GPU-Z "validation" easy to fool?

[Z999z3mystorys]

It seems that it's quite easy to fool GPU-Z into thinking that you've gotten much higher clock speeds than you really have.

I'd think it best not to say how this is done, but is this going to be fixed? There are impossibly high clocks recorded with this now, making comparing the GPU-Z validation records pointless. I just set a new record for my card by "cheating" this way, and tt was way too easy.

 

[Kursah]

I know it reads driver clocks instead of hardware clocks on my GTX 260, which depending on which "clock step-range" i'm in could be lower or higher than my actual clocks. Though it's not a huge difference ever. But you should post your validation link, and explain further how you could cheat. May be useful to W1z in sorting it out.

 

[Z999z3mystorys]

http://www.techpowerup.com/gpuz/dbda5/

1200 core with a 2000 (4000 effective) memory clock...

How was this done? one hyphenated word: Auto-underclocking

Basicly some GPUs will underclock themselves to a low speed and only ramp up to there listed speed when a 3D Application is started.

So you can set your speed sky high, not run any 3D programs, and the GPU will stay at it's underclocked speed, but GPU-Z will think it's running at full speed. GPU-Z need a monitoring system to monitor real Clock speeds and report that, not the set clock speeds.

 

[Kursah]

LoL nice results haha. Yeah another user here by the name of Solaris17 was doing some heavy OC-ing on an 8600GT(S), found out GPU-z was misreporting, that was months ago tho. But with submissions like this maybe that can be caught and fixed.

 

[Z999z3mystorys]

Woo new record, Riva tuner wouldn't go any higher....

http://www.techpowerup.com/gpuz/ku53c/

 

[FordGT90Concept]

One can always fake a score by memory editing...

There is no way to 100% guarentee it isn't happening. The only way I can think of is to establish a secured connection with the validation server and have the validation serve initialize and record the measurements. That way, the window to modify the result client-side is very slim if not completely non-existant unless the encryption is broken on the connection.

Still, the value could be altered at it's origin. If they aren't the actual clocks the card is using as in if the clocks were changed, the card could be damaged. I have no idea how you'd go about finding those memory addresses assuming they are even accessible (not protected by the BIOS and/or OS).

I'll have to ponder that one...

 

[Binge]

I don't think it matters so much since you can't actually use the application.

If you were to use what GPU-Z is actually for then you wouldn't be able to
show those clocks and have anyone take you seriously.

Take Vantage for example. You complete it at a clock and then bring up
CPU/GPU-Z for a screenshot, but you decide to tweak about the graphics
results. Your vantage score looks like crap because someone with your
same clocks pre-tweak would get the same scores.

I never see anyone take validation seriously as we usually trust each other.

It's not like it's hard to sniff out a liar with so many enthusiasts.

-my 2 cents

 

[Z999z3mystorys]

Well I guess my main point is that it's way to easy, if you have to do memory editing to fool the system, that's one thing, but if you can just stumble on this without really trying, like you can with this method, that's another matter.

 

[Z999z3mystorys]

Anyway my main concern is how can you compare if people can cheat this way? That is to say, the absurdly high scores are easy to ignore, but the borderline ones are the ones I'm concerned about, did they really get their core clock to 800, or did the fake it? I mean I know it's not that important in the end, but in the spirit of competition people like me want to compare scores with a fair degree of certainly that the scores aren't faked.

Take 3DMark, people compete for high scores there, and we can be reasonably believe that the scores are real.

Basicly having something that so easy to cheat at makes competing pointless.

 

[LiveOrDie]

you can fall it ezy but people know what clocks are real and not real, u can ezy just oc really high just to get a screen shot then drop it back down even noing that the card would never run that high.

 

[Binge]

Anyway my main concern is how can you compare if people can cheat this way? That is to say, the absurdly high scores are easy to ignore, but the borderline ones are the ones I'm concerned about, did they really get their core clock to 800, or did the fake it? I mean I know it's not that important in the end, but in the spirit of competition people like me want to compare scores with a fair degree of certainly that the scores aren't faked.

Take 3DMark, people compete for high scores there, and we can be reasonably believe that the scores are real.

Basicly having something that so easy to cheat at makes competing pointless.

but I would never compete with people in GPU-Z validations. lol

Then again I don't compete. I only play along. The numbers mean one thing but as long as my name is up on the list of entrees I'm happy.

 

[W1zzard]

there has never been any effort on establishing some kind of 100% validated scoring system into gpuz. this may be something to work on in the future. the 2d/3d clock trick has been around since the first days since gpuz is out