3D Graphics Mark formula

[kvasnyj]

What formula is used for calculation 3D Graphics Mark?
I would like to exclude simple and medium results and recalculate 3D Graphics Mark

[David (PassMark)]

It is bit complex becuase we tried to match the 'mark' values up with previous versions of the software. But effectively it is an average of the 3 values.

I don't think there is any sensible way to do what you want.

[kvasnyj]

There is only one reason: if I start PT6 in same PC 10 times I get next standard deviation of result:
graphics 3d - simple 89,23%
graphics 3d - medium 85,03%
graphics 3d - complex 0,38%
cpu mark 0,08%
2d graphics mark 1,18%
memory mark 0,06%
disk mark 5,77%
3d graphics mark 84,49%
passmark rating4,61%

The result is checked up on various computers.

[David (PassMark)]

Standard deviation is the square root of the variance. It is not measured as a percentage. So your post is confusing.

Maybe you mean that 3D results from some of your runs varied by 89% from some of the other runs. e.g. Run #1 had a frame rate of 100. Run #2 had a frame rate of 11 (an 89% difference). If this is a case you have a problem with your system or test methodology.

[kvasnyj]

I took ratio of standard deviation to average value, that’s why the result is in percents.

We make computers and use PerfomanceTest for examination of quality of assemblage more than 5 years. The examination makes clean system , that’s why problems must not be. We always make simple examination in transition on new version: starting of test from the command line on one machine for some times.

There are the results for Perfomance Test 6:
Sample #1 (Intel(R) Pentium(R) D CPU 2.80GHz)
Test Res1 Res2 Res3 Res4 Res5 Avg Dev Dev%
graphics 3d - simple 12 12 856 804,8 857,5 508,46 453,70 89,23%
graphics 3d - medium 7,1 7 168,8 165,6 164,9 102,68 87,31 85,03%
graphics 3d - complex 28,8 28,9 28,8 28,6 28,8 28,78 0,11 0,38%
3d graphics mark 15,4 15,4 337,2 319,7 336,4 204,82 173,06 84,49%

Sample #2 (AMD Athlon(tm) 64 Processor 3500+)
Test Res1 Res2 Res3 Res4 Avg Dev Dev%
graphics 3d - simple 816,3 797,1 785,2 7,7 601,575 396,1239316 65,85%
graphics 3d - medium 135,2 136,2 135,1 4,6 102,775 65,45188436 63,68%
graphics 3d - complex 31,6 31,3 31,8 31,2 31,475 0,275378527 0,87%
3d graphics mark 314,6 308,7 304,7 13,9 235,475 147,7726266 62,76%

Script:
# Set the test preferences
SETDISK "C:"
SETDURATION 10
SETCOMPUTERNAME "COMPUTER No.000000"
SETCPUTESTPROCESSES 10
SET3DMAX
SUPPRESSWARNINGS ON
# Run the test 3 times
LOOP 3
{
RUN CPU_ALL
RUN G2D_ALL
RUN G3D_ALL
RUN ME_ALL
RUN DI_ALL
}
# Save TXT and HTML reports for the last test run
EXPORTHTML "c:\sysprep\tests\pt6\perfres6.htm"
EXPORTTEXT "c:\sysprep\tests\pt6\perfres6.txt"
EXIT

What may be wrong?
Excuse my English

[David (PassMark)]

In your initial post you said you did 10 runs. But the script you posted loops only 3 times. However the numbers you posted seem indicate that you did 6 runs or maybe 7 or 12 runs depending on if you are using the comma as a decimal point (European style) or if your commas are number separators (USA & Australian style). So I am still pretty confused.

If I try to take the first line of numbers and assume these were meant to be frame rates, as follows,
12.0, 12.0, 856.0, 804.8, 857.5, 508.46
then take the standard deviation of this series, it is 405.8, which doesn't match your numbers. Now does the average of 508.46. Nor could I get any ratio to come out at 89%.


So I am still pretty confused.

But nevertheless, a frame rates of 4 and 7 are NOT normal at all. Does it really look this slow? These low frame rates should be really obvious when you watch the test execute. Maybe the machine has some other background tasks running, maybe it was your AV software. Without examining the machines I can't tell you why the result was so poor.

[kvasnyj]

I perfomanced test different times, 3,5,10, so count is different
Last value 508.46 is avarange value. Deviation is calculate for 12.0, 12.0, 856.0, 804.8, 857.5
No other service in buckground.
I dont known how is look like, becouse test perfomance in termo-chamber
It's true for different PC.

[David (PassMark)]

If you can't see the machine during the test, how can you be so sure about what is happening? Can you pull the machine out for 30min and watch it.

Or if you are doing testing in a temperature chamber, maybe the CPU or graphics card overheated and reduced it's own clock speed or shut themselves down?

Do both these machines share the same graphics card? Did you try another card? Are the video drivers up to date?