Anyone else notice that the gpu usage is now at 80% :)
Printable View
Anyone else notice that the gpu usage is now at 80% :)
But the real question is, is it actually any faster at crunching work units with higher GPU usage?Quote:
Originally Posted by =[PULSAR]=
However, I have found that by suspending and restarting GPUGRID, the GPU usage percentage can change by up to 10%. So I believe it also depends on the conditions of your computer when you start the work units and not just the optimisations for Fermi GPUs.
I have to agree with eternal_fantasy because I've seen a similar variation in GPU usage even on the same work unit after starting up BOINC after a gaming session. I'll play with it some more and post back with any conclusive findings.
On another note, does XP or Linux give better performance with Fermi cards? I plan on creating a dual-boot with Win 7 for gaming and all other PC tasks and switch back to XP for better folding performance.
From pwolfe's 2x470 on Linux numbers of ~80k/day total it looks like XP wins (at least for now).Quote:
Originally Posted by Danger30Q
well guys, it's been quite fun with this card, but i just sold it. i couldn't justify the cost and power consumption for how much it outputs. it did however pull me up all the way to #6 in RAC, and how my production is going t plumet :(. i may pay this card a visit again in the future when prices drop, or if there are better revisions. i'll still keep checking back to this thread to see if they get more optimized WU's.
good luck with your cards!
So is it too soon to say what's the best bang for the buck? (480 vs 470 vs 295)
On a projected basis my OC'd 295 (shaders at 1602) can do 93+k per day.
I say projected because up until yesterday I was running 6.03 exclusively.
Currently 6.72 WUs are running me 12500 seconds per card.
This is on XP32 on an i7-920 HT ON clocked to 3.8 GHz.
I could probably squeeze a little bit more but this seems to be OK for now.
Still too many problems with 6.72 in general (especially the new KASHIF WUs)
to tweak these to their max.
I will not have twaeked numbers for my 480 for probably 2 more weeks.
Anyone else having issues running their card? My card is now at stock clocks and has been erroring out for the past 2 days straight. Not one completed WU is it possible we are about to see the first dead Fermi?
My 480 has been ticking like clockwork.
It looks like you are getting lots of different errors, I'm just guessing but I would try wiping and reloading the driver.
How many gpus are you currently running? Also what are they? thanksQuote:
Originally Posted by Snow Crash
I'm getting new work units, "KASHIF_HIVPR_auto_spawn", which completed in 8638 seconds, and granted 6642.02 points.
This is around 660 second faster then my fastest "TONI_HERGunb" work units, which is 7.1% faster, and only granted 113.59 less, or 1.68% less points.
1 x GTX480 Win7, not using SWAN_SYNC, i7-920 D0 HT ON, 4.075GHzQuote:
How many gpus are you currently running? Also what are they? thanks
1 x GTX295 WinXP, not using SWAN_SYNC, i7-920 C0 HT ON, 3.8GHz
I plan on switching things around to better match various pieces of hardware. I want to find out what this 480 can really do :up:
I am probably going to take the 295 off the grid and put my 285 in it's place(summertime is near) and then sell the 295.
here's a nice rumor. dual GPU fermi card gtx490, could potential give loads of points compared to gtx295. not sure how they're gonna keep it cool with a TDP of 375w. the gtx480 already runs in the 90's with a lower tdp.
:down: No more dual cards for me nothing but issues trying to run windows.Quote:
Originally Posted by WhiteFireDragon
No more dual gpu cards for me either. Lost one 295 to unknown causes and lost another when its EK waterblock cracked after 6 months of just sitting in my case.
So anyone noticing a difference with the new 6.05 WU's seem to be the same as the 6.73's...
Word of caution, 256.25 nvidia drivers break boinc on Linux. Do not upgrade yet! Now I have to find a ppa that has the previous driver. Rediculous. All the ppa's have replaced an official driver with a beta... sigh...
no difference, they are exactly the same ... they pulled an Nvidia ... the way it's meant to be renamed :rofl:Quote:
Originally Posted by =[PULSAR]=
Seriously ... they had seperate apps the old and the new, and then the fermi. They are consolidating so we stop asking why the app versions don't match the options in GPUGrid preferences. I'm pretty sure it is all acemd2 all the time. :up:
are these v6.05 WU's only for fermi? my rig with gt240's are pulling these WU's and erroring out :(. gotta wait a day before gpugrid lets me download anymore work.Quote:
Originally Posted by =[PULSAR]=
Maybe :confused: I'm no longer getting any 6.73 wu'sQuote:
Originally Posted by WhiteFireDragon
GTX285/ 197.45/ Win7 checks out fine for me.
There seems to be memory problems with OS / driver/ card combinations.
I think the new 257.15 drivers have giving a slight performance increase roughly 5% it looks like. I'll wait to complete a couple more WU's till I call this find proven.
People that are using gtx 470 are you able to get some wu validated at overclocked clock. With my evga gtx470,if i even overclock by 25mhz the shader i get wu error. It seem very strange.
Anyone know how long it takes to get new work units after you error out 2 in a row? Been testing my memory overclock, with 2200Mhz I can run artifact free in games, but errors out with anything over 2050Mhz.
Check out the projects tab and should have a deffered time next to the project. Usually the wait time is 24 hours.Quote:
Originally Posted by eternal_fantasy
Overclocking memory does not help much for GPUGrid so unless you really need to push it for gaming I would not spend much time trying to find your card's limit for GPUGrid.
Didn't check that, but I got new WUs after around 2 hours.Quote:
Originally Posted by =[PULSAR]=
I'm finding the highest stable memory clock for GPUGRID so I don't need to change the clocks every time I switch from playing games to crunching WUs. :)Quote:
Originally Posted by Snow Crash
Its not only memory overclock but also shader overclock, even if i just overclock it by 10 mhz it get me wu error.
It might just be the card will not handle an OC?
Have you tried going back to the 197.45 driver?
I've also gotten some strange behaviour with memory overclocking on my GTX480. The results are as follows:
1900Mhz Stable
2000Mhz Stable
2050Mhz Stable
2100Mhz Fail
2125Mhz Fail
2150Mhz Stable
2175Mhz Stable
2200Mhz Stable
2225Mhz Fail
It is now running 100% stable and error free with 850Mhz/1700Mhz/2175Mhz C/S/M. I would suggest you try jumping a few tens of Mhz with upped volts to see if you just have a bad range of clocks for your GPU.
Not overvolted yet ... (I'm posting this in an Xtreme forum??? what the he|| is wrong with me :shakes:)
but I just turned HT off on my WinXP box with a 480 (shaders@1632) and it looks to be turning the WUs VERY fast :sonic:
I'll post back when I get a result for a WU that was run entirely with the new set up.
<update1> It's going to be a little longer ... the first WU was one of the TONI_GA's which run shorter anyway and I don't have an old result for that type on this card to compare to.</update1>
<update2>OK ... There is almost no different under WinXP 32 between HT on and HT off when SWAN_SYNC=0. I am only seeing a minor benefit of 3-4 minutes per WU which I think may actually be because I cranked the CPU clocks from 3.8 to 4.0. While there was an almost 30 minute difference on the one _long task I ran, overall I think if you are using SWAN_SYNC then you can better crunch WCG by leaving HT ON.</update2>
I am really starting to get annoyed by that. I can only complete workunit at 1250 shader, anything higher with 1.087 volt will not work and thats without any memory overclock. I am maybe looking to step up from my gtx 470 and get a gtx 480
I've been running my 2 480 GTX cards in XP wit SWAN_SYNC=0 and I've been averaging about 7,400 seconds to completion on v6.05 work units. Both cards are running at 1450mhz on the shader clock.
7,400 seconds per unit with a credit of 6,750 points per unit works out to be about 78,800 points per day per card. I've been trying to find other users with a GTX 295 under XP and running SWAN_SYNC but I haven't had any luck. I may try turning off SWAN_SYNC for a day to see how my points change more in line with everyone else.
try the new 257.15 drivers, i've read good things about them for the 400 series cards.
I tried those beta drivers when my Fermi was still working, it actually dropped performance.Quote:
Originally Posted by WhiteFireDragon
ah really? i tried 3 different drivers: 197.44 OpenGL 4.0, one of the beta drivers (forgot which one exactly), and the official 197.75 driver. i got the best performance with the beta one. i really can't remember which beta driver it was though :(
The 197.75 has been the best driver so far, at least on win7.
is anyone getting better #'s yet? These should be getting well over 100k if not 120k.
How did you figure what it should get?Quote:
Originally Posted by INFRNL
My ppd is 83+k and still climbing but likely will top out around 90k. It is the top performing 480 on GPU grid, I have it over volted and overclocked running on XP.
not even close. i think they're more around 50k - 70k.Quote:
Originally Posted by INFRNL
what are your clocks at? and you should unhide your computers so we can see what you're running :p:Quote:
Originally Posted by Snow Crash
Quote:
Originally Posted by Snow Crash
I was going off of the initial claim of 2.5x gtx285. my gtx275 gets avg. 40kppdx3 would be 120kppd for gtx480. 2x is 80k avg.Quote:
Originally Posted by WhiteFireDragon
you can get or could get a gtx275 for 175-200 for now 40k. 2 would be $400 at most for 80k. 480 getting a lucky 80k like you for the cost;; not sure its worth it and not as good as they initially claimed.
I am glad I held off, but hope that all the owners are enjoying them. Also hope they optimize for the cards better for you guys.
they're slowly getting optimized, but not near the cost/PPD ratio expectation yet (which is why i sold mine). but it's sure better than the 32k PPD when the fermi WU's just started working.Quote:
Originally Posted by INFRNL
Been a frustrating day or so ... I decided to bring the monitor to the basement and do some maintenance
update the drivers
bump vCore to max with vNidia Inspector
get a little more from the shaders
bump my CPU up a bit.
Windows updates
JKDfreg
full AV scan
stopped a couple of services
... Yes I can see you all shaking your collectives heads ... way to many things at the same time. I know better but sometimes I get impatient. :rofl:After a few error runs I stepped everything back except the drivers but I continued to get errors.
I reversed the drivers, and bumped CPU back up again and I think I might be back into stable territory. :yepp:
1.125 for Voltage and shaders set to 1701, I'm keeping mem at stock because not only does it not help runtimes but it is also the place that fermi has the most problems with OC. :shakes:
:sofa:As for hiding my machines ... the truth is I get embarassed when I screw around and then you all get to see the errors I am throwing.
I'll go unhide them now but reseve the right to let my insecurities make me hide them again at some point in the future :D
Okay so after receiving my RMA the other day I installed the 257.21 drivers and was really disappointed with my card's clocks. Then saw stasio posted the 258.19 drivers gave those a shot and still bad clocks so I went back to the older 197.75. Wow clocks are way better able to get an extra 35mhz on the core 24/7 and it seems the gpu runs 10C cooler also. Just thought I would post my findings and tell you guys :up:
Does the Folding@Home MemtestG80 program work with Fermi cards? When overclocking my cards I found this was a much easier way to detect instability than any other method.
Yes it does run, but when I put in my mem/gpu clocks it only runs the @ 70% usage is that normal?Quote:
Originally Posted by Jacka
CUDA 3.1 just released.
http://developer.nvidia.com/object/c...downloads.html
:clap:Quote:
Support for 16-way concurrency allows up to 16 different kernels to run at the same time on Fermi architecture GPUs
CUDA 3.1 coming to GPUGrid
http://www.gpugrid.net/forum_thread.php?id=2209
Quote:
In the next few days we will update the Fermi applications to CUDA3.1. If you have a Fermi card, please update your drivers to at least 257.21 for Windows and 256.35 for Linux. No action is required from users with other cards.
Thats great news I wonder how much we will see out of it. Hope win7 gets a huge bump as I rather not revert back to xp.
I am going to switch my XP/ 480 over to the new drivers now to try and get a baseline before they release the new app. After that I will have to switch hardware around so if you want to wait it might be a while. Of course the timing depends on the stability of the new app and/ or how long it takes to iron out any major bugs. In the past it seems to take about 3-4 weeks from release to stability mainly because GPUGrid does not have a large local stable of machines to spin through the different OS/ driver combinations. I think we may not see too much improvement this round because GDF usually tells how much they have see in the lab, hopefully after this first port to 3.1 they can start working on optimizations. IIRC NVidia is still working with MS as most of the slowdown on Win7/ Vista is the new driver model at the OS level.
can anyone that still have these older fermi cards try installing the new 258.96 drivers to see if there are any improvement in overclocks or time it finishes the WU?
GTX480 750/1500/2000 runs through ACEMD2 WU's in just over 3:15 for me it seems. WinXP-64, no swan_whatever tweak, 8c/8t @ 3.44ghz (dual e5450), 4x2gb FB-DIMM.Quote:
Originally Posted by WhiteFireDragon
edit
overnight (no pc useage, 800/1600/2000 clocks) went through ACEMD2's in 2:57:58-ish