PDA

View Full Version : VERY interesting discovery wrt 6800gt voltage, stability


Pages : [1] 2

myshkinbob
09-10-04, 12:02 AM
I should probably start this post by pointing out I DO NOT smoke crack.

Skip this paragraph if you don't want the background info. I've got the 6800GT (leadtek) and running it in my old pc i could hit 425/1100 easily, but after upgrading the cpu and motherboard, i found i couldn't top 380 core with fast writes enabled, or 405 core without fast writes. I also got little stutters when i first loaded games, like a 5 second freeze. I kept the same PSU as my old rig, an enermax 430w with 20A (22A peak) on the 12v rail, and my 12v voltage was a lot lower than with the old rig, from 12.3v to 11.7v. So although i should be fine for amps, i put the stutters and reduced overclocking ability down to the droopy 12v rail. A little test with rivatuner indeed showed my card droped to Low Power 3D mode occasionally, and instantly if i pushed the core speed higher than 380 with fast writes enabled. So i took a look in the stutter/freeze thread an hour ago, and noticed some people with the same issues had bought bigger 500+W PSUs and fixed their problems. I can't afford one right now, but it got me thinking. I'm a bit of a quiet pc freak, and if you ever read http://forums.silentpcreview.com/ you'll know undervolting is the big thing for quietness. I undervolt my A64 3000+ to 1.4v (0.1v below default) and still overclock it by 250mhz (10x225), all perfectly stable. You can see where this is going. Default processor voltages are a most-likely-to-work value, to allow for minor defects and poor voltage regulation, you can usually run at stock or close to stock on less voltage. So, i've not got enough 12v current to overclock my 6800GT like i used to, maybe i can undervolt it, reducing the resistance, and impedence to current flow, therefore reduce the current needed.

So basically i ripped the bios off my 6800GT to disk, loaded Omni Extreme, let's try 1.2v for 3D. No 1.2v option there (i'm going to email the dev about that tomorrow, hopefully he'll add it to the next release if i'm polite :) ). Realising a 0.2V undervolt even on a cpu means lower clock speeds, i thought i'd set 3D to 1.1v just to see how low i'd have to go for stability.

I was pretty amazed to see the old startup overclock of 380/1100 held up after leaving rthdribl on for 20 minutes. But it get's better, i tried 400 core, no drop down to Low Power 350! And this is with fast writes on, on an nforce3 250 motherboard! Pushing it to 425 did cause the low power 3D to kick in though, but at 1.1v for 3D that's not suprising in the least. Back to 400, i've just run 3dm03 looping tests 1 to 4 for half hour, the gpu hit 85c and 63c ambient, but no Low Power slowdowns anymore, and no stutters.

so, what do you make of that?! the 6800GT runs at 1.1v core with a +50/+100 overclock as well, with fast writes enabled, on an nforce3. My advice to people with sub-500W PSU's and brief lockups or Low Power mode issues, is have a go at 1.1v, though don't expect to go much beyond 400mhz core with such a large undervoltage. I'm really interested to know if i'm just lucky or what, i really didn't expect the results i'm getting here. I'm also keen to see what a 1.2v option would reap in terms of overclock. Just goes to show, you don't need to chuck voltage at chips to get them to play nicer.

oh some numbers, 3dm03 went from 11632 to 12221, doom3 timedemo demo1 from 66.6 to 73.7 (16x12HQ). The old numbers are from driver 66.00, new numbers 65.76, so that does play a factor in the difference i'm sure.

cambrian
09-10-04, 02:14 AM
I don't know why people keep insisting on the importance of wattage when it doesn't mean jack. A good 480W pc power & cooling psu can kick a standard 550W's butt any day. What matters is the rails and the other specs of the psu. This is kind of off topic from what you are saying, though. I have a 480W Neopower and it runs perfectly fine with my system. By the way, probably something to prove your point: I actually flashed my 6800gt to a gainward ultra bios and I changed the voltage from 1.4 to 1.3. The 6800gt runs perfect fine with 1.3V at 420/1200. It'll probably work fine with 1.2V too.

mustrum
09-10-04, 05:41 AM
So 11.7 on the 12 volt rail kills OC and leads to stuttering?
I was about to buy a Gainward GT Golden Sample but my Enermax puts out 11.7v on the 12 volt rail (Even when i undervolt/clock my CPU).
I'll better stick to my ATI card then i guess. :(

Templar
09-10-04, 07:57 AM
11,7 is within ATX specs, if that rail is steady at that voltage the PC should be stable.

Simon

mustrum
09-10-04, 08:10 AM
11,7 is within ATX specs, if that rail is steady at that voltage the PC should be stable.

Simon
Yeah SHOULD. If you read above his 6800 stuttered and had trouble at 11.7volts. The XT PE works with my PSU that i know allready.
I wanted a 6800 ULTRA but i am not willing to buy a new PSU since my enermax was not cheap either. It's friggin impossible to buy an ULTRA in europe as well.

Knot3D
09-10-04, 08:24 AM
^Then your expensive Enermax isn't doing so well...

I got a Tagan 480 on a OC'd P4 3,2 @ 3,4 with a Gainward golden sample GT @ 400/1100 ; works great. 12V doesn't dip below 11.87

mustrum
09-10-04, 08:45 AM
^Then your expensive Enermax isn't doing so well...

I got a Tagan 480 on a OC'd P4 3,2 @ 3,4 with a Gainward golden sample GT @ 400/1100 ; works great. 12V doesn't dip below 11.87
hat may be the case but the x800 XT PE works flawless and rochstable with it so i'll just keep it. That's all i said. (Was about to buy a 6800 ULTRA but this was the 1st hit into my face. Second was that i couldn't buy one even i i wanted to.)

Pandora's Box
09-10-04, 09:48 AM
i just tried this. seems to be working great.

at 1.3volts i idle at 48C
at 1.1volts i idle at 48C

detect optimal frequencies with 1.3: 396 for the core, 1.09 for mem
detect optimal frequencies with 1.1: 406 for the core, 1.11 for the mem.

manual overclocking with 1.3: 415 for the core, 1.10 for the mem.

havent tried manual with 1.1 yet

edit: results with manual at 1.1volts: 425 for the core, 1.15 for the mem. impressive! but weird lol.


nice to know i can overclock more, but i run at stock anyway. so even better to be able to run at 1.1volts instead of 1.3volts.

Templar
09-10-04, 10:05 AM
The PSU might be delivering 12v+ but the mobo might not be doing it's job, having low end capacitors and cheap 2 phase power design.

Simon

mustrum
09-10-04, 10:31 AM
The PSU might be delivering 12v+ but the mobo might not be doing it's job, having low end capacitors and cheap 2 phase power design.

Simon
It's an asus a7n8x Deluxe 2.0. One of the best nforce2 motherboards out there. I doubt it uses low end cpacitors.
It could be motherboard monitor 5 though. Everest gives me readings of 12.4volts!!! MBM gives 11.7. Strange isn't it.
@Pandora: More OC with less voltage. Strange for sure. If i were you i'd leave it on stock. There must be something else affecting the GPU. You never get more mhz with les voltage. This is just wrong. Be careful there.

Pandora's Box
09-10-04, 10:56 AM
in the bios for the a7n8x deluxe it lists voltages for 3.3, 5 and 12 i believe. check there. i get around 12.04 or 12.4 cant remember.

mustrum
09-10-04, 11:08 AM
in the bios for the a7n8x deluxe it lists voltages for 3.3, 5 and 12 i believe. check there. i get around 12.04 or 12.4 cant remember.
Thanks for the reminder. I forgot about that one. 12.41v on 12 v rail. Sounds high but is perfectly fine. Read up on that on the net and it looks like 12.4 are way better than 11.7.
My 5v rail is struggling a bit though: 4.70v only and going below that even when running prime95. There i know why: An Athlon XP @2.3ghz sucks a lot on the 5v rail.
Anyways: The decicion is made: XT PE it is. I know it works so i will stick to it. It's not like i'd be not satisfied with the performance. Missing stere 3d drivers is the only downside so far.

Pandora's Box
09-10-04, 11:27 AM
yeah the x800 xt pe is one heck of a card. sure it doesnt support sm3.0 but we dont even know if that is going to be important or not.

myshkinbob
09-10-04, 11:37 AM
How'd my thread get so off topic? :)

11.7 is within ATX spec yes, so it doesn't worry me, but a droopy voltage is indicative that the rail is near it's amperage limit.

But back to the topic, do you people not find it odd that not only does the GT run at stock speeds at only 1.1V, but it also runs at ultra speeds at 1.1V, and ulras are defaulted at 1.4V. That's 0.3V lower, over 20% less than nvidia sets these cards to.

And in terms of overclocking it's very backwards, the common rule is to get a better OC, up the voltage, but i (and Pandora's Box, thanks for testing too :) ) get a better OC from seriously lowering the voltage.

A little math to back up what's going on with reducing the current needed, everyone knows from school that V=IR, so I=V/R, and 1.1/R < 1.3/R.

Knot3D
09-10-04, 11:38 AM
Pandora's Box,
^ what if i say ; Splinter Cell 3, Stalker and Vampire Bloodlines are on sm3.0 list.....

mustrum
09-10-04, 11:41 AM
yeah the x800 xt pe is one heck of a card. sure it doesnt support sm3.0 but we dont even know if that is going to be important or not.
Yeah agree. About SM3.0. Maybe it's no advantage at all (SM2.0b seems to get supported by many programmers wich is great) and if 3.0 will be great i'll have some performance loss over the 6800s someday in the future. I guess i can live with that.
Th 6800's are todays high tech cards while ATI countered with brute force.
Funny thing is how close the match is with those totally different attempts.
I don't think that the XT PE is the better card or other way round. I just bought what is available. There were no GTs and no ULTRAS around but there were x800 PRO VIVOs. I know it's the other way round in the US though.

zbose
09-10-04, 12:35 PM
Sorry your thread go sot hi-jacked. Personally I am tired of watching people publicly justify their reason to buy one card or another every chance they get.

Anyway, it seems to me that there is a logical explanation for the results you see. You said before that your card was kicking into lower power 3d mode and that was causing the stuttering (not artifacts). So it sounds like the issue you ran into was purely related to heat on the die. Your GPU was getting hot and clocking itself down, but it was never having stability issues.... could be that your thermometer is poorly calibrated and the GPU thinks its hotter than it really is.

My point is, if you assume that the GPU was still running with in its spec, but clocking itself down due to heat, then undervolting effectively reduces the heat... now that your GPU is cooler, it can clock higher w/o slowing itself down when it "thinks" its too hot.

So what changed between systems? Did you upgrade to a hotter processor that puts out more heat inside your case?

Just my 2 cents anyway.

How'd my thread get so off topic? :)

11.7 is within ATX spec yes, so it doesn't worry me, but a droopy voltage is indicative that the rail is near it's amperage limit.

But back to the topic, do you people not find it odd that not only does the GT run at stock speeds at only 1.1V, but it also runs at ultra speeds at 1.1V, and ulras are defaulted at 1.4V. That's 0.3V lower, over 20% less than nvidia sets these cards to.

And in terms of overclocking it's very backwards, the common rule is to get a better OC, up the voltage, but i (and Pandora's Box, thanks for testing too :) ) get a better OC from seriously lowering the voltage.

A little math to back up what's going on with reducing the current needed, everyone knows from school that V=IR, so I=V/R, and 1.1/R < 1.3/R.

zbose
09-10-04, 12:39 PM
Oh, and typically, putting more voltage into the processor only helps you gain an OC when your chip is still properly cooled. IE, if you are unstable because you OC'd so high your chip is hot, then adding more voltage won't help a bit... but if you are like me, and watercooled and put out 40C max under load then you have to start upping the voltage to go any farther.

I think in the case of the GPU you are maxing out due to heat, so its not suprising that reducing the heat (even at the cost of voltage) is gaining you more clockspeed.

mustrum
09-10-04, 12:42 PM
Sorry your thread go sot hi-jacked. Personally I am tired of watching people publicly justify their reason to buy one card or another every chance they get.

Anyway, it seems to me that there is a logical explanation for the results you see. You said before that your card was kicking into lower power 3d mode and that was causing the stuttering (not artifacts). So it sounds like the issue you ran into was purely related to heat on the die. Your GPU was getting hot and clocking itself down, but it was never having stability issues.... could be that your thermometer is poorly calibrated and the GPU thinks its hotter than it really is.

My point is, if you assume that the GPU was still running with in its spec, but clocking itself down due to heat, then undervolting effectively reduces the heat... now that your GPU is cooler, it can clock higher w/o slowing itself down when it "thinks" its too hot.

So what changed between systems? Did you upgrade to a hotter processor that puts out more heat inside your case?

Just my 2 cents anyway.

Sorry. I didn't wantz to hijack this thread but itS' true it went off topic.
I'm not justifying my bought card. I use a GF3 TI200 right now. I sold this modded XT and just bought another one. (profit)
I was considering a 6800 ULTRA instead to see the competition.
Pandoras Box is one of very few people that seem unbiased. I can discuss with him without dumb comments for any side.
I came into this thread because of this power issue but slided into the off.
:lol2:
Sorry to the thread starter.

zbose
09-10-04, 12:54 PM
:) No hard feelings. Video cards are neat. Now back to why this man saw a performance gain from undervolting.

:clap:

-zbose

Sorry. I didn't wantz to hijack this thread but itS' true it went off topic.
.....

{Sniping}Waste
09-10-04, 01:16 PM
:) No hard feelings. Video cards are neat. Now back to why this man saw a performance gain from undervolting.

:clap:

-zbose


Looks like heat is the thing when under load. The HSF on the GT is poor at best for the amount of heat the core kicks out. The lower voltage will lower the heat at full load.

myshkinbob
09-10-04, 01:24 PM
Sorry your thread go sot hi-jacked. Personally I am tired of watching people publicly justify their reason to buy one card or another every chance they get.

Anyway, it seems to me that there is a logical explanation for the results you see. You said before that your card was kicking into lower power 3d mode and that was causing the stuttering (not artifacts). So it sounds like the issue you ran into was purely related to heat on the die. Your GPU was getting hot and clocking itself down, but it was never having stability issues.... could be that your thermometer is poorly calibrated and the GPU thinks its hotter than it really is.

My point is, if you assume that the GPU was still running with in its spec, but clocking itself down due to heat, then undervolting effectively reduces the heat... now that your GPU is cooler, it can clock higher w/o slowing itself down when it "thinks" its too hot.

So what changed between systems? Did you upgrade to a hotter processor that puts out more heat inside your case?

Just my 2 cents anyway.

I really do not think it is a heat issue, the evidence being the card ran hotter in my old system but could reach 425 core with perfect stability. The reason it won't overclock that far now is down to my cpu upgrade, from an AXP 2400+ to an A64 3000+. The main difference is my old 2400+ mobo didn't even have an ATX_12V connector.

There seems to be a belief that a 6800 core running over 80c is too hot, but these cards are built to run at such a temperature. It may seem un-natural for anything to run well at such a temperature, but A LOT of peoples 6800s get into the 90s without issues. Even nvidia don't throttle it down until you're into 120c territory.

I can see why you'd think that though, if i saw a cpu running over 65C i'd think there was a problem.

zbose
09-10-04, 01:38 PM
I don't think 80C is too hot, my card is always up there when I stress test it...

But I just re-read your post and realized your were talking about low-power mode on the card... not that it was clocking down for heat... which throws my theory out the window :)

So yeah, the stuttering probably comes from transient loads sucking up whats left of your already overstretched 12V rail.... but as you yourself acknowledge, your OC is only limited by PSU constraints... IE, if you downclock/undervolt your A64 even more, I would expect that you could get back up to 425 at regular voltage on your 6800GT w/o having it drop into lowpower mode...

You are right, it is very interesting :) I think you should just go ahead and buy the NEOpower (dual isolated 12V rails). I love mine. Then you can run it all like it wants to be run, fast and hot!

I really do not think it is a heat issue, the evidence being the card ran hotter in my old system but could reach 425 core with perfect stability. The reason it won't overclock that far now is down to my cpu upgrade, from an AXP 2400+ to an A64 3000+. The main difference is my old 2400+ mobo didn't even have an ATX_12V connector.

There seems to be a belief that a 6800 core running over 80c is too hot, but these cards are built to run at such a temperature. It may seem un-natural for anything to run well at such a temperature, but A LOT of peoples 6800s get into the 90s without issues. Even nvidia don't throttle it down until you're into 120c territory.

I can see why you'd think that though, if i saw a cpu running over 65C i'd think there was a problem.

myshkinbob
09-10-04, 01:41 PM
I'm not going to downclock my A64, never :) At the moment i overclock/undervolt, to 2.25ghz@1.4v (stock is 2ghz@1.5v), another un-natural overclocking phenomenon eh.

zbose
09-10-04, 01:48 PM
What kind of cooling do you have on the A64? I am not suggesting that you actually downclock it, but rather that you buy more juice so you can overvolt/overclock them further :)


I am so close to just buying a damn a64.... but I want to wait for PCI-e
:drooling: