sisoft sandra improperly benchmarks mobile XP???

[Guest]

Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

I ran 2005 sisoft sandra on my Asus A7M266-D and the arithmetic benchmark
indicates a Pentium 3 at about 800mhz. I assume their benchmark program
somehow picks up the bios reported cpu speed 800mhz since the system boots
at 800 and failes to properly compute that real speed? I am using
crystalcpuid to set the multiplier to 15x and see 2000mhz as reported by
crystal after windows boots up on the pair of mobile 2500+.

I emailed sisoft about this but have not heard back. Can anyone else
confirm that the cpu benchmarks are incorrect.

AFAICT my system is really running at 2000 as a comparable 2100+ dual AMD MP
(MSI K7D) produces a SETI unit in just over 4 hours and my pair of mobiles
do it in under 3.


--
=======================================================================
Beemer Biker joestateson@grandecom.net
http://ResearchRiders.org Ask about my 99'R1100RT
http://TipsForTheComputingImpaired.com
=======================================================================

 

[Guest]

Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

Beemer Biker wrote:
> I ran 2005 sisoft sandra on my Asus A7M266-D and the arithmetic
> benchmark indicates a Pentium 3 at about 800mhz. I assume their
> benchmark program somehow picks up the bios reported cpu speed 800mhz
> since the system boots at 800 and failes to properly compute that
> real speed? I am using crystalcpuid to set the multiplier to 15x and
> see 2000mhz as reported by crystal after windows boots up on the pair
> of mobile 2500+.

Can't be bothered typing it all again, so just read here:
http://www.ocforums.com/showpost.php?p=3625698&postcount=4

Same problem, different program. XP uses the TSC on MPX boards, so Sandra
doesn't pick up the speed change correctly. I've mentioned the problem to
them several times over the last couple of years (and even sent them correct
code) but have yet to see any improvement in their software.

A solution is to use a program to do the multiplier change prior to Windows
booting. I've written a thing to do this called BOOTFID, which currently
works as a BIOS plugin (tested in Bochs, untested in the Real World) and a
bootdisk (tested and works in the Real World on a K7D, though not
perfectly). I'm currently working (rather slowly) on a better version that
will work as an option from the Windows boot manager (and GRUB/LILO as well
hopefully) in addition to the previous methods. Send me an email if you want
a copy.

[...]

--
Michael Brown
www.emboss.co.nz : OOS/RSI software and more
Add michael@ to emboss.co.nz ---+--- My inbox is always open

 

[Guest]

Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

"Michael Brown" <see@signature.below> wrote in message
news:42cbd529$1@clarion.carno.net.au...
> Beemer Biker wrote:
> > I ran 2005 sisoft sandra on my Asus A7M266-D and the arithmetic
> > benchmark indicates a Pentium 3 at about 800mhz. I assume their
> > benchmark program somehow picks up the bios reported cpu speed 800mhz
> > since the system boots at 800 and failes to properly compute that
> > real speed? I am using crystalcpuid to set the multiplier to 15x and
> > see 2000mhz as reported by crystal after windows boots up on the pair
> > of mobile 2500+.
>
> Can't be bothered typing it all again, so just read here:
> http://www.ocforums.com/showpost.php?p=3625698&postcount=4
>
> Same problem, different program. XP uses the TSC on MPX boards, so Sandra
> doesn't pick up the speed change correctly. I've mentioned the problem to
> them several times over the last couple of years (and even sent them
correct
> code) but have yet to see any improvement in their software.
>
> A solution is to use a program to do the multiplier change prior to
Windows
> booting. I've written a thing to do this called BOOTFID, which currently
> works as a BIOS plugin (tested in Bochs, untested in the Real World) and a
> bootdisk (tested and works in the Real World on a K7D, though not
> perfectly). I'm currently working (rather slowly) on a better version that
> will work as an option from the Windows boot manager (and GRUB/LILO as
well
> hopefully) in addition to the previous methods. Send me an email if you
want
> a copy.
>
> [...]
>
> --
> Michael Brown
> www.emboss.co.nz : OOS/RSI software and more
> Add michael@ to emboss.co.nz ---+--- My inbox is always open
>

Michael: Check your email. I forwarded a response from a Sisoft engineer
who essentially admitted the problem while blaming microsoft and windows for
the clock handleing problem.

 

[Guest]

Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

Beemer Biker wrote:
> "Michael Brown" <see@signature.below> wrote in message
> news:42cbd529$1@clarion.carno.net.au...
>> Beemer Biker wrote:
>>> I ran 2005 sisoft sandra on my Asus A7M266-D and the arithmetic
>>> benchmark indicates a Pentium 3 at about 800mhz. I assume their
>>> benchmark program somehow picks up the bios reported cpu speed
>>> 800mhz since the system boots at 800 and failes to properly compute
>>> that real speed? I am using crystalcpuid to set the multiplier to
>>> 15x and see 2000mhz as reported by crystal after windows boots up
>>> on the pair of mobile 2500+.
>>
>> Can't be bothered typing it all again, so just read here:
>> http://www.ocforums.com/showpost.php?p=3625698&postcount=4
>>
>> Same problem, different program. XP uses the TSC on MPX boards, so
>> Sandra doesn't pick up the speed change correctly. I've mentioned
>> the problem to them several times over the last couple of years (and
>> even sent them correct code) but have yet to see any improvement in
>> their software.
>>
[...]
> Michael: Check your email. I forwarded a response from a Sisoft
> engineer who essentially admitted the problem while blaming microsoft
> and windows for the clock handleing problem.

It is sort of MS's problem, in that they don't provide a mechanism for the
PerformanceCounter recalibration. However, there are good reasons why you
shouldn't do PerformanceCounter recalibration so don't expect to see it
"fixed" by MS any time soon. The reason why there are systems (such as
laptops and K8 Cool'n'Quiet) that handle transitions correctly is that most
recent systems have an ACPI timer, which is used for the PerformanceCounter
instead of the TSC. Unfortunately, although the ACPI timer exists on the
MPX boards (IIRC ... I'm pretty sure I saw it in the programming manuals),
XP does not use it and instead uses the TSC.

However, it's quite easy to write your own code to do high-accuracy interval
monitoring. Essentially, at the start of your benchmark you do your own
calibration using a PLL-like method. You can easily get an accuracy of 10ppm
or so in a few RTC ticks (~1/10 of a second on an XP box is sufficient). The
code to do this is not complex, and I've even offered it to Sisoft gratis
(and got no reply). The ball is very much in Sisoft's court to get this
fixed, and I think it's quite poor that they've known about it (and known
the method to fix it) for at least two years and done nothing about it.

PS: Got your other email too, I'll respond to it a bit later when I have
some more time.

--
Michael Brown
www.emboss.co.nz : OOS/RSI software and more
Add michael@ to emboss.co.nz ---+--- My inbox is always open