ah sorry! i made a mistake in explaining it to you because i read your settings too fast. do you have AIM or something we can live chat? i'll try to give you live help lol
Printable View
ah sorry! i made a mistake in explaining it to you because i read your settings too fast. do you have AIM or something we can live chat? i'll try to give you live help lol
yeah I had that all set correctly then, thanks for working it out with me. I guess that just means I have bad luck pulling available units.
I wonder if having 2 of the 480 GTX cards running at the same time is causing my 40% cpu usage? I'll try removing one tonight and see what happens when BOINC only recognizes one card. I'll also try turning HT off.
Using EVGA Precision instead of Afterburner to overclock the cards had no effect on improving things. GPUz shows that my overclocked speeds are holding while running GPUgrid.
I'm running out of ideas at this point. My work units are still taking almost 1100 seconds to complete.
i'm not really sure what else to say other than maybe try detaching GPUgrid from BOINC, and reattach it and restart BOINC to see if you can download the WU's. the only difference between your system is that you're on linux, so that might be the cause of it all. i'm going to try linux on these machines later to see if it works or not, and i'll let you know.Quote:
Originally Posted by pwolfe
how many WU are you seeing crunch at a time? it should be two right? try to disable the SLI and have the gtx480 as two separate cards. it may be your drivers. i noticed that you're on 197.55 (probably got that from the CD it came with?), and i'm on 197.41 downloaded from the nvidia website. HERE's the link. also to be on the safeside, go to you cc_config.xml file in C:\ProgramData\BOINC and have these lines in there:Quote:
Originally Posted by Danger30Q
<cc_config>
<options>
<report_results_immediately>1</report_results_immediately>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>
i could not get win7 to recognize multiGPU to crunch unless i attach a monitor to each card. you can put a dummy plug into one of them if you don't want to have two monitors.
another thing that worries me is I'm still using 195.36.18. 195.36.24 (official fermi driver) hasn't hit the ubuntu repositories yet and the one ppa i installed a dev version from hosed my system.
Thank you much WhiteFireDragon, I'll try out all of your suggestions tonight.
I wouldn't worry about not getting any WUs at the moment. Beta WU that is required for Fermi GPUs are quite scarce for now. I was getting some yesterday morning but they seems to be dried up as of now, GPU running idle for a while... :(
GDF over at GPUGRID forums just posted:
No idea what it means though have asked if it meant we can now start accepting non-beta work units.Quote:
The beta has been promoted to the application acemd6.73 cuda3.
gdf
Well I set up my pref's to pull all WUs and promptly error'ed out 6 units. So I'm done for the day for sure.
I'm now pulling beta work units just fine. GDF said they ran out of beta work units and just replenished the supply pool.
To get the release version 6.73, which are full size production WUs, select ACEMD in your GPUGrid preferences.
DO NOT select ACEMD2.
Same question posted on their forum; Did you accept "Run test applications" in the preference menu?Quote:
Originally Posted by Snow Crash
i just updated my preferences to run the new WU's, we'll see how it goes after these two beta WU's are finished.
for the new units that are not beta, i believe you don't need to , but just to be on the safe side, i would leave it as "yes"Quote:
Originally Posted by eternal_fantasy
Keep us posted on how the points are looking. I have my 480 just sitting on my desk running a 295 till they up the point values.
Selected only the ACEMD preference and accept test applications, I'm only receiving Beta test units, but not selecting it I get nothing when the manager requests new tasks... Have left it on for now so the GPU have something to do :(Quote:
Originally Posted by WhiteFireDragon
i finally just got the new v6.73 (cuda30) WU about an hour ago, with the "test application" checked yes. at first, i didn't get anything either, so i guess you'll just have to wait a few hours for something to update. i can't see how fast it's actually crunching until i get home from work.Quote:
Originally Posted by eternal_fantasy
if you didn't already see it, skgiven on the other forum gets about 60k/day on his tweaked gtx470 with these new WU. our 480's should get a bit more. this is a HUGE improvement over the 32k/day with the beta WU's.
this is good to hear, hope they can make it even better though:shrug:Quote:
Originally Posted by WhiteFireDragon
So how do you get these 6.73 work units are they ACEMD, ACEMD 2.0, or Beta?
they are the regular ACEMD, so make sure on you set that profile on your rig with the fermi card. i'm letting both the ACEMD and ACEMD beta run in case the WU pool from one category runs out. you should get much more points per time of run for the new 6.73 WU's though.Quote:
Originally Posted by =[PULSAR]=
Starting to look nicer than 295's I was a little apprehensive about selling them soon but if this keeps up it'll be worth it :D
well don't get your hopes up too much yet, those numbers were only from one person's results on a 470, and it finished a WU in 4.8 hours. my 480 should be a little faster, and it has been on for 4 hours already, still not finished yet. i can't see how far the WU is until i get home. i though it'd be done by now...
Well going to wait 2 hours for the current WU to finish so I can change cards. Looks like he is running xp with beta drivers also...
Looks like he also just returned a WU with his shaders OC'ed and averaging 66.5k ppd.
Just checked only ACEMD and it is still sending me beta 6.23 hmmm....
mine's at 72.7% complete in 5h 20min, meaning the WU will complete in 7h 20m. hmm... not looking good. at this rate, i'm looking at 39k/day, which is still an improvment from beta units, but not by much. i guess biggest factor here is OS and drivers. i'm on win7 64, i7-930 with SWAN_SYNC enabled, but also running WCG fully loaded, and a slight overclock of 765/1570/1850.
ryan, you may need to wait a few hours for the older WU's to get flushed out or for gpugrid to recognize your updated preferences :shrug:. i was still getting beta units for a little bit until a few hours later. or maybe restart BOINCQuote:
Originally Posted by =[PULSAR]=
Are you talking about the queue thats on the computer? because I already aborted all the other WU's. Well either way I'll leave it running overnight to see what happens but I don't think I'll get a piece of PIE tomorrow.
Just last night I was able to pull one 6.73, which error'd out in ~10 seconds, then was back to beta units. Then hour or so later of betas, another 6.73 popped up, which has been running for 5:32, at 94%, so should be completed with a total time of 5:53. This is on a PC running at 4.16Ghz 980X, HT ON with 12 WCG work units. 480 at 800Mhz core, 1900Mhz Mem, and with 57% utilization crunching the 6.73 WU.
I've only selected the "ACEMD" application, with the "ACEMD ver 2.0" and "ACEMD beta" unticked.
If you only want the new 6.73 WUs you can opt out of "Run test applications", but be warned that the new WUs are quite scarce, so your manager may need to spam the GPUGRID server for a while before getting a unit to crunch. I selected "yes" to "Run test applications" to receive some beta test units to keep my 480 busy whilst waiting for one of the new units to show up, which seems to be around an hour or so of betas.
Quote:
Originally Posted by WhiteFireDragon
once you deselect the 2.0 and beta ACEMD applications, "no" to "Run test application", and "If no work for selected applications is available, accept work from other applications?" to "no", you will not be getting anything else other then the new 6.73 units. But as I have said above, they are scarce at the moment so be prepared to wait a while whilst the GPU sits idle for the new WUs.Quote:
Originally Posted by =[PULSAR]=
What you are already crunching/waiting to be crunched on your computer is irrelevant to what work units you will get next.
Also, my preference set over on GPUGRID seem to be almost instant, once set and selected to "update" the GPUGRID project in the manager.
Just got a 6.73 WU and basing on the 5% mark looking at 40k ppd. Thats only running at 0.10 cpus on win7 64.
And it is done;
2284093 1440600 5 May 2010 23:52:27 UTC 6 May 2010 6:08:25 UTC Completed and validated 21,265.93 20,168.32 7,954.42 11,931.63 Full-atom molecular dynamics v6.73 (cuda30)
Total Run Time 05:54:25, 11,931.63 credits, so around 48476 PPD just with this one work unit... however it seems I'm back to crunching betas for now. :(
edit: Ah sweet just downloaded another 6.73 :cheer:
How much is your cpu set to? Do you have a dedicated core for it?Quote:
Originally Posted by eternal_fantasy
I have the full CPU thread setting configured. I don't have a dedicated core for the GPUGRID, as there are some downtimes with WUs from GPUGRID drying up for periods of time, and don't want to waste CPU cycle.Quote:
Originally Posted by =[PULSAR]=
However I have a lower priority set for the WCG, at 80, so when I do have something to crunch GPUGRID gets priority on one of the cores, while one WCG WU thread waits with 0% CPU utilization.
Okay just just changed the SWAN_SYNC=0 and running 12% instead of the typical 5% lets see what happnes :D
mine finally just finished, it took 7h 20min (26,452sec GPU, 26,158sec CPU). i'm not sure why it's significantly lower than everyone else's. it's slightly overclocked, and i dedicated a whole thread to it with SWAN_SYNC too :(
btw, congrats ryan, you made it to the first page of our team :eleph:
WFD I think it because me and you are the only ones using official drivers if you look everyone else is running betas :confused:
EDIT: First page of what? lol
alright, time to switch drivers to see if there are improvements. i'm just going top copy etherna_fantasy's and use 197.44 unless there's any objections. it's weird how official nvidia driver is 197.41, which seems older than 197.44. so where can i get these "official" older drivers without having to DL from some untrustworthy site?Quote:
Originally Posted by =[PULSAR]=
first page, as in top 20 of our gpugrid team :clap:. i'll meet you there in a few days :D
http://img.photobucket.com/albums/v2...th_1stpage.png
I would just go with 197.75 found here. Seems like some people are having goodluck with them.
http://www.nvidia.com/object/win7-wi...7.75-beta.html
I'm waiting till tomorrow to see how this first 6.73 WU does. They need to fix that OC bug for Fermi's I can fold all day at 850 core but it would be pretty beneficial to unlink the shaders and raise them more :D
I downloaded my 197.44 driver from HERE, which is Nvidia's official OpenGL 4.0 driver, provided by Nvidia's official site, which is not supported by any other version.
Just gave the beta 197.75 drivers a shot my 6.23 beta went from 590 seconds down to 497 so I would call that a vast improvement.
EDIT: was looking at skgiven's stats and showed a huge difference.
v6.23 beta
XP x86 - 325 seconds
win7 x64 - 600 seconds
That pretty much sums it up.
i tried the 197.75 beta as well. the v6.73 WU's when from 7h 20m down to 7h exact, and the v6.23 beta WU's when from 685sec down to 650sec. that's about a 5% in time reduction on both. still not as good as i wanted, but better than nothing. i'll be trying out the openGL 4.0 driver later to see if i get even more improvements.
yeah i mentioned earlier on skgiven's results on his gtx470. he's getting 4.8 hours on XP with v6.73 WU, while my 480 gets almost 8hours on win7, and only difference is OS and drivers. for people crunching with both WCG and GPUgrid, there's a dilemma of which OS to use. XP gives much better results for the GPU, while win7 gives better results for the CPU. which to use :shrug:? how does linux compare to XP for gpugrid?
I still don't understand how you guys are able to pull WU's. I mean I can't possibly be that unlucky to have not pulled a single beta or acemd unit this whole time. I wish some of you guys ran Linux so we could compare notes. At this point I'm assuming there is something wrong with my setup that is preventing me from getting a WU. My gpugrid pref's are identical to all of yours and I set this machine up with a clean install exactly as I have many times before. Frustrating.
i think i might know why. which drivers are you using? the profile of your computer with the 480 doesn't show which nvidia driver it's using, so if you're using generic drivers, that might not work. i think when i used the drivers that win7 auto-installed, it could not crunch because i believe it didn't have CUDA drivers.
if you're having this much trouble with linux, i would just switch to win7 or XP (preferable XP since you get almost double the computational advantage) so your card has something to do. it's one expensive card to just be sitting idle.
I haven't been able to get any new work units today but my time is still around 1100 seconds for my GTX 480s at 775/2000. I updated to the 197.75 beta drivers and BOINC tells me no GPU detected. I really haven't had time to seriously troubleshoot these problems but hopefully I can tonight or tomorrow assuming I get more work units to test.
i hate babysitting my these new cards too, and without the application fully supporting the cards yet, i can't afford to keep mine. i'll be listing both of them in the FS section soon, so if any of you guys want it, PM me and i'll sell it a little cheaper than to other non crunchers.
:(
Which cards are you selling? Don't give up just yet :D
i'll probably be selling one, maybe both of my 480's. i'm glad i left one of them unopened. the numbers on these cards are too low for how much i spent on them. i'll be replacing these two cards with an army of gt240's
I'm using 195.36.24, the latest stable Linux drivers, BOINC has never reported my driver version correctly, even when things are humming along smoothly.
How about a trade for a gtx 295 :p:
Hmm.. I'm currently crunching a 6.73 which looks to be complete just over 5 hours... :eek:
It is currently at 75% and 3:54 hours in... I'm running Windows 7 64-bit too, but I disabled Hyperthreading, and only allowed WCG to run on 5 cores, with one left for GPUGRID, which when the 2 are running the CPU is at 100% utilization for all 6 cores.
It's interesting that the CPU plays a much bigger part then the GPU for these WUs at the moment. :D
Mine finished in 5:57 looks like the extra cpu percentage really matters. What kind of clock were you running on your cards?Quote:
Originally Posted by eternal_fantasy
800 Core, 1900 Memory. However seeing the GPU usage is only around 60-65%, I'm not sure how much it matters at the moment, at least until the code is more optimised to be crunched on Fermi GPUs.Quote:
Originally Posted by =[PULSAR]=
I do wonder how upping the CPU speed would affect GPU usage, as it seems is what's bottle-necking the GPU at the moment...
The work unit that completed just over 5 hours;
2287511 1443883 6 May 2010 13:13:18 UTC 6 May 2010 23:13:40 UTC Completed and validated 18,799.97 18,055.79 7,954.42 11,931.63 Full-atom molecular dynamics v6.73 (cuda30)
I found out why my SLI GTX 480s were taking on average 1100 seconds to complete the beta work units. I turned off SLI and that did the trick. My last few work units have completed in 700 seconds at stock speeds. Oddly, it still shows my gpu clock speed as .81ghz. I've noticed other users will get this to 1.4ghz or so.
My gpu usage is around 55% for both cards.
is this a crunching card only or will you be playing games on it also? if you're only going to crunch with it, only the shader clocks matter, i just leave my memory speed at stock. i wish i could leave my core clocks at stock too, but too bad these fermi cards can only have linked shader and core clocks.Quote:
Originally Posted by eternal_fantasy
65% GPU load looks like good numbers. i'm only getting 50% load, and as i overclock the card, that number drops a little further. my guess is because the card can only do a certain number of work, so a faster OC'ed card doesn't need to work as hard :shrug:
i mentioned you need to disable SLI a while ago :p:. and my gpu usage sits at 50% only. perhaps if i OC my CPU more, this number can go upQuote:
Originally Posted by Danger30Q
what version of the boinc are you all using?
*oops, double post*
it looks like you've switch your cards from linux to win7? are you getting WU's now? i just installed ubuntu on the same computer, but could not get my wireless adapter to work with linux. i don't have time to troubleshoot it, so back to win7 :(.Quote:
Originally Posted by pwolfe
based on the one's who've responded to this thread, there are only 6 of us with fermi cards. you can view each of our profiles to see what cards we're using, how much points they're producing, and which nvidia driver and BOINC version is being used:
snowcrash (hidden)
WhiteFireDragon
=[PULSAR]=
Danger30Q
eternal_fantasy
pwolfe
I play for around 3-4 hours per day, 4-5 days a week on this PC, and when I do, I suspend the GPUGRID from crunching.Quote:
Originally Posted by WhiteFireDragon
The 800 Core/1900 Memory overclock is already hugely toned down, as I have gaming OC profile of 900Mhz/2200Mhz. I'm watercooling the card so It doesn't make much difference if I lower the Memory clocks even more.
As I said in the post you quoted, it seems that the GPU is limited in how much work it can do by how much work the CPU feeds it. This theory fits our observations, that when HT is turned off, and one CPU core is dedicated to feed the GPU, instead of one HT thread, the GPU usage increases, which in turn reduced the time to completion. In your case the CPU is already bottle-necking the GPU with a set amount of data, increasing the GPU Core speed just means it requires even less work to crunch what the CPU is feeding it.
I've currently turned HT back on for this Supercomputer Week, but once it's done I'll turn it back off, and see how much the CPU speed affects the GPUGRID work units by reducing my overclock, which I suspect plays a significant role.
I do have you to thank WhiteFireDragon for turning SLI off. I actually had a long list of planned changes to make but then I was stuck without a work unit for quite some time. Now I need to get in the habit of turning SLI back on before I game.
Quote:
Originally Posted by WhiteFireDragon
Thanks for the links and the help. I upgraded to BOINC 6.10.52 just to see if I could pull a WU.
I think theres nothing more to it than they have not uploaded any Linux compatible WUs. As far as the windows 7 thing, I just booted into windows to see if it would pull a WU, but as it turned out there were none available on the status page.
I'm just gonna leave well enough alone and not even attempt to think this through any further until I see gdf say they have uploaded the Linux batch of WUs. He normally posts such things when they do. Until then, I'm tired of stressing over it.
Can I run a GTX 295 and (2) GTX 480s on a EVGA Classified together?
I have all 3 cards and I'm about to put the GTX 295 up for sale but I thought I would crunch it along with the 480s until it sold, assuming it works with GPUgrid.
yeah i don't think there are any WU's for linux, but there should be plenty for win7. occasionally, you might run dry, but it's odd that you never got any of them. i've been getting a steady stream of them for a while now, so there's gotta be something you're not setting correctly.Quote:
Originally Posted by pwolfe
nope, most likely won't work since 480/470's use drivers designed only for themselves, while all the other cards use the same nvidia drivers. i tried this on my gtx480 and gt240 and it was a no go. you could always try it and report back here, there might be a slim chance it works.Quote:
Originally Posted by Danger30Q
I have tried a 295 and 480 on a classified and it does not work.Quote:
Originally Posted by Danger30Q
Thanks for confirming my suspicions that a GTX 295 and GTX 480 won't work together.
197.75 Official Drivers are now released.
EDIT: Downloaded and installed, testing now will report back tomorrow when 6.73 WU finishes.
I found this thread about different BIOS versions on EVGA's forum:
http://www.evga.com/forums/tm.aspx?&m=347659&mpage=1
Has anyone a card with the new BIOS?
From what I'm reading there seems no performance gain, but the voltage reduces -> possible cooler cards.
Well I went against my principles and am runnng the grid on my windows partition for the time being, just to see how many ppd I can generate until such time they decide to release the linux app. So far so so good, no errors.
But could also mean not as high of an overclock ;)Quote:
Originally Posted by mreuter80
I am using the newest one but no see overclocking voltages loss/gains.
70.00.21.00.82 from an EVGA SC
If i get boinc to run on this windows 7 x64 , will test some at 850+, keep get cant connect to client if dont install as service.
Pfff boinc .exe just crash on start with an access violation :(
it's crazy how unoptimized these cards are for win7. i got nearly a double in point production by switching to XP. on XP, each WU takes about 3h 45min to complete, HT off, swan_sync=0, mild overclock, but also loaded at 100% with WCG.
GUPGRID currently utilising almost 80% of my GTX 480, With my 980X at 4.32 Ghz and HT Off.
http://gallery.me.com/jeffrey_lee/10...20ulti/web.png
I think they might be optimising more for Fermi now, and I found the utilisation of the GPU very work unit dependent.
As this is my gaming machine, I have to stick with Win7, so hopefully it will get more optimisations.
Good to see how much cpu usage is ACEMD using?
It seems that the WU you see in the screen shot above are set to complete in 2 hour 40 min... will be interesting to see if It will develop an error or grant full points, because if it did it would be very intriguing. :D
Pulsar, It is using one full core while WCG is using the other 5, so 100% CPU usage. :up:
ah it looks like you also got the new WU they just released for fermi. mine finished a few of these new WU's in about 1 hour 52min for 6,756 points each. at the rate mine's going, that's 87k points per day :eek:. seems like these new WU's are more optimized?
Just got one of the new WU's also looks to complete in 2:45, GTX 480 owners say hello to owning the PIE :D
So that WU finished around 2h37m, which works out to be around 61800 PPD. Hope there's room to improve even further. :D
Whats your shader running at eternal?
1600Mhz
if you can maintain this or better; then we are making good progress:up:Quote:
Originally Posted by WhiteFireDragon
In win7 they are about on par with a gtx295 now...I guess that is good.
My gpu usage is around 70% compared to barely 60% a few days ago. It seems like the work units are a bit more optomized.
whats a gtx295 produce? approx 60k ppd. I think these need to produce around 100 or so to be on par with the initial claim of 3.5x gtx 285...or was it 2.5x:shrug::rofl:Quote:
Originally Posted by =[PULSAR]=
if you have an extremely tweaked 480 in overclocks, OS, and running by iteself without WCG to hog resources, you can get close to 100k/dayQuote:
Originally Posted by INFRNL
A overclocked 295 is right around 60k a day. Yeah the initial claims was for a 480 to outperform the 285 by 3.5x. I'd be really happy if we ever saw that :)Quote:
Originally Posted by INFRNL
thanks guys...it is definately looking better, but i thnk i am also liking WFD's results on his 240's cheap cards after rebates/cb and very good results for the price:up::yepp:
optimizations are not over for fermi. i bet there will be some large improvements both from gpugrid dev team and nvidia's compiler(cuda3.1 awaits!)
i agree, just thinking; how much more will they get:shrug:Quote:
Originally Posted by Chumbucket843
I realize some people bought Fermi cards solely for folding and may be disappointed thus far with the results. I primarily game with my 480s and so any folding I do while not gaming is a plus. I'll definitely be sticking with my 480s for quite a few months.
from the GPUGrid forum ...
I will release today the beta for Linux.
Promised.
gdf
:clap:
pwolfe ... please post up your deets when you get this running :up:
Anyone else getting new 4 hour WU's? They seem to be the same point value as the 2 1/2 hour ones just longer :(
this would be me lol. i bought it solely to crunch with, and a little disappointed with the power consumption and output from nvidia's top card.Quote:
Originally Posted by Danger30Q
if i have time and can get linux working, i'm willing to switch from XP to linux as well to see if there are any noticeable differencesQuote:
Originally Posted by Snow Crash
are you sure it's a new WU or something just happened to you rig to make it longer? mine's still getting a consistent 1.85hour completion. check to see if yours down clocked, or check the actual WU nameQuote:
Originally Posted by =[PULSAR]=
I'm with WFD on this ... I have not seen any new timed WUs ... looks like one of your GPUs got kicked in the ascii yesterday morning ... sometimes the WU does not fail even if the driver has to reset (nice resiliency), which typically puts you into 2D mode.Quote:
Originally Posted by =[PULSAR]=
Finally able to crunch on Linux. While they are only beta units, they are completing in about 425 secs vs 8xx secs while I was booted into win 7. Very encouraged so far.
is this the same WU that gives 280.92 points each? if so, my beta WU finishes in 277 sec in XP. i guess XP > linux in this case?Quote:
Originally Posted by pwolfe
They are giving me 280.92 per. My best time was 402.22, so I guess xp is better.Quote:
Originally Posted by WhiteFireDragon
I blame driver maturity :D
:D pwolfe
I have ordered a gtx 470 today and i will put it on gpugrid. Look like i will be the guinea pig for gtx470 number.
actually my 2 cards are in fact 470s
and what is your ppd ?
hasn't run for 24 consecutive hours yet, we had a power failure last night.
I want to see some of these numbers too...Quote:
Originally Posted by pwolfe
And have the points stabalized for the 480s? I remember reading only 60k, then 87k...
Is that 87k stable? or just a projection?
my projection is about 88.3k, my actual looks like 87.7k:Quote:
Originally Posted by Otis11
http://img.photobucket.com/albums/v2.../gtx480PPD.png
I guess that the GTX480 can yield a higher score than what a 5870 can do in Milkyway@home when its full potential is used for GPUGRID.
By the way, how much can a 5870 do in Milkyway@home?