Increase in VideoCard clocks by +5% adds +20% to 3DMark scor

[TabrisDarkPeace]

Increase in VideoCard clocks by +5% adds +20% to 3DMark scores

That is my 'great puzzle' of the day.

Last time I took 3DMark seriosuly the company was called Mad Onion.... around 3DMark 2003SE I stopped believing in it.... now they are called Futuremark. Someone else on these forums got me back into 3DMark so figured I'd try and get my score 'up there' for a Radeon X800XL.

I knew that linear progression in 3DMark led to exponential increases in results (check the ReadMe), but wasn't expecting it to be so drastic as this.

Taking a ATI Radeon X800XL from 380/465 (underclocked) to 400/490 (stock defaults), which is about a +5% increase in performance for both GPU and Video RAM leads to a +20% increase in my 3DMark05 score.

From: 380 GPU / 465 VRAM giving 4,201
To: 400 GPU / 490 VRAM giving 5,168

So would going from 400/490 to 420/515 increase the 3DMark05 score by another +20% or so ?, to say around 6,200 ?

Now I ain't changed anything else, Driver forced texture quality, FSAA, etc is not in use. Geometry Instancing is even off, and using Catalyst AI 'Standard' only. Not AI 'Advanced'. Just stock default Catalyst 6.1 driver paired with the 'nForce4_amd_6.70_winxp2k_english.exe' driver on a Tyan K8WE (S2895, BIOS 1.02).

None of my hardware is 'modded' either. Not my scene anymore. I've had some 'weirdness' with my video card in the past, (search for other posts under my name) but in the last day or so of testing it has appeared fine.

The only variables during the test where the GPU and Video RAM clock speeds. Nothing else has changed to isolate the score change to just 3DMark.

I find this incredibly suspicious to be honest, and it wouldn't be the first time 3DMark has been caught in the spotlight for doing 'interesting' things with scores.

Perhaps TomsHardware could investigate this, if enough users can confirm it (via underclocking if desired, as it is safer than overclocking).

Has anyone else noticed this ?

Anyone else care to try and confirm or re-create the 'inflated score' inaccuracy.

 

[raven_87]

Theorectially standing...your statement raises many questions in my mind.
I think I'll give it a whirl when I get home. I'll use my 7800GT to.
Just gotta re-download 05 now.

Also: To anybody who reads this thread. Underclocking is not necessarily as safe as overclocking. Lowering your speeds is about equal to over volting. Trust me, I've seen far to many people think that their GPU's are going to overheat, so they underclock to compensate not realizing this can cause DOT.

 

[RichPLS]

It uses CPU benching too, and you got 4 CPU's, x 5% = a 20% boost!!!

 

[GeneticWeapon]

To anybody who reads this thread. Underclocking is not necessarily as safe as overclocking. Lowering your speeds is about equal to over volting.

:lol:

I feel sorry for anyone who ever takes what you say seriously.

Maybe folks could underclock safely if they flash their video drivers like you recommended someone before eh?

 

[TabrisDarkPeace]

I am not overclocking my CPUs at all, (Tyan K9WE has no overclocking options), or my Reg ECC RAM.

3dMark05 only uses 2 CPU cores anyway during the CPU tests, I've watched the process. 3DMark06 uses all 4 though.

Thus: Isolated to just video card overclocking.

That is why it is so weird.

 

[TabrisDarkPeace]

Yeah, there is 'some' truth to what you say, but only if they underclock significantly, up to -15% is quite safe even at the same voltage / amperage.

Underclocking might lock a system up, but it is extremely unlikely to lead to component damage in 'this test'.

==========================

Just to isolate it further, I'll try locking CPU affinity of the 3DMark process to ensure it doesn't use all 4 cores. It only appears to have 2 isolated threads though.

 

[raven_87]

Ok, let me explain this then.

Most GPU's have 4 states of power management.

Off state: If you dont know this, your dumb..lol
2D: 3D component will sit as Idle - consuming next to nothing
Lower consumption 3D : not under full load
Full 3D consumption- power management circuit is distributing equal amounts of wattage to keep your card at peak preformance. This is why there is a large amount of QA testing, cards will run forever @ stock speeds.

Your GPU itself has a voltage/power regulation system. ( I know its a general explanation, but its true) So by underclocking either the 2D or 3D mode, you disable power management, putting large amounts of stress on the GPU.

This is a very general statement and has flaws, but eventually it is correct.
Underclock CAN and WILL damage a card. I know from experience, becuase when I first recieved my 9600XT, I underclocked it becuase I thought it was overheating (I didnt know any better) I scaled it from 500/600 to 350/350...in time it produced artifacts just like overclocking would produce.

@ GW
Hey, did you ever PM any of those "other" bad advice posts?
You said I had a ton...I just checked my box, didnt see any...maybe you havent had time? :lol:
About those drivers, you know what I meant...and your still taking it out of context. Grow up...

I really really would like for you to make some constructive posts bud.
So, your kinda wrong again. Dont believe me? Take that X800GTO you have, scale it down without a BIOS flash...see what happens.
Actually, take that 9800XT you have laying around. Underclock it daily
for about 2 weeks and see what happens, I can guarantee I'll be right.

It's 1:03AM, I'll see if I can post some results before I fall asleep.

EDIT:

@ DarkPeace

Yeah, I know a significant underclock can be damaging. So I apologize if I wasnt specific. Your right about your numbers, I just didnt want anybody to see this thread, then do something along the lines of what I didnt 3 or so years ago. A mild underclock...dont worry. To be honest, my X800pro ran 465/875 for a year and a half. Then I didnt have a system to use it in, so there's nothing considerable in a small scale.

 

[GeneticWeapon]

Actually, take that 9800XT you have laying around. .

The card you saw in the photo was a vanilla 9800.

 

[GeneticWeapon]

So by underclocking either the 2D or 3D mode, you disable power management, putting large amounts of stress on the GPU.
.

:lol: :lol: :lol: :lol: :lol:

I simply dont know what to say, but I have a feeling that I'm arguing with a very young person....and I should be ashamed of myself.

I'll seriously try to leave you alone now

 

[GeneticWeapon]

AARRRRRGGGHHhhhhhh!!

 

[raven_87]

Well, I'm not that young bud.

Reading over some of your posts, I can see why your probably 12 years old.
However you just happened to be the one to say something to begin with,
my guess is you should have left me alone quite some time ago.

EDIT:

Do I need to go and dig up more proof again?
I mean, damn, you thought I was a moron when I said a 9800SE
could be flashed to a 9800. Then I provided you with successful
evidence...you were nowhere to be found.
It takes alot to admit when your wrong, I can do it...why cant you?

 

[GeneticWeapon]

I cant do it.

Every time you post, you make something up that sounds stupid, though you think people will look at you as being knowledgable.

This post up above is a perfect example.

......also....you lie about the hardware you own.

Did you ever give me that 3dmark link?

 

[raven_87]

Sorry, somewhere around the time of family, job, another job, girl, essays, Exams, classes, putting in new electric lines, taking care of my grandfather and staying sane...I guess I somehow forgot....sorry??

Sure, I'll get you your link within 24 hours.
Not like anybody here doesnt believe me.
While I'm getting you your link, provide me with some details as to how
my post above is in anyway what your refering to.

You have some time to think, its 2:15 am...

 

[ShortyNumber45]

whhooo... at last the fight is over.

 

[TabrisDarkPeace]

I think we are getting just a 'shade' off topic.

Still, if it involves looking at technical specs, ain't really that fussed about it myself.

Should have added a warning about not underclocking more than -15% for the test, but generally underclocking will cause a system / component to lock up before anything potentially bad happens. Ditto for overclocking but I personally wouldn't recommend more than a +5% overclock (or -15% UC) to a complete beginner, or anyone concerned about warranty or longtivity of their hardware.

General concensus is that underclocking is the 'safer' way to recreate the strange (non-linear) scaling of said 3DMark results. If clock speeds where totally locked there would be far less deviation of PC's in the world.

 

[raven_87]

It didnt die, it just changed venues...its now here.

http://forumz.tomshardware.com/hardware/POST-3DMARK05-SCOREftopic-171160-days0-orderasc-150.html

 

[TheGreatGrapeApe]

So by underclocking either the 2D or 3D mode, you disable power management, putting large amounts of stress on the GPU.

I won't bother with the rest because it's entertaining and not of my making, but simply put I disagree with your theory. You're saying an underclocked card is more stressful than a card OC'ed under load, or even stock under constant load? I wouldn't be worried about power management so much as HSF controls, but then again those to can be overridden if you know what you're doing.

This is a very general statement and has flaws, but eventually it is correct.
Underclock CAN and WILL damage a card. I know from experience, becuase when I first recieved my 9600XT, I underclocked it becuase I thought it was overheating (I didnt know any better) I scaled it from 500/600 to 350/350...in time it produced artifacts just like overclocking would produce.

Sounds to me like you had a bad card, and your underclocking experience led you to an interesting, but flawed conclusion.

 

[TheGreatGrapeApe]

That is my 'great puzzle' of the day...
...Anyone else care to try and confirm or re-create the 'inflated score' inaccuracy.

Gimme a bit of time (just got back from Kananaskis skiing, gonna play poker and watch the Rumble with the gang at a co-worker's later today [sleep between now and then]),

I'll see if I can still re-create, with the current version, the errata that I created for Lars/Borsti and Futuremark. IF it still works, then I should be able to show you then undclocking has a positive effect or overclocking has a negative effect.

 

[raven_87]

No, I'm not disagreeing with you. However, what I"m trying to state is that most people believe many problems with a GPU can be remedied with underclocking in significance. When in truth they could end up damaging their card...thats all.

As for a card being under constant load while Oc'ed...hell yes thats worse,
no arguement there, just something I failed to mention.

I had the 9600XT replaced, then given to a friend as a gift.
So, I'll take something older that I have and see what happens (most likely again)

EDIT: Heatsink/Fans controls would have nothing to do with what I'm talking about. I could go into a 3rd party tool, or Nv and set my GPU's fan rotation to 100%....the same can be said for ATI tool. Its nothing a chimp couldnt do. I'm talking about the PCB itself, the integrated system that distributes voltage throughout the card depending on its current use.

Maybe this helps. If I'm going to play Doom3, yet I've underclocked my 7800GT to 270/800, and forcing the clocks, the voltage will stick be there, however in quantities that are not being used becuase the clock speeds no longer require them. Almost like having too much of something, it eventually becomes unhealthy.

Just google the subject, I'm sure you can find a number of examples.

 

[borandi]

I have an X800XL...

system specs below...

Normal 400/490 was 10800 on 3dMark03
Oc'ed 420/535 was 11250 on 3dMark03

= 4.2% increase in benchies

This was before I Oc'ed my X2 4400 from 2x2.2Ghz to 2x2.475Ghz

After: Normal 400/490 was 10880
Oc'ed 420/535 was 11400

= 4.8% increase in benchies after OC

So I dont see that 20% rise..... wish i did

 

[TabrisDarkPeace]

I don't think 3DMark 2003(SE) will do it, especially if the score is already over 10,000 points (in '03) to begin with.

Check the difference in 3DMark05 and '06 if you get the chance though.

 

[raven_87]

Ran 3Dmark05 using the reference speeds for a 7800GT = 400/1000.

Achieved 6434

I left my CPU @ 10x200 again
There is no volt mod's or overclocking...
Now I'll give a 5% OC to my GPU on both core and mem....see what happens.

Be back soon.

With 5% Overclock

6442......

I have to leave for work in 5 minutes...I'll post my thoughts later.

 

[TheGreatGrapeApe]

Ran 3Dmark05 using the reference speeds for a 7800GT = 400/1000.

Achieved 6434

I left my CPU @ 10x200 again
There is no volt mod's or overclocking...
Now I'll give a 5% OC to my GPU on both core and mem....see what happens.

Be back soon.

With 5% Overclock

6442......

Remember that the GF70 does not overclock in even amount it divides it's overclocking into steps where the geomertry engine and pixel engines overlock asymetically.

 

[raven_87]

Took me a while to realize that too :lol:

Thanks for clearing that up...