Personal computing discussed

Moderators: renee, Starfalcon

 
credible
Gerbil XP
Topic Author
Posts: 499
Joined: Fri Oct 07, 2011 7:47 pm

Breaking! Windows 8 Benchmark Records No Longer Accepted

Sun Aug 18, 2013 12:35 pm

What do you all think about this, seems like they have a point to be sure, its a little beyond my pay grade but I get what they are talking about.


http://hwbot.org/news/9824_breaking_windows_8_benchmark_results_no_longer_accepted_at_hwbot
 
jihadjoe
Gerbil Elite
Posts: 835
Joined: Mon Dec 06, 2010 11:34 am

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Sun Aug 18, 2013 1:16 pm

Wow this is certainly the first I've heard of that, and it seems really silly that the OS doesn't figure out the BCLK first before using it in calculations for the RTC.
 
Waco
Maximum Gerbil
Posts: 4850
Joined: Tue Jan 20, 2009 4:14 pm
Location: Los Alamos, NM

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Sun Aug 18, 2013 4:50 pm

Holy crap...if that's true, that's an idiotic RTC. This has implications that run much deeper than simple benchmark scores!
Victory requires no explanation. Defeat allows none.
 
ChronoReverse
Gerbil Elite
Posts: 757
Joined: Wed Dec 12, 2007 4:20 pm

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Sun Aug 18, 2013 4:59 pm

Do they have HPET off or something? I'm on Windows 8, overclocked and my clock is perfectly synced.
 
jihadjoe
Gerbil Elite
Posts: 835
Joined: Mon Dec 06, 2010 11:34 am

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Sun Aug 18, 2013 11:40 pm

^ The article mentions the problem is with downclocking, not overclocking.
 
ChronoReverse
Gerbil Elite
Posts: 757
Joined: Wed Dec 12, 2007 4:20 pm

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Mon Aug 19, 2013 12:56 am

jihadjoe wrote:
^ The article mentions the problem is with downclocking, not overclocking.

Nope, it specifically says and even shows a video with it happening with _both_ overclocking and underclocking.
We can also demonstrate the Windows8 RTC problem with the Windows Time. Have a look at the videos below. In one of the videos, we underclocked the base clock frequency by 6%. After 5 minutes, Windows Time was already 18(!) seconds behind real time. When overclocking the base clock frequency, we can see the opposite effect. We overclock by roughly 4%, and after two minutes, Windows Time is 3 seconds ahead of real time.



I submit that they rushed to a conclusion before thorough testing. Of course, this being bad new for Windows 8 if true, this is going to turn into "common knowledge" without verification.
 
Glorious
Gerbilus Supremus
Posts: 12343
Joined: Tue Aug 27, 2002 6:35 pm

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Tue Aug 20, 2013 9:16 am

I posted about the ultimate source of this problem 7 years ago:

Glorious wrote:
The crux of this really comes down to the problem that there isn't any reliable way to get a high-precision, high-accuracy timer on Windows. Depending on your windows boot switches and hardware combination the HAL might actually be using anything from the TSC to some PIC off the PCI bus on the motherboard.


viewtopic.php?f=2&t=41625&p=571319&hilit=RDTSC#p571319

The ultimate issue here is that the windows API has to get a timer from somewhere, and thus depending on the situation the HAL can pick any number of different things. We were never guaranteed an accurate timer, and simply relying on the generic API timers means you are implicitly taking whatever the system gives you.

Now it seems that Windows 8, probably in an attempt to accommodate strange hardware/save power, either doesn't default to relying on the RTC in the HAL, or at least it didn't with that particular test system. Instead it relies on the TSC or some other similar cycle-counting scheme.

What I am emphasizing is that this was always a possibility. It's not unique to windows 8, although windows 8 *might* explicitly default to cycle-counting these days. The point is that windows 7 could be rigged the same way, or due to the buggy hardware it could end up that way.

At any rate, given the reliance of on cycle counting, the abnormal results are completely understandable: changing the base clock like that while the system is running is probably not handled by the OS, nor should we expect otherwise. If the timer-calculation portion of the code assumes that the base clock remains the same as it was at boot time, but it was actually adjusted to be faster without actually notifying the OS, of course the timer will run too fast. It'll run faster directly commensurate with the clock adjustment. The algorithm is correct, you've just mucked with the inputs without the system knowing. GIGO.
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Tue Aug 20, 2013 9:57 am

I wonder if there is a boot-time option to force Windows 8 to use a particular timing source. Might be worth looking into.

I recall doing something like this a few years ago (this was on Linux, not Windows though...) when the system clock was running way too slow. Setting a boot-time option to force the use of the HPET instead of ACPI timer (or vice-versa, I forget which...) cured the issue.
Nostalgia isn't what it used to be.
 
Glorious
Gerbilus Supremus
Posts: 12343
Joined: Tue Aug 27, 2002 6:35 pm

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Tue Aug 20, 2013 10:00 am

JBI wrote:
I wonder if there is a boot-time option to force Windows 8 to use a particular timing source. Might be worth looking into.


There are new bcdedit settings that look exactly like that, I don't have windows 8 or I'd try them out myself. I posted about that same idea on the frontpage:

Glorious wrote:
There are new options in bcdedit that look interesting:

useplatformtick
tscsyncpolicy

It would be interesting to see what enabling them would do in reference to this problem.

Setting the venerable useplatformclock would be interesting too.


Heh, I kind of doubt you have windows 8 too JBI. :P

Maybe someone here can try it out.
 
Ryu Connor
Global Moderator
Posts: 4369
Joined: Thu Dec 27, 2001 7:00 pm
Location: Marietta, GA
Contact:

Re: Breaking! Windows 8 Benchmark Records No Longer Accepted

Tue Aug 20, 2013 10:11 am

Those are the wrong commands to do what you're asking.

There is a way to force Windows to use the HPET timer, but it is unwise as HPET is no longer the best timer in modern PCs. Invariant TSC is a superior solution.

bcdedit /set useplatformclock true

Edit: Microsoft discourages the use of the three items we've listed. They are marked for debugging use only.

useplatformtick would break the OSs new power management features.

tscsyncpolicy should have a negative performance impact. I have some QPC benchmarks I could test that theory with later.
All of my written content here on TR does not represent or reflect the views of my employer or any reasonable human being. All content and actions are my own.

Who is online

Users browsing this forum: No registered users and 1 guest
GZIP: On