GTX260 gives ABYSMAL performance!

[manveruppd]

Here's a little problem that's bound to have even the best among you scratching your heads...

I recently upgraded from a Powercolor Ati 3850 to a BFG GTX260. For the first few days, I was having terrible stability problems with it: every time I would launch any kind of game the driver would crash, sometimes BSODing, sometimes dumping me back to the desktop in 320x200 with a "nv4_disp has crashed" error message, sometimes simply freezing. Some would show the little corporate logo vids and then crash, others immediately on attempting to launch.

At first I assumed it was a driver problem, so I uninstalled and reinstalled the drivers several times, trying several different versions, thoroughly wiping the previous one with Driver Sweeper before installing the next one. None made a difference. Then I dropped my CPU back to stock speed and all my BIOS settings back to default. None of that made a difference either. I monitored my rails, thinking my PSU might be insufficient, but they were rock solid. After a few days, I managed to stop it crashing by upping the PCI-E frequency to 101MHz and the PCI-E voltage by +0.1V. Every game I tried would launch and run, and wouldn't crash at all! Through experimentation, I found that, if I upped the PCI-E frequency to 105MHz I could even restore my overclock, bringing my CPU back to 3GHz, where it's been running for the past 2 years or so, and the system would still be rock-solid.

EXCEPT... all 3D games run SLOW! Not slow for a GTX260. Not slow compared to my previous 3850, but slow compared to the Nvidia 6800nu I was running up until 3 years ago! The system scores about 6,600 in 3D Mark 06, when every other PC with a GTX260 and a ~3GHz C2D gets 11,000 or more on Orb!! I'm talking 12-24FPS in games as old as Mount and Blade or Guild Wars! In fact, the only games which AREN'T running substantially slower than they did before are Saints Row 2 (which ran like a dog before anyway, so I really can't tell, and in any case is CPU-limited) and Burnout: Paradise, which alone of all the dozen or so games I've tested inexplicably shows an improvement!

I did all the usual things I could think of, tried different driver versions (thoroughly uninstalling and wiping the previous ones before installing a new one). None of them made a substantial difference. I'm currently sitting on the latest official Nvidia drivers, though the ones from a few versions back (186.18) gave me about a 5% better performance. I also tried running these games with RivaTuner and GPU-Z monitoring in the background. I've found that rails were steady under load, temps never went above 42, and the GPU was running at its max speed and (according to GPU-z) at 80-100% utilisation, so I know it's not being throttled or held back by insufficient bandwidth. My CPU would be at around 60-80% utilisation all the while (with the exception of Saints Row 2 which maxed it) so I know it's not a question of my being CPU-limited.

I've tried tweaking a bit more in the BIOS, making minor adjustments to timings and voltages, but none of it made any difference whatsoever. Apart from reinstalling Windows I've completely run out of ideas, and I'd really appreciate some advice! Any suggestion you can think of, no matter how outlandish, please throw it at me!

Specs: C2D E2180@3GHz on a Gigabyte P35-DS3 and 4GB Crucial Ballistix RAM, under Windows XP SP3. All other drivers and stuff are completely up to date, except for my BIOS which is F11 (latest is F12, but for some reason the Gigabyte auto-update utility won't download it - in any case, the only fixes in F12 are identification codes for new CPUs).

 

[Rroff]

I'm on F11 here for my P35-DS3R and the GTX260 in that PC works fine. TBH the simplest way to troubleshoot this rather sounds like your gonna have to reinstall windows - theres just so many variables with a dirty OS install. I'm assuming your PSU is upto the job?

 

[manveruppd]

PSU was one of my first thoughts, especially since overvolting the PCI-E bus stopped it from crashing, but BFG's website says this card requires a combined 12V rating of 38A or more, mine is well above that (3 +12V rails rated at 17A each). And BFG's minimum requirement is based on having a top-of-the-range quad CPU, I'm only running a relatively piddly dual core and 2 HDDs on here!

 

[manveruppd]

A bit more info from the BIOS:

Robust Graphics Booster: Turbo
CIA2: Disabled (anything else causes the crashes)
PCI Express Frequency: 105
Performance Enhance: Standard (anything elses causes the crashes)
RAM timing is by SPD
DD2 overvolting: +0.1V
PCI-E Overvolting: +0.1V
FSB Overvolting: +0.1V
Vcore: 1.3375V
CPU Enhanced Halt: Disabled
CPU EIST: Disabled

My CPU temps stay at 42C under Orthos, and I've never seen the GPU go above 42 when gaming while Rivatuner is logging either.

 

[varkanoid]

if the only item you changed was the 260 then is it not safe to say its faulty ? Plus bear in mind if anything else is plugged into the 12v rails then the available Amps for the card is diminished.

Can you not try the GTX260 in someone elses PC ?

 

[RavenXXX2]

I had many performance issues with a 260 and XP, drivers can be very temperamental with XP, suggest you move up to windows 7 ASAP.

Regarding performance issues, check CPU-Z or GPU-Z and make sure your bus interface/link width is at 16x.

 

[mame]

Best to try it in another pc. Was the card new?

 

[manveruppd]

if the only item you changed was the 260 then is it not safe to say its faulty ? Plus bear in mind if anything else is plugged into the 12v rails then the available Amps for the card is diminished.

Can you not try the GTX260 in someone elses PC ?

A faulty card is a possibility but I would very much like to rule out everything else first! There is one friend of mine whose computer I could try this on, but his PSU is only 450V so it would be VERY borderline whether it would handle this! I'll try and talk him into letting me hijack it tomorrow but I doubt I'll be able to get it working!

What else draws on the 12V rail? I know the motherboard does, but do the hard disks?

I had many performance issues with a 260 and XP, drivers can be very temperamental with XP, suggest you move up to windows 7 ASAP.

Oh, it goes without saying that I'm moving up to 7 as soon as I'm able to! That's why I've been trying to avoid reinstalling XP for so long! For now it's not an option though, and surely, however temperamental it is, ONE of all the driver versions I've tried should have worked by now?

Regarding performance issues, check CPU-Z or GPU-Z and make sure your bus interface/link width is at 16x.

Now that was a very interesting suggestion! GPU-z says "The graphics card reports that it supports PCI-Express x16 v1.1. It is currently running at PCI-Express x1 v1.1"!

However, it also points out that "power saving features may affect the results displayed", and I think that's what is happening here: when I was benchmarking, GPU-Z's hardware monitor reported that the GPU core was seeing 80-100% utilization at full speed (depending on the game), so it's utterly impossible that the bus was running at x1 - the GPU would have choked in insufficient bandwidth, there's no way it would have been able to see that high GPU load! Nevertheless, I'll launch some game or another and then task-switch to GPU-z just to check in a second.

 

[RavenXXX2]

Check with CPU-z in the mainboard section and link width, the bus speed will never dop under normal circumstances, a number of times in XP my bus would be stuck at x1, had to re-seat the card or clear the CMOS to correct it.

 

[drunkenmaster]

I was going to say that aswell, for no reason I could fathom when I went from a perfectly fine 4980 in a 16x slot, to a 5850, I was getting no where near proper performance, it turned out to be running at 8x speed which still shouldn't of given me the problems I had, eventually put the card into the second slot, started running at 16x speed fine(its a 790fx, both 16x slots) and performance was spot on.

The mobo as it turns out, died a few months later, something had gone wrong and that slot hadn't ever worked as it should till it finally went kaputski.

So try to force your current slot at 16x speed(power saving should not make a difference, only for devices that can turn off, maybe, your primary gpu it shouldn't afaik), I had several options in the bios and not one worked. try another slot, even if its only an 8x slot, it shouldn't limit you much. While mine ran at 8x speed, it wasn't performing as such at all. A proper working 8x speed slot should be more than enough for pretty much anything, a 5870 loses less then 2-3% in a slower slot.

I'd also consider putting the pci-e speed back at 100mhz, or auto, just to get into windows and see if the slot runs at 16x at stock settings.

 

[manveruppd]

**** it, CPU-z says it IS stuck at x1!!! Unfortunately the P35-DS3 only has the one x16 slot so I'll have to try your suggestions (re-seating the card and then clearing the CMOS jumper if that doesn't work) and hope that it gets unstuck. If it doesn't I'll have to put back the old card and see if that nudges it out of hibernation.

I find it REALLY unlikely that that's the cause of the performance though - GPU-Z's monitoring thing was reporting AT LEAST 80% GPU utilization under load! There's no way the bus could've achieved enough throughput to feed the GPU enough data to get it up there if it was working at x1 all this while - it MUST be sojme kind of power saving feature!

On the other hand, I turned all those power saving features off in the BIOS so I have no idea why it would be reporting x1 either....

 

[RavenXXX2]

x1 will obliterate your performance, it did mine, you can still have 100% GPU usage but the low bandwidth is killing the cards performance big time.

 

[drunkenmaster]

Its possible, theres plenty of boards with slightly odd implemenation of monitoring things and CPU-Z can misread certain boards, however the 1x thing is a very likely culprit, if at least because its telling you something is going on with the slot.

did you check with gpu-z (techreport) and hit the continue to monitor box to check temps and actual speed while in game. It might be running at idle speeds.

If you have any other pci-e slot I'd try it though, if its an 8x it really won't matter(as long as its actually working right), a 4x will limit performance, but if performance increases noticeably over what you have now, you know what the issue is. 4x is slower than 8x, but not as much as you think, 1x is the real killer.

But again, its more an indication something is wrong with the slot than the actual speed being the problem, bit of both probably.

 

[manveruppd]

Right, well...
Reseated the card, still x1. Cleared CMOS and (re-)re-seated the card, still x1. Uninstalled the nvidia drivers, cleared CMOS and (re-(re-))re-seated the card, still x1.
Removed the card and put my old 3850 back in, x16, without even installing the ati drivers.
Do I assume the only possible explanation is a faulty card?

PS. @drunkenmaster, all the other PCI-E ports on my mobo are x1 I'm afraid, it's only a DS3 And yeah, I did monitor while gaming (and while benching with 3dmark too). Used both GPU-z and Rivatuner, both reported the core/shader/vram clocks to jump up from their idle settings all the way up to full pelt. And even after a good few runs core temp never went above 42.

 

[manveruppd]

PPS. Still keen on getting the new card working, so if anybody can think of any other outlandish idea to try and kick-start it I'm willing to entertain it! I'll talk my friend innto letting me try it on his computer tomorrow, though with his PSU I doubt it'll work properly.

 

[Alex74]

My suggestion at this point would be try another card (other than the 3850), sounds to me at this point like a problem with the 260 itself..

 

[drunkenmaster]

I wouldn't jump to faulty card, as I said, I went from a 4890 to a 5850, could not get it to work at 16x no matter what I did. worked at 16x in the other slot without issue, the 4890 would continue to work at 16x in either slot when I tested.

I'd say its a motherboard issue, but its also very possible its a gpu issue, aren't computers a pain in the behind

Its worth having a look, checking if any of the pci-e connectors on the graphics card are missing as the gold bits can come off, or if one is corroded or dirty. If dirty/corroded, try to clean it up with isopropyl alcohol. Have a look for dust/debrie in the slot on the mobo aswell.

AFter a quick google I've seen other people have the same issue, its generally random taking card out and putting it back in that worked finally. Its probably dirt in the pci-e slot/dirt on the pcie connectors on the card as I said, clean them up and see if theres a difference at all.

Also someone said they couldn't get their 4xxx working at 16x they bumped pci-e voltage up by 0.2, and cpu voltage a touch, might just be mobo dropping a little juice under a higher load.

The other option is of course, beg someone to let you whack the gpu in their computer, if it works at 16x, its your mobo, if it still fails to work, its most likely the card.

 

[manveruppd]

I just came from my friend's house, the card only worked at x1 at his as well. The gold connectors are all there, flawless and shiny, I gave them a quick wipe with isopropyl and some lint-free paper just in case, but it made no difference. I literally tried every combination of BIOS settings imaginable, tried booting with only the bare essential components in case it was a power issue, plugged the card in and out so many times my motherboard's had multiple orgasms, and reset the CMOS settings more times in the past 3 days than I needed to in the past 3 years. I'm out of ideas!

 

[shahpor]

Apologies for the question, but is it the Bus Interface you are referring to on GPU-Z?

I also have a GTX 260 and mine reads PCI-E 2.0 x16 @x16 2.0

Is this correct?

Thanks.

 

[Zenduri]

yes shahpor