Home
A lot of acronyms....
Benchmarks
Extraction - The best way of cooling your CPU?
Linkz
RoadRunner's own 'puter
About the author
Running
Svenska delen
Pictures
Athlon o/c
 

 


 


AMD K6-2 vs K6-III comparision, TNT ref drivers

Some questions has been pretty popular amongst BBS posters since the debut of AMD’s K6-III, how does it perform in games vs the old K6-2, is it performing better only because of bad 3Dnow! implementation etc. Even heard people claim that that would be the reason why the K6-III spanks the K6-2 with so much with a TNT-card compared to a Voodoo2.

As some of you know, Nvidia’s latest BETA drivers (1.73 and 1.76) do contain some 3Dnow! Support, well definitely not too early if you ask me ;) But, since my primary setup is based on a Asus P5A and the Ali V chiset for some strange reason isn’t compatible with these drivers, I’ve been unable to test them before.

The last weekend I had the opportunity to do some benching since I had my brother’s Ax59Pro based computer in for a reinstallation. So, let’s get to work, shall we?

First of all, a few comments on the test computer: this machine was not in anyway optimized for performance, only had a very limited amount of time to do the testing on so didn’t have any time to go out chasing drivers for the Ax59Pro mobo etc. Simply take these numbers as a comparision between different drivers and CPU’s because that’s what it’s aimed to be.

Also, these tests were run with sound quality set to high, although the K6-III might shine even brighter if I went with low quality I simply didn’t this time. Somebody will always want the test run a different way but I made it this way this time. This way we get a "worst-case" scenario and not any maxed out test setup.

For making a "completely fair" comparison I didn’t o/c the card, no tweaks from def settings in either Q2 or TNT drivers (except for the refresh rate etc ;)

At the end I also benched the 1.76 drivers as my usual test setup is, TNT o/c to 120 / 130 and sound quality set to low. This to test and see what the performance difference would be on that kind of a high level of TNT o/c…

 

Test computer setup:

Mainboard: Aopen Ax59Pro
Video card: Asus V3400 TNT
Memory: 128 MB RAM, Samsung cas 2
Harddrive: Seagate 3.2 GB, Quantum 1.3 GB
Soundcard: Soundblaster 16 ISA
Desktop resolution: 1024x768 & 16 bit color mode
Write Combining was enabled with 0.48 drivers, but for some reason it simply didn’t work with 1.73 and 1.76. This might have infected the scores a bit negative in lower resolution but nothing to do about it. Maybe the new Setk6v3 would've made it possible but didn't have it last weekend so...

Q2 K6-2 vs K6-III 400 MHz High sound quality



Demo 1



Ref 0.48

640x480

800x600

1024x768

K6-2

37,7

38

34

K6-III

44,9

43,2

35,2

K6-III vs K6-2

19%

14%

4%

Ref 1.73

640x480

800x600

1024x768

K6-2

45,2

43,1

32,5

K6-III

51

46,3

32,7

K6-III vs K6-2

13%

7%

1%

Ref 1.76

640x480

800x600

1024x768

K6-2

53,4

48,1

32,7

K6-III

61,4

49,8

32,8

K6-III 300 MHz

52,5

47,4

32,7

K6-III vs K6-2

15%

4%

0%

K6-III 300 vs K6-2 400

-2%

-1%

0%

Perf diff 0.48 vs 1.76 K6-2

42%

27%

-4%

Perf diff 0.48 vs 1.76 K6-III

37%

15%

-7%

Max performance, no Q2 tweaks but TNT o/c to 120/130

Ref 1.76

640x480

800x600

1024x768

K6-III

68,3

60

40,8

K6-III TNT def

63,3

47 ?


 

Crusher




Ref 0.48

640x480

800x600

1024x768

K6-2

19,8

19,4

18,9

K6-III

22,2

22,2

21,3

K6-III vs K6-2

12%

14%

13%

Ref 1.73

640x480

800x600

1024x768

K6-2

24,7

24,7

23,3

K6-III

28,1

28

24,9

K6-III vs K6-2

14%

13%

7%

Ref 1.76

640x480

800x600

1024x768

K6-2

28,5

28,3

24,6

K6-III

33,4

32,5

25,8

K6-III 300 MHz

27,5

27,5

24,5

K6-III vs K6-2

17%

15%

5%

K6-III 300 vs K6-2 400

-4%

-3%

0%

Perf diff 0.48 vs 1.76 K6-2

44%

46%

30%

Perf diff 0.48 vs 1.76 K6-III

50%

46%

21%

Ref 1.76

640x480

800x600

1024x768

K6-III

36,2

36

31,3

K6-III TNT def (no o/c)

35,9

34

N/A

 

Ok, what do we see here? Well, first we can see that the later TNT 1.76 drivers probably do contain a nice amount of 3Dnow! Support, seems like a performance increase of 30-50% (vs 0.48 ref drivers) is the least to be expected from final TNT 3Dnow! drivers, at least this is the case in Quake2 and lower resolutions.

Second, well yes the K6-III doesn’t improve as much as the K6-2 in demo1with the 1.76 drivers, but the difference is small and since the drivers are pretty poorly optimized for higher resolutions I don’t care too much about this test.

Ok, the Crusher numbers. Here we can clearly see the true impact of bad drivers, a performance increase of as much as 50% only by switching drivers seems nice enough for me :)
And judging from this more demanding test, the K6-III needs the better drivers even more than the K6-2. But remember that this is only one test, Quake 2 isn't the only game in the world but I simply didn't have the time to bench more than one game :(

And finally, we can also see that a 300 MHz K6-III performs pretty equal to a 400 MHz K6-2, this surprises even me a bit since Q2 doesn’t make that much use of L2 cache and the CPU is running 100 MHz slower. The K6-III should also scale a lot better than the K6-2 as CPU speeds increase, unfortunately the Ax59Pro mobo didn’t allow me to bench either CPU at 450 MHz which is the speed I’ve been running both CPU’s on with my P5A mobo :(

&nbp +

A few final comments

We have waited a long time for Nvidia to release any kind of 3Dnow! Support, and now it seems like we are pretty close to the final showdown. If they don't come up with nice drivers for TNT / TNT2 the AMD part of NVidia's customers would soon go over to Matrox or 3dfx, maybe even Ati and S3.
With 3DNow! improvements like this or better, the TNT and TNT2 will become a pretty nice performing option for AMD users, but we will have to wait a little bit longer to see how the final drivers will perform. But the rumors say that they are very close to release now, if there weren't a few bugs still remaining they should already be out so in early June I think we'll see the real drivers.
And of course we still might have some compatibility issues with the Ali chipset :(

Since I didn’t run these tests with my usual setup there was no time to run any other tests, it would sure be nice to see if they have improved other kind of performance as well. But at least I’m pretty sure that both OpenGL and D3D will perform better than ever once Nvidia finally releases those drivers, stay tuned to LC BBS to find out when they comes out :)

To discuss this test and the results, let's go over to Lost Circuits BBS. At least that's where I'm gonna spend the time talking about it :)