8800GTX LoD Testing for 3D Mark2001SE

[BenchZowner]

Since I'm a 3D Mark2001SE addict, I decided to go...hardcore and test some LoD settings with the 8800GTX for it.
LoD Range tested = from 0 to + 9 [ at 0.1 increments ]

I'll update this post with more results, more tests, testbed config ( system specs, hardware & software ), BIOS Settings, etc and more.
At first I'll run each test with integer LoDs, then I'll test them with FP LoDs.


I've run each test with each LoD setting 3 times and averaged the results ofcourse.
In Car Low ( a.k.a. "the good run" test ) I've excluded the "good runs" from the average, but I'll include them in the results table.

----------------------------------------
--------------System Specs-----------
----------------------------------------
Motherboard: Asus P5W DH Deluxe ( BIOS v1602 ) [ only crucial benchmarking hw enabled ]
CPU: Intel Core 2 Extreme X6800 @ 3.3GHz ( 300MHz FSB , 11x Multiplier )
RAM: OCZ PC2-6400 S.O.E. Urban Elite 2GB Kit @ 600MHz ( FSB/DRAM = 1/1 - Timings 4-4-4-8 , Write Recovery = 3 )
VGA: Twintech GeForce 8800GTX ( Stock Clocks )
OS: Windows XP SP2 ( only basic services started, disabled themes )
Chipset Drivers: Intel Chip. Inst. Util. v8,1,1,1010
VGA Drivers: nVIDIA ForceWare v97.92

Car Low:
LoD 0: Average = 750.66 , Runs = 756.4 749.6 746
LoD 1: Average = 750.16 , Runs = 749.5 795.6 752 749
LoD 2: Average = 749.83 , Runs = 752.4 793 749.9 747.2
LoD 3: Average = 751.66 , Runs = 750.3 751.3 799.8 753.4
LoD 4: Average = 751 , Runs = 750.1 752.6 750.3
LoD 5: Average = 750.56 , Runs = 746.6 797.9 752.1 753
LoD 6: Average = 747.43 , Runs = 746.3 748.8 747.2
LoD 7: Average = 746.8 , Runs = 790 747.1 747.7 745.6
LoD 8: Average = 744.03 , Runs = 746.7 744.5 740.9
LoD 9: Average = 744.9 , Runs = 742.2 743.7 748.8

Car High:
LoD 0: Average = 217.6 , Runs = 217.8 216 219
LoD 1: Average = 218.6 , Runs = 218.3 219.2 218.3
LoD 2: Average = 218.56 , Runs = 218.8 218.1 218.8
LoD 3: Average = 217.03 , Runs = 216.3 217.6 217.2
LoD 4: Average = 218.76 , Runs = 220.1 218.4 217.8
LoD 5: Average = 218.53 , Runs = 218.9 218.2 218.5
LoD 6: Average = 219.4 , Runs = 219 220 219.2
LoD 7: Average = 217.16 , Runs = 218.7 216.4 216.4
LoD 8: Average = 218.53 , Runs = 219.2 217.2 219.2
LoD 9: Average = 219.86 , Runs = 220.8 219.5 219.3

Dragothic Low
LoD 0: Average = 1010.1 , Runs = 1014 1007.4 1008.9
LoD 1: Average = 1010.36 , Runs = 1010.7 1010.1 1010.3
LoD 2: Average = 1010.4 , Runs = 1010.5 1010.3 1010.4
LoD 3: Average = 1009.36 , Runs = 1009.7 1009.4 1009
LoD 4: Average = 1010.33 , Runs = 1010.6 1010.1 1010.3
LoD 5: Average = 1010.36 , Runs = 1011.4 1010.4 1009.3
LoD 6: Average = 1008.16 , Runs = 1008.7 1007.7 1008.1
LoD 7: Average = 1006.4 , Runs = 1006.5 1005.6 1007.1
LoD 8: Average = 1011.73 , Runs = 1011.4 1012.1 1011.7
LoD 9: Average = 1007.66 , Runs = 1007.3 1009.6 1006.1

Dragothic High
LoD 0: Average = 540.9 , Runs = 540 540.8 541.9
LoD 1: Average = 540.36 , Runs = 540 540.9 540.2
LoD 2: Average = 540.73 , Runs = 541 539.3 541.9
LoD 3: Average = 540.13 , Runs = 539.5 540.3 540.6
LoD 4: Average = 540.26 , Runs = 540.6 540 540.2
LoD 5: Average = 540.8 , Runs = 539.9 541.4 541.1
LoD 6: Average = 539.8 , Runs = 539.6 540.3 539.5
LoD 7: Average = 539.73 , Runs = 540.5 538.8 539.9
LoD 8: Average = 539.56 , Runs = 541.6 538.1 539
LoD 9: Average = 539.9 , Runs = 539.4 540.1 540.2

Lobby Low
LoD 0: Average = 782.26 , Runs = 779.4 785.4 782
LoD 1: Average = 782.56 , Runs = 784.4 779.1 784.2
LoD 2: Average = 782.36 , Runs = 783 780.6 783.5
LoD 3: Average = 782.4 , Runs = 783.1 783.7 780.4
LoD 4: Average = 780.9 , Runs = 779 779.6 784.1
LoD 5: Average = 783.23 , Runs = 783.7 782.9 783.1
LoD 6: Average = 781.33 , Runs = 780.4 783.3 780.3
LoD 7: Average = 780.06 , Runs = 777.7 781.5 781
LoD 8: Average = 781.76 , Runs = 783.4 781 780.9
LoD 9: Average = 780.23 , Runs = 783.1 777 780.6

Lobby High
LoD 0: Average = 315.13 , Runs = 315.3 315.9 314.2
LoD 1: Average = 315.3 , Runs = 315.3 316.1 314.5
LoD 2: Average = 315.5 , Runs = 315.2 316.4 314.9
LoD 3: Average = 314.8 , Runs = 314.6 315.1 314.7
LoD 4: Average = 313.96 , Runs = 315.6 313.6 312.7
LoD 5: Average = 314.3 , Runs = 314.5 314.8 313.6
LoD 6: Average = 315.76 , Runs = 317.1 315.7 314.5
LoD 7: Average = 315.13 , Runs = 315.4 315.4 314.6
LoD 8: Average = 314.23 , Runs = 314 314.1 314.6
LoD 9: Average = 315.33 , Runs = 315.9 315.2 314.9

Nature
LoD 0: Average = 535.16 , Runs = 535.2 535.1 535.2
LoD 1: Average = 559.6 , Runs = 559.6 559.5 559.7
LoD 2: Average = 564 , Runs = 564.2 564 563.8
LoD 3: Average = 570.16 , Runs = 570 570.2 570.3
LoD 4: Average = 599.56 , Runs = 599.9 599.8 599
LoD 5: Average = 621.73 , Runs = 621.8 622 621.4
LoD 6: Average = 616.76 , Runs = 619.7 615.5 615.1
LoD 7: Average = 607.53 , Runs = 607.8 607.2 607.6
LoD 8: Average = 598.36 , Runs = 596.7 601.7 596.7
LoD 9: Average = 583.56 , Runs = 583.6 584 583.1

Once I finish with the FP LoDs, I'll go for a quick AquaMark 3 LoD testing as well
Actually I'm thinking of playing with the decimals in-between the "beneficial" LoDs only ( lets say 4.1 to 5.9 for Nature ).

Your ideas ?

 

[BenchZowner]

Updated

 

[thilanliyan]

Forgive me for asking but what exactly does this show?? How much detail is kept and the performance hit it takes for keeping that detail??

 

[BenchZowner]

In benchmarking, you're hunting for the highest scores possible.
You don't care about Image Quality, so you mess with some stuff, like MipMap's LoD ( Level Of Detail ).
With this test you can find out which LoD setting gives you the best performance in each 3D Mark2001SE test.

Example:
If you run 3D Mark2001SE without messing with your LoD setting ( which defaults to 0 ), you'll get 50.000 points ( lets say ).
If you run 3D Mark2001SE with LoD tweaking ( different LoD setting for each test ) you'll get a better score, 53.000 points ( lets say ).

You can use RivaTuner to change the LoD setting or mess with nVIDIA's settings in the registry.

 

[allies]

This is why I don't understand "competitive benchmarking"...

 

[Extelleron]

You should do the same thing with a more recent benchmark like 3D Mark 06.

3D Mark 2001 is so CPU limited the results don't really matter; the 8800GTX isn't being stressed at all no matter how high you set the detail. Seeing how this impacts scoring on 3D Mark 06 would be interesting however.

 

[thilanliyan]

Originally posted by: BenchZowner
Example:
If you run 3D Mark2001SE without messing with your LoD setting ( which defaults to 0 ), you'll get 50.000 points ( lets say ).
If you run 3D Mark2001SE with LoD tweaking ( different LoD setting for each test ) you'll get a better score, 53.000 points ( lets say ).

You can use RivaTuner to change the LoD setting or mess with nVIDIA's settings in the registry.

Isn't it just getting a higher score because it is rendering less?

 

[BenchZowner]

@Extelleron:
3D Mark05/06 show just a little ( almost considered none ) improvement.
While 3D Mark2001SE & AquaMark 3 show some.

3D Mark2001SE is far more realistic than any other 3D Mark, as it's really stressing what it should.In gaming the CPU & VGA are stressed out at most, and RAM just a little bit, that's exactly what 3D Mark2001SE does ;-)

In a "real" bench, with a Core 2 Duo over 4,5GHz the 8800GTX shows its muscle against the other cards.

@thilan29:
Yup, not only less, but even with lower detail.

 

[allies]

I don't understand that at all.. it's like seeing how fast a car can do a quarter mile, but reducing the length of the track so it's X feet less than a quarter mile.

 

[BenchZowner]

@allies:

Well apart from that ( LoD tweaking ) overclocking is 'fair'.
You can always run it without LoD, just overclock your system and check the difference ( in performance ).

The thing with LoD, is since most people use it, you've got to use it too, if you want to stay in the competition ( and there's some big competition going on, check hwbot.org )

 

[BenchZowner]

I've just finished testing the decimals in-between the "beneficial" LoDs, and higher LoDs ( 9.x - 12 ) as requested by massman and AquaMark 3 as well

My final numbers:

---------------------------
-----3D Mark2001SE-----
---------------------------
Car Chase Low = 3.0
Car Chase High = 9.1
Dragothic Low = 8.0
Dragothic High = 5.0
Lobby Low = 5.0
Lobby High = 6.0
Nature = 4.9

---------------------------
------AquaMark 3-------
---------------------------
Just like 3D Mark2001SE, I've run each LoD test 3 times, and averaged the results.
Unfortunately, the highest score & average was achieved with...LoD 0.
LoDs from 0 to 12 were tested.

-----------------------------
Now...who's willing to test all 49 possible test sequences ?