Faking Benchmarks, The Good Benchmark List, How To Spot Cheats and More...

[Kunaak]

This post has a few purposes.

1. to show people that even the most reliable cheat free benchmarks, aren't perfect, or cheat free.

2. to create a list of known benchmarks that are either cheat free, or extremely tough to cheat on.

3. to expose Cheat methods, and to expose methods to stump a faker.

What is not to be talked about her, is HOW to fake a benchmark.
Links to programs that allow you to fake benchmarks.
Any Method, or Hint at how to fake a benchmark.
XS is not a Place for trading cheat secrets or a place that will allow free information about how to fake a benchmark.

If you know of any method for Cheating on a benchmark... Then Please, PM me, Saaya, Karnivore or another Mod here.
We are trying to create a definate way to verify Overclocks, Benchmarks, and other things that would be benificial to overclockers in General.

Any Person Talking, or Attempting To Share Cheat Programs, Or Information On How To Cheat On a Benchmark, will have thier Post Deleted immediatly, with a warning from me, or any other mod, that sees the post first.
No Mod of XS likes to hand out warnings or anything more, unless absolutly necessary, but given the nature of this Post, we must be strong on this.
You can say "well I have freedom of speech..."
I will argue with you...
one cannot walk into a movie theater, scream "FIRE", then walk out and claim freedom of speech.
so don't argue on this, you wanna cheat on benchmark... go somewhere else.

we are here to try an inform any and all members willing to listen, on ways to spot fake benchmarks, to bust cheat programs and methods, and to try to create a list of Benchmarks, CPU programs, or anything else that can be used as a tool for overclockers....

Current Methods I know of cheating on benchmarks.

PSNulling.
Wire Frame Hacks.
Pause Programs.
Windows Slow Motion Programs.
Bios Cheating.

PS-Nulling seems to be limited to a few Nvidia Drivers, and are easy to spot usually, cause they give unusually high results, by lowing the pixel shaders to the lowest possible.
Unfortunatly-one can install a PS null driver, uninstall it, install a driver over it, then still have the PS Null on by default.
So driver versions, mean little to nothing for verifying a PS-Nulled Cheat Benchmark.

fortunatly, overclockers seem to be extremely adept at spotting PSNull cheated benchmarks, and its barely a issue here at XS.

Wire frame hacks are exactly as it sounds....
they make 3d benchmarks run with not graphics, just a wire frame of what they should be.
these benchmarks are usually limited to very old videocards, and if used on a modern card, the results are so ridiculous, that you'd have to a fool to not be able to spot them.
These are of little to no risk to XS and Benchmarks just cause they are so damn hard to fake in a reasonable way. unless you wanna be the first person to try and pass a 1,000,000 point 3dmark score as legit...

Pause Programs...
these are the first of a new gen of cheat programs.
I know 2 so far.
1 seems to have been removed from google.
the other is aparantly old, and meant for something completly different.
(DO NOT TRY TO FIND IT< OR LINK TO IT HERE)
these programs, are exactly as they sound...
they allow you to "pause" windows doing anything you want.
which for a benchmark like 3dmark, it mean you can just pause, and watch your frames go as high as you want, unpause it... and then get what looks like a real score.

these are fairly easy to spot so far.
they look like super bugged scores for 3dmark2003 and 2005.
it doesn't work on 3dmark2001se.
I will explain why soon...
on aquamark 3. these cheat programs, look completly undetectable.
thats not a good thing....
they are also completly useless for sisoftsandra, pi benchmarks and such.
it's biggest advantage comes from if you looking for super bugged scores.

This next cheat method, is by far the worst of the lot.
it effects 90% of benchmarks I know of, and CPUZ, WCPUID, Sisoft sandra, and all forms of 3d benchmarks are vulnerable to it... except 3dmark2001se... again, I will explain that soon.
this Cheat method, works by slowing windows down to "slow motion".
meaning everything, runs in slow motion.

all game benchmarks-every single game benchmark you can name, is vulernable to this. from Quake3 to Doom3... they can all be cheated on with this cheat method to give nice high faked benchmarks.

Super PI is extremely vulnerable to slow motion cheating, in which you can get world records in a second if you want.

CPUZ, WCPUID, CBI and Crystal CPUID are all extremly vulnerable to these cheat methods, and can be used to fake any OC... even fooling the Checksum in CPUZ to look 100% legit.

Aquamark3 is completly vulnerable to this benchmark.
Aquamark3 seems to run, by counting the amount of time it takes your computer to render the 5203 frames of the benchmark, then divides it by the time it took, or something like this.
by slowing down windows, it makes the benchmark think the benchmark took alot less time then it really did to complete. so it's extremely easy to fake a benchmark with this cheat method.
I know of no way to fix this...

3dmark2003 and 2005, are both EXTREMELY Vulnerable to this cheat method as well, and appear to have No Protection for this kinda cheating.

what happens in 3dmark 2003 and 2005, is your videocard still renders the same amount of frames per second, just like normal, but the benchmark takes much longer to complete.

so if you card is doing say 30 frames per second, and the benchmark is 1 minute long... thats 108,000 frames a minute.
by slowing down windows to say half speed.
it will still render 30 frames per second, but with the benchmark going in slow motion, or in this case, half motion, it will then render 216,000 frames, in that same slowed down 1 minute.

neither 3dmark 2003 or 2005 have any sort of protection against this kinda cheating.

oddly enough, only 3dmark2001se has any sort of protection against this kinda cheating... again, I will explain soon, when I get to my part about programs and benchmarks that can still be considerd atleast mostly reliable or cheat free.

lastly are Bios Cheats.
Bios Cheats are something I only heard of recently....
kinda pathetic cheat method really.
but still didn't bring a smile to my face when I heard about it.

bios cheats are cheats, where say you want to take the ti 4200 record.
some people resort to buying ti-4600's. erasing the bios, and installing a Ti 4200 bios on it.
it gives a severe unfair advantage to the cheats, and is basically for only someone really desperate to take a record in some catagory.
I know of no way we or anyone could verify a cards bios...
so basically-theres nothing we can do about it.
it's just something you gotta live with, when competing in odd catagories.

below I am gonna show you a series of benchmarks I have been able to cheat on with the various cheat methods I have learned recently.
every single one of these are faked.
not real in any sense.
but theres no photoshop used to alter the results.
the pictures have a big "FAKE" written on them, and are edited alittle to remove any trace as to the program used....cause like I said, this post isn't about showing you how to cheat.
just to show you... it can be done.

then to show you programs that are still good and solid for being cheatfree... atleast for the moment.

[Kunaak]

Every Benchmark Here is Fake.

[perkam]

Bump...lol..way too long Kunaak. Could you seriously (no joke) organize that a bit better, with some bold fonts here and there for subheadings etc.

Thx.

EDIT: DANG !! You can actually fool the CPUZ Checker, whoa !! I think XS is gonna need some new software tech in a few months when dual cores come out and further complicate the issue: "Look Ma, 6 Ghz with no effort (3Ghz Dual core). LOL (Yes, I know very little about dual cores)

Perkam

[charlie]

made a sticky
well done, K!!

[Kunaak]

This is a list of Benchmarks that cannot be considerd 100% cheat free.

Sisoft Sandra.

Vulnerable to programs that allow you to run benchmarks in slow motion.
easily manipulated to show fake shots of any MHZ, any Mips, Any bandwidth.
can easily be passed for real, if someone doesn't get greedy.
hard to spot the fakes, unless just obviously faked (dude check out my 1 ghz ram at cas 2-2-2-5)

Super Pi.

This benchmark is easily manipluated, with slow motion cheat programs.
it seems to run on the windows clock timer. so if you slow the clock, the benchmark can still take the normal amount of time it should, but since the windows clock can't register the real time, it gives artificial results.
easy to pass off as real, cause it's impossible to check a faked Super Pi shot.

3DMark2003 and 2005.

Both are extremely Susceptable to Slow Motion Cheats.
cause it renders frames by the second, but the benchmark runs in slow motion.
easy to pass off as real, if your not a greedy faker.
no way I know to detect a fake, slow motion ran benchmark in either 3dmark 2003 or 2005 yet. But the solution may lie in 3dmark2001se...

Aquamark3.

easy, easy, easy benchmark to fake results on.
vulnerable to PSNulling, forcing Antialiasing off, Forcing Aniostropic filtering off, via the Driver on Nvidia cards.
Extremely Vulnerable to Cheating with Slow motion Programs.
Extremely Vulnerable to Pausing programs.
No way I know of to detect a cheated, or even bugged score on Aquamark.
Aquamark is by far my favorite benchmark, but definatly the least reliable from what I have seen in the last week.
very very easy to cheat on, with any card you can name.
especially Nvidia based cards.

CPUZ, WCPUID, Crystal CPUID, Central Brain Identifier.

versions 1.25 of CPUZ and less, aren't considerd solid or cheat free.

each of these programs can be used to fake a OC.
each one are extremely easy to fake the MHZ readout, with no need from Paintshop or anything else.
Even CPUZ's Checksum isn't able to bypass these.
Windows slow motion cheats change the mhz readout.
you can do anything, from get unbelievable 400 mhz, cas 2-2-2-5 shots.
to get somewhat believable OCs and claim it was done with Liquid Nitrogen if you wanted...

(More to the Story on this in the next part- CPUZ may be the leader in the Anti Cheating area, then any other program I know)

CPUmark99

This Program isn't cheat free.
it's susceptable to the slow motion cheat, and theres no way to spot the cheat, if used on this program.

sad day, this was one of my favorite benchmarks ever.
simple and easy to use.
unfortunatly, with nothing to indicate if its been cheated on, it's nothing you can use to verify a overclock.

RightMark CPU Clock Utility V1.3 or less.

in short, this is the worst CPU utility I seen.
even with no cheat programs, or cheat methods enabled.
the moment you enable it, it jumps all over the map for about 30 seconds, before it finally reads the proper CPU speed.

even when you try to wait for it to work, it barely worked, and read half the information of my PC wrong.
including on my laptop.

but when you try CPU cheat methods like windows slow motion cheats, it is extremely easy to mess with this program, but has the tendancy to set your CPU load to 100% when faking CPU speeds...
so atleast it's partially possible to detect a cheat when using this program.

Fast Floating Fractal Fun

by far one of the best benchmarks I ever played with.
it's a beautiful benchmark in it's simplicity, but unfortunatly easy to cheat on.
this program is vulnerable to both windows slow motion cheating, and usual manual cheating methods.

this program gets a definate thumbs up for design, and usefulness on a non cheated benchmark, but cannot be considerd cheat free, or even hard to cheat with. Great program though...


More to be added to this list as I find programs that aren't 100% reliable.

this list is just to show what programs aren't entirely reliable, good to show results from honest members with real OCs. but shouldn't be considerd 100% real if the results are questionable.

[bachus_anonym]

amazing job Kunaak :thumbsup: eye-opener for all of us....

[perkam]

Thx for the more understandable syntax Kunaak, it helps

BTW, your faked SuperPi shot is funny--> An 18s SuperPi WOW, man, what kind of knocked out system would you need to get THAT in real life.

EDIT: Though Kunaak, I appreciate, as we all do, stuff on programs that CAN be cheated, are there any 3rd party programs that are more reliable/credible then those listed above (I mean as an alternative to CPUZ, etc).

Perkam

[Sheik]

Thanks Kunaak !

I never even knew such a thing as Windows Slow Motion Programs existend Can't believe people can be so retarded.

[Tek]

Great thread Kunaak. It had to be long, as there is alot of info needed to be passed along.

First off, force the benchprograms to use internet based realtime clock checks. Meaning forcing people to have a small program within benchmark or running in background as proof of validity, that checks for incorrect timings down to miliseconds. Only way to pull that off without that being affected or tampered with as well, would be to have it linked to a server of some sort, that checks all of the neccery settings for OC.

Checksums running all over the place. In memory, clocks, OS etc.

Any cheaters should have their IP permanently banned for any future benchmarking and information should be shared openly about them. If you have done it once, you have shown the mentality regardless of reasons. It is like doping, a 2 year ban is a rediculously low punishment. If you have cheated in the least of ways, you have by default excluded yourself permanently.

3DMark01 shows what programs run in the background, but programs can be renamed. Make an avarage kb needed in memory list for these programs. Granted you could rename them to similiar programs, but running something unneeded and even hindering performance for OC could be a possible hint of cheaters. Granted, I run Firewall and Anti Virus program when I OC, but that is because I am too lazy to unplug the ethernet cable, but that is because I need the link and safety when uploading information.

Make something like the 3D Analyzer, that shows in 3D bench programs, so all of those settings are shown and logged, just like the program listing thing.

Make it obligatory when benching, that a program logs events, timings etc from boot to shutdown. That log for x number of hours, days etc. should also be tied to the programs.

Just suggestions. I hope somehow all of this is solved and a reliable way of benching is made, or at least make it way more hard for cheaters to override.

If this isn't solved or a solution(s) is found, reliable OC/benchmark is dead. No less, there isn't a future for it. Alot of the software developers that makes bench programs, should have an interest in this too, as without reliability, their programs and rankings are no more worth than a pixel on a screen.

[Kunaak]

This is a list of benchmarks, or programs that I have had ZERO Luck fooling, or Faking benchmarks with.

CPUMathMark. V3

This program isn't perfect.
the CPU speed reported with this program can easily be manipulated.

but it has 1 strong point.
you can't alter the results, with Windows Slow motion cheating.
so if your at stock, or a faked 1,000 ghz OC, it still will report the same speed everytime.

Good program, considering it's the only math program I know thats fairly cheat free, or atleast, hard to cheat one.

CPUZ Version 1.26 and Up.

this CPU reading utility can be considerd cheat free at the moment, assuming its version 1.26 or higher.
lower versions of CPUZ can be cheated on very easily, this is why this program is on both the good and the bad list, according to version.
this is the only CPU reading program that is cheater free at the moment.
hopefully this lasts...

CPUMark2.

This is a relativly new benchmark I think.
it's not too common, but I highly encourage members to atleast try this program.

this program is a benchmark that tests your CPU in 3 ways.
its seems to be almost invincible.
I mean nothing I did would vary this benchmark even by 1/10th of a point.
it did the same points every single time I ran the benchmark.
to top it off....

IT DETECTED MY CHEAT PROGRAM

see, you can submit results online.
when I tried, it detected my cheat program I was testing, and basically gave me the finger, and wouldn't let me submit my results, haha...
it instantly became a new favorite program when I saw that.

I can also say, this program never varied its CPU Speed reading, it always detected the real CPU Speed, no matter what cheat method I tried to fool it with.

This program can be used to verify a OC, atleast to some degree.
I can't claim it's 100% cheat free, as I am still testing it, but so far, it passes every test I can throw at it.

Windows Properties.

not a benchmark, or anything too interesting.
easily cheated in painshop.
but windows gets its CPU speed from the Bios.
it won't change on the fly, so it won't report anything, other then the boot up speed. but it's atleast a method of helping to verify a overclock.
should not be considerd 100% reliable, but a means to help verify a OC atleast.

Super PI Cheat Free Version.

theres a few versions of Super PI out there.
some allow cheating, some have some questionable patchs...
some are just inacurate using false timers.

this version here, is one a member here, named SNQ modded himself, and I can verify that its just as accurate as the normal version of super pi, but it does implement the best Anti Cheating efforts I know of.

this is the only PI program that should be used for anti cheating.

Here

Clockgen.

Probably the most suprising of the lot.
Clockgen seems to be completly immune to reading faked CPU Speeds.
Clockgen gets its CPU Speed by reading the FSB from the PLL on the motherboard, and the multiplier from the bios itself.
it then does alittle math, and reports the CPU Speed.

I know of no method to alter the readings, as the readings are hardware based, and have nothing to do with windows itself, other then it runs in windows.

should not be considerd 100% fool proof, but atleast highly reliable still.

A little new ABout CPUZ, Clockgen and Overclocking.

Frank, the Author of CPUZ, is by far, the best advocate I know from the programming community, that is on board with finding ways to bust cheat programs, and find ways to keep thier programs cheat free.
I have been talking with frank a few times a day, beta testing various new CPUz's sharing ideas, and trying to find ways to get a better CPUZ to everyone.
while CPUZ at the moment is not cheat free, it's still the best CPU program I know of, especially, when mixed with the checksum, and clockgen readings.
till then, we are working fast to get a better CPU Check program out to everyone

Lastly.

3dmark2001se.

you might be sick of it.
you might call it old.
you might say it's synthetic.
you may just hate it.
but plain old fact is...
I just can't cheat this benchmark, in the same ways I can with 3dmark2003 and 2005.

3dmark2001se, seems to have a built in timer.
if the benchmark should take 5 minutes, and takes 10 minutes cause of a slow motion cheat... the benchmark errors out, and gives you a "timer insanity" error.
its interesting to see this, cause it gave me a idea.
I recommended to futuremark that they should include some sort of timer on thier 3dmark2003, and 2005 programs, as well as PCmark, just like 3dmark200se has.
it's not a perfect protection against cheating, but its alot better then the no cheat protection that both 3dmark 2003 and 2005 have at the moment, and can easily be updated with some sort of timer, with a Patch.
I have since recieved a reply from someone at futuremark, who said they will look into it.
but honestly... I don't see anything happening soon, if they don't take me serious.

fortunatly, we have a few connections here, to people at futuremark, who are investigating the cheat program, and ways to bust these cheats.
so, hopefully, me bugging the hell out of them+ the people here working on it, will speed up futuremark to incorperate some sort of cheat protection in thier next patch.

This list will be updated, as I find and test more programs that I feel I cannot cheat on, with any methods I know.

Again, this list shouldn't be considerd a list of infallible programs, but it's atleast better then nothing.

[corrupt]

When I used the "multiprocessor" tweak I noticed some very weird results with cpuz, If I ran prime95 and had CPUZ open I would get some crazy numbers and all I would have to do is take a screen shot and BAM 490x12 with a A643400+ on water. Don't know if the OS tweak has anything to do with any "cheats" but it was very weird none the less.

[Tek]

CBID looks like it shows a locked startup clock. If I shut down program and change the clocks with clock gen, then restarts CBID it doesn't change the mhz/rating to match clockgen. Dunno if this is good or bad, as I remeber old versions changed the number just as CPU-Z does.

Should future benchmarks given to the community, be with HexPifast and CPUMark2 links/pics then?

Would be good if all good agree on a common bench/OC setup, at least to offer more validity.

CBID. The real clock rate is 12x200mhz when screenshots were taken, except for the one in the middle(real one taken before clockgen). Current tempurature jumps however to 255C when doing this, sometimes??? The rate reported flickers a little, as if it was reading the clocks realtime, which it obviously don't. I don't regard this is a cheat, as you would have to boot at the reported level. Just wanted to contribute a finding.

[The Doctor]

Hi Kunaak,

I have some experiance with cheats because i am running a pifast score site with some friends and i have noticed that it is possible to cheat on the hexus_pifast bench.
I wont post how its done but if you are intrested you can pm or mail me (vnrswk AT kabelfoon DOT nl).

I'm at school at the moment (screen made at school) so i cant reply to you mail. If you want a fast awnser PM me.

[Gokou]

not much of a cheat if you can spot it (ie minus computing time)?

if this can be varied then yeah, slightly more problematic

[The Doctor]

not much of a cheat if you can spot it (ie minus computing time)?

if this can be varied then yeah, slightly more problematic

It can be varied. The screen is only to show it is possible to take time of the score.

[k|ngp|n]

WTF?? Those screenies all look so real. This really sucks hard.

So sounds like 01' and cpu mark are the best for legit results?

[Snowman89]

i find it pretty funny that 3dmark01 is the oldest now used 3d benchmark and thats the only one you cant cheat on

[Tek]

It can be varied. The screen is only to show it is possible to take time of the score.

But you can you upload it and it passes their control? Didn't Kunaaks attempt "laugh" at him? If it does, then HexPiFast is out of the loop, if not, then a link to the uploaded results could be a good checkout.

[The Doctor]

But you can you upload it and it passes their control? Didn't Kunaaks attempt "laugh" at him? If it does, then HexPiFast is out of the loop, if not, then a link to the uploaded results could be a good checkout.

There's no reason why a score that's faked the way i did should not be aproved. When i make the first 6 steps 1 or 2 seconds faster its very hard to see the difrence beetween the faked score and a real score.

EDIT:
Attached 2 screens just to show how easy it is to fake.
Ofcourse 1,5 sec is a lot but i could say i was running 3-4-4-8 on the real and 2-2-2-5 on the fake one.

[Kunaak]

Docotor, can you PM Me the Information about this Cheat.

I will try to break down the severity, and how easy it is, and any way possible to detect it. but your right... from the looks of it, it might be undetectable.
let me look into it, and see what I can find.

[NWEng]

Maybe I'm getting old, but I just can't see the value of cheating on something as insignificant (in real world terms) as overclocking, other than becoming a legend in your own mind in a world of anonymous screen names. The truth will out.

[masterofpuppets]

Cheating is pathetic. If your gonna bs, you might as well bs about something less geeky. BS'ing about overclocking is for complete wierdos.

[Kunaak]

Things don't look good for Pifast.
it does appear to be cheatable.
this isn't a easy cheat method, but not hard either,
not done with a program. so atleast it's not easy.

at the moment, I will say... hexus pi fast, can still be consider 50% solid, and a grey area benchmark.
cheating through programs on it, isn't easy as far as I know.
cheating via a actual method you need to learn though, is very real.

still testing the cheat.

on a side note.

CPUZ looks to be taken off the Cheatable list, and added to the "reliable" list.
the reliable list isn't meant to say a benchmark or CPU tool is uncheatable.
just means to say, if theres a way to cheat it, I don't know it.

I was given the newest beta version of CPUZ to test, and things look REALLY good for it.
in the last 25 minutes, I tried a few methods for changing the CPU speed.
it passes 90% of the cheat methods I know.
it's still susceptable to CPU Spiking, no CPU program I know can avoid this, as it's more a windows issue then anything, and unavoidable so far, as far as windows tools are concerned.
but it's far better then any other CPU read program I know where they all failed every cheat method I know.

the new version of CPUZ is beta.
it should be out soon, but at the moment, its not available for download or anything, as it's still being tested, tweaked, and added to and such.

the next version of CPUZ should be a good solid tools for overclockers, as well as having a few new interesting features I'll leave for the release for you to find out

anyways....
this is just a preview of the new CPUZ.

this was on my laptop.
I faked the clock, and took my 750 mhz P3 laptop CPU to a retarded 3.2 ghz.
it was able to produce faked benchmarks comparable to what a 3.2 ghz P4 should do... it's able to produce over 3200 mbs in bandwidth from the ridiculous 450 mhz cas 2-2-2 ram fake reading.
its able to fake the old versions of CPUZ, 1.24.

but didn't fake the new beta version.
showed my laptops 750 mz, as the real speed, not the faked 3.2 ghz.

[Kunaak]

This are is for benchmarks that can be considerd "grey area", neither cheat free, or easy to cheat on.

Hexus Pi Fast.

Hexus Pi fast.

a cheater, is very unlikely to fake this program out.
it's not a obscure benchmark, it's rather common, alot of people run it, and compete with it. most benchmarkers with world records will atleast give a shot with this program.
not having a shot from this, is alittle odd.
but not reason enough to call a benchmark a fake.

this benchmark is vulnerable to cheating.
not through any program I know of.
but through a actual method of things you need to do to cheat the benchmark.

since the knowlege of this is something one must actually learn, and can't be easily done with a simple program cheat, I am labling this benchmark as grey area.

it's neither 100% proof that a OC is genuine.
but it's not definatly a program thats common for cheating on.
it's possible to cheat this program, but shouldn't be too common from what I saw.

I wish someone would write a new Pi Program though...

[Firefoxx]

Well done, but it's sad to see, people need to do this and not having their own skills to do a legit benchmark or overclock.