Futuremark Tightens RTC Validation Check Following Tecmundo and Rbuass Findings

Futuremark is improving the detection of all types of 'gaming' of their benchmark software. Two weeks ago we reported on the removal of a ton of Hall of Fame submissions due to LOD alteration and following the Rbuass' findings shared in our forum and reported by Tecmundo, Futuremark is tightening the detection of the benchmark timer. For those who are not familiar with the problem, here is a brief overview of the Windows 8 / 8.1 / 10 RTC problem.

On August 8, 2013, we published an article detailing issues concerning the use of timer in Windows 8. The use of the RTC timer impacts the veracity of benchmark results in such a way that decreasing the base clock frequency at run-time will cause a drift in timer. In short, a second no longer consists of 1000 milliseconds, but consists of 1000 milliseconds times the decrease in base clock frequency. For example, a decrease of 5% in frequency would increase the 'time in a second' by 5%. Because the system is unaware of this, the benchmark result will seem 5% better. For more information and supporting data, I suggest you to read through our original article.

In a response, Intel hot-fixed the XTU benchmark and uses the more accurate HPET since v4.2.0.8. Early September Futuremark announced a fix for the Windows 8 RTC bug though the Systeminfo service. From that moment, the HWBOT moderators accepted 3DMark benchmark submissions on the Windows 8 (8.1/10) operating system provided a supplementary verification link as proof that the RTC was not tampered with.

On November 20, Rbuass showed in a YouTube video that it's possible to validate 3DMark submissions even if the RTC had been affected. Based on data provided by Rbuass it seems that if one stayed within a 2% margin, a benchmark result would still validate correctly. As we have learned over the past couple of days, this is possible because the timer validation process Futuremark uses on their online database compares the data from the HPET and RTC timer and flags the result as invalid if the difference is too large. For competitive overclockers 2% is a large margin as even with a single GPU in Fire Strike Extreme it can easily mean a difference of a couple of hundred points. It's important to note that any RTC drift, where the OS time is not real time, means the benchmark score is NOT valid.

Christian Ney, Head of Moderating, got in touch with Futuremark to see if anything could be done about the situation. At the moment of writing, Futuremark has already tightened the margin of the RTC and HPET timer for new submissions, meaning any result with RTC drift will be invalidated. In addition, based on historical data Futuremark will also sort through the current Hall of Fame submissions and re-evaluate the validity of the scores.

As Futuremark (in)validates the top benchmark results, the HWBOT staff will proceed checking the Global rankings as well. More information to follow.


15

Belgium Massman says:

Hope this clears up some air.

Good to see Futuremark is quick on the ball and great to see they'll be able to apply this retro-actively :)

Germany der8auer says:

Good work guys!

Poland Xtreme Addict says:

That was fast :)

Indonesia Lucky_n00b says:

Awesome! :D

Brazil Rbuass says:

I hope it helps.
Looking forward a fast FM solution.

Portugal Allgommes says:

Grande trabalho o vosso a recompensa está aí á vista de todos.

Brazil CarlosSchenckel says:

Nice ;)

United States steponz says:

Too bad its not going to change anything... lol

Brazil CarlosSchenckel says:

What you mean, steponz?

United States xxbassplayerxx says:

CarlosSchenckel said: What you mean, steponz?


I think he means that no one was using this cheat... so all scores that were valid are still going to be valid. HWBot leaderboards won't change.

Christian Ney says:

http://www.3dmark.com/fs/6522749
http://www.3dmark.com/fs/6520016
http://www.3dmark.com/fs/6522848
"Time measurement inconsistencies detected during benchmark run. This result is not eligible for Hall of Fame or overclocking contests."

In the Top 100 FM HoF database (including all rankings of currently supported benchmarks, about 5800 results), only 157 results were affected. There are very few users with multiple match. It is safe to say nobody was exploiting this on purpose.
There are as few as 5 results from those 157 that have been posted on HWBOT. Only one being a top score.

Sweden elmor says:

Doesn't seem to detect this, from 126.22 bclk down to 124 http://www.3dmark.com/fs/6564667

Christian Ney says:

This one was submitted before the fix went into production. Server is checking every new score and finished checking HoF (5800 results). Will take some time before all old results outside HoF are checked.

Sweden elmor says:

Christian Ney said: This one was submitted before the fix went into production. Server is checking every new score and finished checking HoF (5800 results). Will take some time before all old results outside HoF are checked.


Ah ok, thanks. I'd say 157/5800 is not insignificant though, any data on which boards are affected? Trying to find out why this is only happening on SOC Champion and not R5E.

Bulgaria FinFinFTW says:

elmor said: Trying to find out why this is only happening on SOC Champion and not R5E.


+1
I will subscribe for that :D

Please log in or register to comment.