Pi-Fast tweak discussion

[I.nfraR.ed]

I found some great tweaks for pifast. They have similar effect of the super tweaks in PCMark05.

Managed to beat the current record with cpu at only 5.1GHz.

 

Before:

 

After the tweaks:

 

 

Video of the tweaked run:

 

Don't know if my calculations are right, I just wanted to show it's relatively easy to fake a pifast score without photoshopping it.

 

PS: This is not really related to the case here and the screens from John, but thought it would be worth showing.

Edited October 10, 2012 by I.nfraR.ed

[BeepBeep2]

Infra, you don't mean super cheats of PCM05?

I still don't get the text edit tweak, "legit" or not (what defines legit?), why still no level playing field?

[zeneffect]

Infra, you don't mean super cheats of PCM05?

I still don't get the text edit tweak, "legit" or not (what defines legit?), why still no level playing field?

 

why not keep this about pifast? Infra makes a very good point about pifast. the video about pifast is pretty informative though unsuprising.

 

pifast.

[BeepBeep2]

Ok, my bad.

Pifast.

 

So one could write out a file that looks like Pifast, but does not Pifast?

[zeneffect]

Ok, my bad.

Pifast.

 

Could one not write out a file that looks like Pifast, but does not Pifast?

 

it would be pretty easy to do as a bat.

[BeepBeep2]

Okay, so a wrapper would need to be written to check validity(?)

[rbuass]

@Infrared.

 

Your tweak gave a big difference in the computation numbers (computing time).

That's easy, even to brake world record without photoshop.

Is just change the windows time and back 4 or 5 seconds.

 

 

[xXSebaSXx]

Many ways to skin a cat on this one.

The less scrupulous could even make an entirely new .bat file that echoes all the text on the cmd window and makes it look like the benchmark has been run and enter whatever time they please.

[knopflerbruce]

@Infrared.

 

Your tweak gave a big difference in the computation numbers (computing time).

That's easy, even to brake world record without photoshop.

Is just change the windows time and back 4 or 5 seconds.

 

 

 

Nope, series computing time is way too low. This would definately be reported as "bugged run". You need the same relative gain in each subtest, approximately.

[zeneffect]

what about protecting the hexus.txt? i can shave 2 seconds @ 4.6 by changing the amount of digits of pi that are calculated and still see all 6 steps. im pretty sure thats how infra is demonstrating it. it would be very easy to just compile it all into a single executable which would add a layer of protection against these novice exploits.

[GENiEBEN]

what about protecting the hexus.txt? i can shave 2 seconds @ 4.6 by changing the amount of digits of pi that are calculated and still see all 6 steps. im pretty sure thats how infra is demonstrating it. it would be very easy to just compile it all into a single executable which would add a layer of protection against these novice exploits.

 

I can fix that if needed. Always wondered why it was left so exposed lol.

[zeneffect]

I can fix that if needed. Always wondered why it was left so exposed lol.

 

can you make it calculate and display a checksum for the pi text that is generated as a 2nd/redundant means of validation? a proper calculation should result in a checksum that matches a range, an improper (say if the calculation is done to 8970000 vs 10000000) should calculate to a different range due to the file size.

[I.nfraR.ed]

No, I made a "fake" pifast. Simple c++ project for the sake of showing it running, otherwise a simple bat file would output whatever you want just for the screenshot. This benchmark is the most insecure and have a lot of loopholes.

Strange I haven't thought about this before.

 

What I did is illegal ofcourse. But I made it obvious. If I had bad intentions, I could do a "normal" result that won't be suspicious.

I don't think everyone can do it, because the specific way pifast calculates the times, but it's not that hard.

[TerraRaptor]

in case of any changes to # of digits to calculate the allocated memory value and index # will change. Standart values are 61372 and 705226 accordingly. The wrapper should just calculate checksum for sub-times and the .exe of pifast.