"new" 9800GTX+ and/or SLI issues

[laywill]

Hi everyone!

I have been running my inno3D 9800GTX+ overclocked for about 2 or 3 years now (Phenom 940 and Crucial DDR2 RAM also overclocked). I recently bought for £30 off eBay a second 9800GTX+ as a cheap way to leap into the world of SLI

...BUT...

I have set up SLI in the nVidia control panel and removed the overclock on the graphics cards, however whenever I load up the GPUs (e.g. 3DMark, Skyrim) the system will manage about 15 seconds then crash. I get black screens with the audio continuing to work for about 10 seconds then everything is black and silent untill I do a hard reset.

My inclination is to say its a memory issue on the "new" GPU, and only when I load it up is the dodgy memory being accessed / the rate of read/write errors becomes too high to handle or something...


IN A NUTSHELL: What do people think? Is there anything I can do to test GPU memory? Have I bought a damaged card?



Thanks

Laywill

[shaithis]

Put the new card in on its own and stress test it.

You will then see if it's the new card or an SLI/driver issue.

[laywill]

Put the new card in on its own and stress test it.

You will then see if it's the new card or an SLI/driver issue.

Thanks, will give it a go and report back on Monday (probably) when I get a chance to :-)

[007]

Can you provide some more info about your psu? Especially the 12v rail rating, model..I have never come across arctic cooling psus.

[laywill]

Hello again,

I've had a bit of a play and racked up the energy bill a bit courtesy of Furmark :-)

Both cards when installed as individual cards run absolutely fine on stock with a half hour stink of Furmark or OCCT (woo!)

In SLI (tried the 3 way bridge over the 2 way just in case it helped) they are... variable...
I re-installed the latest drivers from nVidia and tried turning the fans onto 100% rather than auto, and can now just about make it through a 3Dmark06 run at 1680x1050 with 4xAA and all other settings at default (I decided worth testing at the settings I want to game at). It will then blackout after a little while

Interestingly - if i boot to windows and just leave it sat on the desktop (fans on auto) it will blackout after about 15 or 20 minutes

POSSIBLY IMPORTANT POINTS:
1) the second graphics card has about 3mm of clearance from the bottom of the case because of where the PCIe slots are on the mobo and my crap case... because of the design of the coolers, the heatsink isn't enclosed however, so airflow is better than it might be.

Worth the addition of a fan somewhere to blow more cold air in at the heatsink maybe?

2) The PSU was reasonably cheap from eBuyer a few years ago - had one blow up and they sent me a replacement "better" model...
According to the label on the side:
AC input: 230V 50Hz 10A
+3.3V - 24A
+5V - 17A
+12V1 - 24A
+12V2 - 24A
+12V3 - 24A
-12V - 0.5A
+5Vsb - 3A

It has a 20+4 pin motherboard connector which is being used, a 4 pin motherboard power connector (currently unused), 1 string of molex connectors (powering 8pin motherboard, 2x9800GTX+, DVD drive and a fan controller), and 1 string of SATA connectors (powering my 1 solitary HDD).

Is it worth removing some of the SATA connectors and replacing with Molex so I can draw off (what I assume will be) a different +12V line?

The PSU also has a tendancy to run quite warm... not worryingly hot, but definitely a LOT of hot air being blown out whatever the load is.


Thanks in advance

laywill

[Golden Dragoon]

W= VxA, this basic equation shows that your PSU seems to be rated at 864w on just the 12v or 1028w overall, now there will be a lower max possible draw when taking power from all 3 rails, but still 3 proper 24A rails is something only expensive psu's have, the fact that it has only a 4 pin cpu power connector, and a general lack of other connectors leads me to think that the label isn't worth the paper it's printed on, though I may be wrong on that, what is the model and it's rated wattage?

[laywill]

You're probably right! I freely admit it was a cheap PSU I bought several years ago when I was a lot more naive than I am now

It was sold as a 750W PSU and there is in tiiiiiiny print on the side some max draw figures that add up to 750W. Though I get the impression you're leaning towards telling me to junk/ebay this thing as fast as possible and get a proper PSU :/

[Agent]

SLI really stresses your system, you just can't get away with a cheap unit most of the time

[Pob255]

My bet is the psu, the arcticpower (please note nothing to do with arctic cooling) are not good psu's
and probably only around half the given wattage.

For your listed specs I'd look a good 600-750w psu, I ran an i5 750 OC'd and sli 9800gtx (original) on my enermax 625w
Although I had tearing issues due to card BIOS miss match but that's a separate issue, although you could run into it if you fix the power issues.

[laywill]

Confusing Arctic Power with Arctic Cooling was indeed the mistake I made when I bought it thinking it was a branch off into a new market (much to my loss as it seems)

So a 600 - 750W power supply from someone reputable e.g. Corsair, OCZ, Coolermaster should see me right we think?

If it is a power supply issue I don't understand why the system crashes if I just leave it on the desktop after 15 or 20 minutes though... its clearly nowhere near maximum load so surely that's not the cards trying to draw too much power...?

[shaithis]

Might also be worth upping the IOH voltage and PCI-Express clock drive strength values in the motherboard BIOS.

[laywill]

Might also be worth upping the IOH voltage and PCI-Express clock drive strength values in the motherboard BIOS.

IOH Voltage? And what do you mean by "drive strength values"?

[andy572]

I'd have been looking at the question of power - and it looks worse when you say that it's not a great unit, is old and gets toasty.

Cooling could be key too. I've got a 9800GT here and that thing's hot. Even at reduced clocks on the desktop, my card runs at around 60 degrees (high 50s to just over 60 at this time of year) and I think load throws it to at least the mid-80s.

I'd be changing the power supply anyway. You may have been and remain lucky, but it could take all of your system down with it too. Of course, there's always that chance with any failure - but it doesn't need to be invited.

Ah, I forgot the cooling again.

The air might be stalled there and the cooling less effective than you think. A few millimetres at least doesn't sound good.

[laywill]

A curious update:

Afternoon all, I've been playing around to try and eliminate any configuration issue and ensure it is actually the PSU that's at fault.

Many interesting things have happened in my testing of the cards:
I can now pass 3Dmark 06 on 1680x1050 4xAA in any configuration (CPU + RAM OCd but not the GPUs) in or not in SLI
I can run OCCT's GPU test in any setting OTHER THAN in SLI or with the second card dedicated to Physix seemingly...

But most intersting of all:
I had one monitor connected to each GPU and was playing Skyrim on one, with GPUz and HW-Monitor on the other to keep tabs on GPU load and temperatures. I got a good half hour of game time then all of a sudden when I did a fast travel - so without any particularly heavy GPU load, BUT a lot of memory work - the primary GPU with skyrim being outputted on it crashed, but left the other GPU with GPUz and HW-Monitor running still working fine (well... given that one GPU had gone fizzle ). Computer was however unresponsive and required a reboot...

This brings me back round to the start of this thread questioning whether there is some faulty memory on a GPU somewhere or whether what I'm still seeing is the power supply going fizzle

If I can have a dig around and find some money I think I'll order a new PSU anyway... thoughts on this 750W OCZ modularone?: http://www.scan.co.uk/products/750w-...-year-warranty

[Bambooz]

OCZ ZT Series are decent units. Built by an off-brand OEM (Great Wall), but decently. All japanese capacitors (Rubycon and Nippon Chemicon) on the secondary (output) side too, which is a bit suprising for a Chinese PSU OEM. Should last a looong time.

However, looking at your sys specs, 750W is total overkill.
A single 9800GTX uses 156W max., a 9800GTX+ 141W

Using 150W as a safe middle, you'd be at 300W for both GPUs together (if they're at full load that is).
The Phenom II X4 940 BE is rated at 125W TDP, with the OC, let's say 150W.
That's 450W on 12V. Add all the other tidbits of the system and you're maybe scraping 500W (if everything's fully loaded to the maximum, which pretty much never happens unless you provoke it by running tons of torture tools at the same time)

A 550W OCZ ZT Series has 45A on 12V, which translates into 540W. In _theory_ that's enough already.
The 650W version has 54A (648W) on 12V, which is already way more than you need, even if you somehow managed to load down both GPUs and the CPU and everything else to the max.
However, it would give you a quite sizeable headroom for upgrading your GPU(s) to something more recent later on.

The 750W version is just.. plain overkill.

http://www.scan.co.uk/products/550w-...-year-warranty
http://www.scan.co.uk/products/650w-...-year-warranty

I'd say go for the 650W and have quite a lot of headroom, rather than spending even more for the 750W which is so overkill you won't ever manage to load it up even to just 2/3rd of what it's capable of.

Going crazy high with the wattage is the usual mistake after someone fell for a crappy china firecracker PSU with "650+W" or something similar on the fake label and a real maximum of maybe 300W if you're lucky and it doesn't blow up at that point.

I've had people call me crazy for using a 430W PSU in a system with an overclocked quad and a HD6870 before (see the specs thingie). They suggested 650W minimum. I laughed
My PSU has just 360W (30A combined) on the 12V rail, and it powers all the crap in my specs thing just fine (and has done so for the past 4 or 5 years).

[laywill]

Hi Bambooz,

So 650W it is then!! I'm quite happy having extra money in my pocket

The other option open to me would be buying a better 650W PSU than the OCZ one, such as the 650W modular Corsair one:

http://www.scan.co.uk/products/650w-...-eps-12v-quiet

Thoughts on whether its worth the extra money? They seem to have excellent reputations!