beta V1.0.3b - fixed downclocking bug

[Massman]

It's been a pain in the ass to debug, discover, solve and re-code this bug. For those who don't know, this was the bug we were dealing with.

Originally Posted by Massman

The problem with Unigine is that the downclocking is not really speeding up the benchmark, it's just making it run for a longer period (so effectively changing the benchmark). We figured this out just a few days ago.

- Unigine score is derivative of the FPS
- The FPS in unigine are equal to "num_frames_calculated" divided by 260, as 260 is the predefined length of the benchmark.
- Adjusting the BCLK in OS makes the benchmark run longer or shorter (depending on downclock and upclock). In normal circumstances, the benchmark runs 260s. Adjusting the BCLK downwards by 10% means that the benchmark will take 260x1.1=286s.

The problem is that Unigine didn't take into account that the benchmark length can vary. So, regardless of how long the benchmark took, the total amount of rendered frames will always be divided by 260. The more time a benchmark takes, the more frames you can render ... so downclocking increases the score.

We've already given the feedback to Unigine, but they've not been very responsive so far. I'm actually thinking about disabling the benchmark from HWBOT as long as this bug is present .

Now, since we discovered the above, we've (and I actually mean MCQ8 mainly) been working both implenting a solution, which was already quite difficult, and working on more security to prevent abuse of the solution, which took roughly 2 weeks to get done.

The result is a V1.0.3B public beta for everyone to test and try to cheat if necessary.

Note: online submission has been disabled, so you need to SAVE the result and submit manually. Based on the feedback, we'll launch a V1.0.3 final release soon!

Download link: http://91.121.148.119/downloads/beta...en V1.0.3b.exe

[T_M]

This is great news.
Now i wanna see people try to catch my ATI runs without this bug :P

[Massman]

Anyone who tested this new version for bugs?

[gekke-gerrit]

i don't have the best way to show it but its something,

it didn't work fine on my windows maybe win7 64bit

i7950
5870 lightning
asus p6t7

200*21= 4200Ghz

170*21= 3570Ghz



bug with submitting score?

[Massman]

Online submission has been blocked for this version of the application, because it's a beta version. So that's normal.

Also, you go up 9 points by increasing the CPU speed from 3570p to 4200p ... I think that's also good (downclocking usually gave a serious boost in score)

[gekke-gerrit]

Originally Posted by Massman

Online submission has been blocked for this version of the application, because it's a beta version. So that's normal.

Also, you go up 9 points by increasing the CPU speed from 3570p to 4200p ... I think that's also good (downclocking usually gave a serious boost in score)

Not good so for us, we (nedernakker & me) are organising a overclocking competition with the 450GTS and we thought this was the perfect benchmark but its cpu dependent, so I'm afraid we have to switch back to nature benchmarks or do you got an idea?

[Massman]

You ... went ... up ... 9 points. 18% increase in CPU speed = 1.1% increase in score

Yeah, it's most definitly CPU dependent.

[Massman]

Moved discussion to the main bugs thread (this one is about the beta version)