A tale of two 7900s ... laugh, then cry

[vois2]

Thanks in advance for reading. If you have the patience to read this, you'll probably laugh at how thorough and plodding my planning/testing has been, only to see it all go down the tubes. In the end, I now seek advice: Is it my power supply?

I recently bought two (2) XFX 7900 GTs, one of which I purchased as a gift for a friend. While I had both cards in my possession for a while, I benchmarked them relentlessly, each card getting several benchmarking days in the same box. The two cards performed exactly the same in every manner, including temperature ranges in each card. My conclusion is that these two cards are thus identical in performance.

These XFX 7900s at stock are 470/1370 . Before any of the below, I extensively tested my 350W PSU for constant and consistent voltage on the +12V and +5 rails. Also, the machine has never had any power-related problems before. My PSU gives 18 amps (rated) on both rails, max combined is 28 amps (rated).

So I shipped off one card to my friend as a gift. He immediately used Coolbits on his card to o/c to 562/1612 . He didn't crawl up the clocks, either. He just went straight to 562/1612 using Coolbits recommendation. Zoom. All of his benchmarks increased 15-20% thanks to this sweet overclock.

I am decidely more cautious, so I inched mine up only a tiny bit to 474/688 ... an almost microscopic overclock. All was well after some days, so based on his experience I decided I would FIRST install a Zalman VF 900 Cu VGA cooler, and THEN follow in his footsteps on a path to a high overclock.

Well, the Zalman certainly does a great job of cooling. All my temp ranges slid colder by 16-20 degrees Celcius. Sweet! So now I was going to have a much cooler card while I started to overclock. This joy was my last moment of success -- it's all been a wash since then.

I used Coolbits to take it to 559/1600 (almost identical to my friend's o/c). Ran 3dMark05 and enjoyed about a 5% increase in performance. Hmm, well, not nearly the gain of my friend, but better performance nonetheless. Then the fun began...

I wound up 3dMark05 for another go ... result: Crash, BSOD, nv4_disp infinite looping. I recovered and reapplied drivers (I use DriverCleaner properly always) ... now for another 3dMark05 test. Result: Fizzle pop random pixelation all over the place, freezing. I cleaned with DC again, now using RivaTuner, and lowered the clock substantially to 500/730 ... result: black screen, monitor gets no signal. Keep in mind that I continue to use DC Pro so that I can start fresh as I switch between Coolbits and Rivatuner ... and by now I have worked my way downward on the clocks ... at this point today it won't even overclock to the microscopic 474/688 I had with the stock cooler a few days ago. "Test settings" for anything above 470/685 and sometimes, pixelation, sometimes black screen, but always Windows crashes.

So basically I am back to a cooler running card, but it is at stock timings, so that the Zalman cooler ends up being a useless expenditure.

Here are my questions:

(1) I did get a full run of 3dMark05 on my FIRST go at 559/1600. Why would it run once and give me a 3dMark score, and then refuse to run 3dMark05 again at any overclock?

(2) The Zalman cooler fan requires power directly from the PSU, and this has to be pulling more power than the stock fan did because the stock fan needed power only from the VGA board itself. Do you think this higher power draw with the new Zalman fan could point the way to a need for a new PSU based on the errors described above?

(3) I have the Zalman fan running directly from a 4-pin molex to 3-pin converter, so she's getting juice straight from the PSU. I have several other molex connector available in there, and I have read about load balancing and what not, but I don't understand it. Is there any point to my trying a different molex connector on a different line bundle?

Thanks again. All thoughts much appreciated.

 

[4Ryan6]

Its almost impossible to advise you when you've only listed the 7900card and P/S w as 350, the rest of your system specs that are drawing power would be most helpful.

The brand name of the P/S is also helpful, is it a quality brand or one that came with the case.

If the Zalman had to be hooked directly into the P/S you still powered up the graphics card also right?

Heres a shot in the dark, the manual that came with the graphics card usually states what they reccommend as a minimum P/S rating, what was it?

And if your P/S is an older model and you had to use a splitter to power the graphics card in the first place its probably not up to the task at hand, it depends on just how much you have on each rail as to whats solidly being delivered to work with, and if the graphics card isn't receiving enough power to do its job, it won't do it, especially OC'd.

 

[3lfk1ng]

Vois2,

List your system plz.

With all processors and graphics cards and even memory modules there are a few runts in every batch. They work just the same as any others but when you try to overclock them, they may be very stuborn. For instance, the new Core 2 Duo e6600 can be overclocked to 3.2ghz with stock voltage, but some people got a runt and can only overclock to that speed if they step up the voltage. Thats just the luck of the draw, you never know what your going to get so always expect stock, and pray for overclock.

Example: The processor that I use is capable of being overclocked from 1.83ghz to 2.4 with ease and up to 3.2ghz when combined on the NF-7s mobo and some coolance system. Certain combinations of mobo/proc, gfx/mobo, or gfx/memory work better than others, thats the way it is. It's all about the runts.

Anyways most people have my processor overclocked to ~2.4 or ~2.6 with minor voltage adjustments. I have the worst runt ever made and I can only overclock my processor to 2.0ghz. If I go any further with any adjustments it fails, and the worst part is that I had to raise the voltage to get that small 204mhz increase which most people don't do till ~2.2- 2.3ghz.

You might have a runt?

So basically what I'm saying is that with all products on the market don't expect yours to be able to do that same as someone elses', athough they are made to be the same, they all have different results when you try to overclock them. Your issue sounds like it's just not capable of overclocking because you may have a runt. You just got lucky on your first 05' run, so I suggest you try the following:

1)I don't think it is your psu because if it was, It would shut down the computer or even worse, shoot blue and orange sparks. The artifacts that you experienced are from reaching a speed that your graphics card cannot maintain. As for it only being able to operate at stock speeds only...thats an entirely different issue that you'll need to troubleshoot.

2) If you have a spare graphics card, that is unmodded (still stock cooler) try it. Although it will use less voltage, you can check to see if It is merely a gfx card issue or something else.

3) I would suggest trying a friends 450+ watt psu, If you have no friends, find some otherwise your going to have to fork out another $80 for a decent psu. When you use their psu use ur gfx card w/ zalman and see if it's just the psu and if #1 failed...try using that card with your friends psu.

4) If all else fails, you may just have to put back on the stock cooler...and see how well it runs then see how well it runs with a slight overclock ...and then if all goes smoothly perhaps attempt to put back on the zalman and try again.

5) If your within 30days...return the card for a new one and pray for the best.

Good luck,
~3lfk1ng

 

[vois2]

The rest of the system specs are not abnormal, especially for non-enthusiast parts. P4 630 3.0 GHz, 925x chipset Dell mobo, 2x hard drives 7200rpm, 1x optical, 3.0 GB RAM, PCI soundcard.

The PSU is Dell. Supposedly Dell underrates their PSUs by a bit. I listed the rated specs of the PSU in the OP.

The card is powered by the direct-from-PSU 6-pin connector.

The XFX manual states in print 350W PSU required. The PSU previously supported a 6800GS 256MB without having any problems. The 7900 draws a little less power compared to the 6800GS thanks to nVidia's die shrink.

The PSU is not older -- it has the direct PCI-E P6 connector made for PCI-E cards.

 

[vois2]

3lfk1ng wrote: "luck of the draw, you never know what your going to get so always expect stock, and pray for overclock."

That's a good point. If I try a newer 450W PSU with at least 22a on the 12v rails, and get the same result, I'll just chalk it up to your point made above.

In reply to the numbered items from 3lfk1ng...
(2) A good testing idea. I have a PCI-E vanilla 6600 that I could do this with.

(1) I am getting the black screen where the computer seems to shut down; it's not a complete down and "off" but it just sticks on the black screen until the monitor reports either out of range, or, no signal.

 

[dberthiaume555]

Call your friend and tell him you sent him the wrong card by mistake and you need to swap...

 

[vois2]

To dberth:

LOL ... it's funny you say that. Actually, when I had both of the cards in my possession, and tested them extensively, his card always performed between 1.0% and 1.5% better than the card I kept here. At the time I did not figure this to be statistically significant. I chose to send him the better-performing card because he was preparing to move overseas, and didn't want him to have to hassle with an RMA (just in case).

Well, hmmm ... darn, too late to be an Indian gift there.

 

[4Ryan6]

So basically I am back to a cooler running card, but it is at stock timings, so that the Zalman cooler ends up being a useless expenditure.

NiBiTor

 

[vois2]

This is the original poster just following up on the resolution of the original problem. That problem was the discovery that I could not o/c my 7900GT by even one megahertz. I suspected the PSU was the problem. Thanks especially to 4ryanc and 3lfk1ng for helping me. I replaced the Dell 350W PSU with a 520W OCZ Powerstream power supply, and the problem went away. I now have the card o/c'd to 562/810 (from 450/685) and this provides about a 5% improvement in benchmarks and game fps averages. Ironically, even at this overclocked state, temps went down by 3 degrees in all ranges of idle and stress. I guess this has to do with the card fan receiving more stable power? The only disadvantage to the 520W OCZ is that it makes my CPU fan spin at astronomically faster rates when the CPU is very stressed. This isn't too often though -- the worst is during the CPU tests of 3dMark.

 

[4Ryan6]

You're Quite Welcome!