and 2.03 (win cuda). Note that the ATI ones all show %99 cpu busy.



Note that 2.02 has %cpu of 0.020, 2.03 has 5.17% but 2.09 shows 99.64 average for 4 results. Compare 2.09 to casinoAlpha which is fully CPU intensive (QuantumFIRE) and has 2 tasks averaging 96.38%
17) Message boards : Number crunching : ati13ati is burning up a full cpu, way more than cuda (Message 7205)
Posted 2833 days ago by Profile BeemerBiker
I just started using boinctasks instead of boincview. One thing I noticed is that boinctasks shows the cpu% that is used by a gpu task. If I interpret the following images correctly then the ati13ati version of collatz uses %99 of a cpu while the "2.02 collatz" uses only 1.71% of a cpu.

As shown in the graph, the ati13ati collatz is highlighted in red (99.72) while it is working its way up and one that finished is shown at the top (99.52). Note that the CUDA collatz "2.02 collatz" is NOT a cpu hog and used only 1.71% of a cpu. That version was running on a gtx280. The ATI one was running on 5850.



In comparision, milkyway uses much less for its ati13ati version (7.42%)



I posted the above over a the boinctasks forum

[EDIT] - I just checked the vista task manager and I dont see a full cpu for collatz even though I see %99 in the above. I expected to see %25 if collatz was indeed taking a full cpu. I actually see 0.0 (zilch) which is not reasuring. Maybe collatz is not reporting the correct cpu utilization info or boinctasks in not calculating utilization correctly for collatz.
18) Message boards : Number crunching : CUDA tasks erorr out after suspend / resume (Message 7148)
Posted 2835 days ago by Profile BeemerBiker
Found out, one of my host's(Q6600+HD5770+EAH4850), errored about 240 WU's, cause, the extended-desk-top feature, wasn't active anymore (?)

Card 1=5770; 2= 4850, other projects (MW), f.i. has difficulties finding the card supporting Double Precision(4850), so discards it.

Running WIN XP x86 SP3 and BOINC 6.10.15.

Maybe an attempt to run Remote Desktop, stopped the 'original -driver', to respond.
Found an alternative feature Team Viewer 5 , which works OK.


Yea, I quit using remote desktop over a year ago when that problem was first discovered. Thanks for the info about Team Viewer. I have been using vnc which also works on linux.
19) Message boards : Number crunching : CUDA tasks erorr out after suspend / resume (Message 7143)
Posted 2835 days ago by Profile BeemerBiker
Collatz 2.03 (cuda23) 1270232359_491722_0 cancelled by me after 43 hours (weekend) processing 38.125% Running boinc-manager 6.10.43 - windows xp pro -
nvidia-driver 197.13 gts250. Normal run about 35 minutes!!

Any special reason??


I looked at your task result and didnt see anything indicating a problem. All I saw was 0 and 0 for the times and that abort by user. One of your wingmen returned a result. Is this system dedicated to boinc? ie: do you do any gaming? Did the progress bar move at all? Did you try suspending and resumeing that task?

By suspending and resumeing the same collatz tasks, I can (after several tries) put BOINC into a state where it thinks that both collatz tasks are running on the same device (which is not presently possible)



Even then, both task complete just fine. The above problem is discussed here but the problem is only cosmetic IMHO as both gpu's are used even though device 1 is not listed by boincmgr.
20) Message boards : Number crunching : [error] No application for linux and cuda (Message 6736)
Posted 2853 days ago by Profile BeemerBiker
The problems I had been seeing all went away when I put Dotsch in. Details here


Next 20

Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.