8800 GTX on LN2

I had my first run with my Tek-9 tonight as well as my first GPU on LN2. Well....first successful/semi-successful run hehehe.

Here is the system.

P5K Vanilla voltmodded (Vcore/Vmch/Vpll, and all 3 measure points)
Q6600 G0 cooled on LN2 by a Dragon Evo 2.4 pot
8800 GTX (stock volts) cooled on LN2 by a Tek-9
Corsair HX620W
WD Caviar IDE 160gb
2 X 1GB Mushkin Extreme Performance 1066mhz

The max core clock I could achieve stable enough for 3D06 was 729 (real).
The max shader clock I could achieve stable enough for 3D06 was 1750 (1728 real I believe?)
The max memory clocks I could achieve stable 3D06 was...stock. 950

I ran the GPU pot anywhere from -30C to -50C. Should I be running colder?

A few others questions.

1. Regarding the memory clocks. Is this because there is a decreased heat dissapation from no timm/heatsinks/fans on them anymore? As far as I have been told its not a lack of voltage. My 8800 GTX has Hynix chips as opposed to the Samsungs, and from what I have been told (personal communication - ViperJohn from OCF GPU forums) the Hynix memory chips do not need any extra voltage to overclock and in fact, more voltage to these chips actually hinders overclocking potential. Agree or disagree with that statement? How can I get to increase my memory clock settings?

2. I was hoping for more on the core clocks than 729 even stock. Is that decent for stock volts LN2? I was hoping for closer to 800ish.

3. I was unable to run 3D05 at the above mentioned settings on 2 attempts (CPU 490X9, GPU 729/1750/950) but was able to run 3D06 repeatedly at those settings. on 3D05 1x it restarted and 1x it hung. I dont recall which test it hung on 3D05 though. More likely CPU or GPU related for 3D05?

4. Did not get to run 3D01 as I ran out of LN2 3D03 I got 1 attempt at, and it hung near the end. Cant recall the test either and settings were same as above.

5. Canyon Test 3D06 I was getting red and green "sheet-like" artifacts every single time I ran that test. Everything else looked 100% artifact free and crisp/clear. Is there something particular about that test that causes this? I have not seen that before. At first I thought it was because the GPU pot was not cold enough, but that was not it.

Even considering that I was not able to OC the memory, Im disapointed in the scores. With the same CPU at 3.9Ghz and 8800 GTX on air at 660/1620/1050 I was able to get 15,800, and only 16,800ish on LN2 for both GPU/CPU. I definitely need to look into some additional tweaks and suggestions are always appreciated

Thanks for looking!

3D06


Aquamark

yikes, 1.896 vcore on that Q6600...that frightens me!

jivetrky said:

yikes, 1.896 vcore on that Q6600...that frightens me!

Click to expand...

When im pushing the CPU to the max I run it at over 2.0 Vcore

So is that how you blow caps on your motherboard? Pushing 2 volts on a cpu?

EmAn said:

So is that how you blow caps on your motherboard? Pushing 2 volts on a cpu?

Click to expand...

Nope. I believe I was at all stock when the caps blew.

Im 99,9% certain the caps in that other board you are referring to blew from a faulty VMod.

you? all stock? that is hard to believe as well.... Alright... Sorry to hear your getting disapointing results on your 8800gtx but... i wouldn't mind having a bum overclocking 8800gtx

just as long as i get it free lol

how high did it get on air? that seems very low for sub ambient

Not bad doc, good experience. The thing to remember is most of the score improvement is going to come from the card clocks on '06 (it's more balanced on AQ3)...basically you need to try this again with a working core mod - 1.6V should do nicely. I'd also like to re-iterate that you should try less vcore on the 6600 next time, if you havn't already. Too high a vcore can make for less clocks, thing is you might have broken the chip in to those voltages already

I have tried (several times allready Mark) at your advice and others advice to go with less Vcore.

It dramatically limits the clocks I can achieve when I give it less Vcore on any of the boards Ive tried it on (Mars and P5K vanilla)

I have my notes at home, but at 1.8ish Vcore on the vanilla last night most I could get with the Q6600 was about 480 FSB. With 1.9Vcore I was able to get 505-510ish.

At least thats the way It seems to me but I still need additional testing.

Edit: Most I was able to get on air was 650ish. Would hang anywhere near there until I dropped down to the next 27mhz increment.

But I was also staying at stock 1350 shader with it unlinked, which if you are aware of my other thread (no need to cross reference) was limiting my max clocks available.

So,.....

Just to reintroduce 1 issue that I still need clarification on....

What are the avg temp ranges of the pot I should shoot for when cooling a GPU on LN2?

Is there such a think as a GPU coldbug?

Should I go lower than -50c?

Someone on XS (TM) told me I should shoot for -80c to -100c.

Is that accurate?

Also, any suggestions on why I cant OC the memory at all would be appreciated if anyone has any advice

Thx!

I'd see if it acts up, some cards may have a cold bug, just have to test it and see really, I'd start out slow and go to dry ice temps, havn't heard of any bugging at ~-75 and go from there

So...-30c to -50 is NOT cold enough to achieve max clocks?

No, as cold as you can run it without hitting the bug is where you will reach max clocks. A GTX should do 900MHz with 1.6V between -60 to -80. When you get cards too cold the screen will garble and you have to warm them up to run again.

Q66 is looking nice, but yeah... the GTX is not doing it for me...

PhysX said:

Q66 is looking nice, but yeah... the GTX is not doing it for me...

Click to expand...

Phys, could you elaborate?