3D Acceleration Benchmarks
These tests were made on the now-classic Abit and Intel Celeron 300A overclocking combo at 450 MHz.
The motherboard had 128mb Micron CAS 2 PC100 SDRAM and an IBM Deskstar GXP Ultra DMA/33 hard drive.
All benchmarks were done in 16-bit color unless otherwise specified.
Quake II was tested with demo1.
The Permedia 2 was slightly blurry in high resolutions at high refresh rates. It also screws up Max's viewports when you change resolutions. After looking at the displays generated by the G200 and the i740, it's a pain to switch back to the eye-straining Permedia 2. I hope the Permedia 3 produces a sharper image. Despite the blurriness and the benchmarks, in my real world tests I had to conclude that for 3D Studio Max users the Permedia 2 is still the best low-cost option.
The Intel i740 was very sharp and pretty, but caused several annoying anomalous effects, and frequently paused or locked up in games. In my tests in 1280 X 1024 with the Starfighter it caused the display to be corrupted at 85 Hz refresh rate, but worked fine at 75 Hz... until I did the Max tests, which is when it caused another lock-up. It's smooth frame rates and crisp display are marred by poor Max performance and unreliability. This unreliability may relate to the fact that it gets very hot despite the heatsink.
The G200 had beautiful, sharp picture quality even in high resolutions. Unfortunately, the OpenGL support under Windows NT seems to be non-existant. Max shows Microsoft's OpenGL driver as active, and Maya doesn't see any 3D accelerator. Hey, Matrox, what's the hold-up?
The Diamond Viper V550 Riva TnT card was surprising in several ways. First of all, it was just as sharp as the i740 and almost as sharp as the G200. I don't know what the deal is with those blurry AccelStar II cards. The game performance was as excellent as expected, but with Windows NT it had problems. The Diamond drivers caused display anomalies in 3D Studio Max, so I ended up using STB's drivers for the benchmarks. Neither prevented lock-ups, so I tried several different ones and the most stable one turned out to be from Hercules. Max performance was better than expected. The card would not use any TnT driver in Windows NT if an IRQ was not assigned to it, and if the bus is not set to 2/3 instead of 1/1 (100 MHz) games would not run.
If you want a TnT card, buy an Asus card with video in/out for around $150 and use the Hercules drivers in Windows NT when you don't need to use the video features.
 
 

Software Only

GLint Permedia 2 8mb 
(AccelStar II)

Intel i740 8mb
(Real3D Starfighter)

Matrox Millenium G200 
16mb SGRAM

nVidia Riva TnT 16mb
(Diamond Viper V550)

Quake II 
640 X 480

 22.2 fps Permedia 2
26.8 i740
26.4 G200 

18.8 fps 

 52.1 fps 

beta OpenGL ICD
30.8 fps 

58 fps 

Quake II
800 X 600

16.9 fps Permedia 2
20.2 fps i740
20 fps G200 

13.7 fps 

38.2 fps 

- 

47.1 fps 

Quake II
1024 X 768

11.6 fps Permedia 2
14.2 fps i740
13.6 fps G200 

8.8 fps 

24.1 fps 

- 

fast, then failed 

Direct3D Forsaken
640 X 480

 14-17 fps (crappy picture) 

 >42 fps 

>60 fps 

85 fps 

120 fps 

Direct3D Forsaken
800 X 600

- 

 >42 fps 

>56 fps 

85 fps 

120 fps
(jerky pauses) 

Direct3D Forsaken
1024 X 768

- 

>28 fps 

 >50 fps
lock-up 

 42 fps 

81-85 fps
(jerky) 

Direct3D Forsaken
1152 X 864

- 

>17 fps 

 >38 fps
lock-up 

 42 fps 

37-75 fps
(jerky) 

Direct3D Forsaken
1280 X 1024

- 

>14 fps 

>30 fps
lock-up 

 28 fps 

42 fps
(lock-up) 

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 1

Heidi software Z-buffer
>10 fps

 >17.3 fps 

 >10 fps 

MCD (crap)
 5.42 

>16 fps 

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 1

Heidi software Z-buffer
>8.9 fps

 >16.7 fps 

 >10.3 fps 

MCD (crap)
5.92 

>16 fps 

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 1

Heidi software Z-buffer
>7.53 fps

 >15.2 fps 

 - 

MCD (crap)
6.7 

>15.8 fps 

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 2

Heidi software Z-buffer
3.25 fps

 4.34 fps 

  >2 fps 

MCD (crap)
1.86 

>3.8 fps 

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 2

Heidi software Z-buffer
3.8 fps

 4.31 fps 

2.26 fps 

MCD (crap)
1.88 

>3.8 fps 

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 2

Heidi software Z-buffer
3.04 fps

 4.2 fps 

 - 

MCD (crap)
1.97 

>3.8 fps 

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 3

Heidi software Z-buffer
>9.7 fps

 >15.5 fps 

 >5.5 fps 

MCD (crap)
>4.1 

>16 fps 

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 3

Heidi software Z-buffer
>8.2 fps

>14.7 fps 

>6 fps 

MCD (crap)
>4.7 

 >15 fps 

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 3

Heidi software Z-buffer
6.9 fps

 >13 fps 

 - 

MCD (crap)
>5.3 

>14 fps 

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 4

Heidi software Z-buffer
>90 fps!!!

10 fps 

 >4.72 fps 

MCD (crap)
3.7 

>124 fps!!!
lock-up at end 

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 4

Heidi software Z-buffer
>78 fps!!!

 9.74 fps

4.76 fps 

MCD (crap)
3.9 

>115 fps!!!
 lock-up at end 

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 4

Heidi software Z-buffer
>65 fps!!!

 >9.27 fps 

 - 

MCD (crap)
4.12 

>103 fps!!!
 lock-up at end 

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 5

Heidi software Z-buffer
>7.2 fps

 >16.2 fps 

 10.33 fps 

MCD (crap)
1.46 

 >15 fps 

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 5

Heidi software Z-buffer
>6.2 fps

 > 15.3 fps 

10.25 fps 

MCD (crap)
1.85 

>15 fps 

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 5

Heidi software Z-buffer
>5.14 fps

 >13.37 fps 

 - 

MCD (crap)
2.43 

>15 fps 

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 6

Heidi software Z-buffer
>14.27 fps

>22.7 fps 

>18.6 fps 

MCD (crap)
3.36 

 >23 fps
 lock-up at end 

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 6

Heidi software Z-buffer
12.8 fps

 >21.3 fps 

 >18.9 fps 

MCD (crap)
3.75 

 >21 fps
 lock-up at end 

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 6

Heidi software Z-buffer
>11.3 fps

 >18.95 fps 

- 

MCD (crap)
4.37 

>20 fps
  lock-up at end 

 

Benchmark Test 4 has me totally freaked out. The software Z-buffer rocks in this test. The TNT drivers seem to have worked out this kink. As for the lock-ups at the end of the benchmarks, the Hercules driver seems to have resolved this problem.

Upcoming chipsets I hope to test include the Permedia 3, TNT2, ATI Rage 128GL, and the Metabyte TNT Dual SLI configuration.

Additional Tests

These tests were done on the same system, but I've switched to a recent issue Celeron 333 running at 500 MHz.
 

 

16mb RAGE 128 GL AGP

    

Quake II 
640 X 480

31.3 fps 

    

Quake II
800 X 600

29.4 fps 

    

Quake II
1024 X 768

20.4 fps 

    

Quake II
1152 X 864

18.1 fps

    

Direct3D Forsaken
640 X 480

 >120 fps 

    

Direct3D Forsaken
800 X 600

 >63 fps 

    

Direct3D Forsaken
1024 X 768

>58 fps 

    

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 1

 >11.6 fps 

    

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 1

 >11.5 fps 

    

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 1

 >11.3 fps 

    

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 2

 2.5 fps 

    

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 2

2.5 fps 

    

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 2

 2.5 fps 

    

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 3

 >13 fps 

    

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 3

>12 fps 

    

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 3

 >12 fps 

    

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 4

>16 fps 

    

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 4

>16 fps

    

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 4

 >16 fps 

    

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 5

 >11 fps 

    

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 5

 >10.7  fps 

    

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 5

 >10.7 fps 

    

3D Studio Max 2.5
Windows NT 4
1024 X 768
test 6

>21.3 fps 

    

3D Studio Max 2.5
Windows NT 4
1152 X 864
test 6

 >21 fps 

    

3D Studio Max 2.5
Windows NT 4
1280 X 1024
test 6

 >20.6 fps 

    

Back to Graphics Cards Review

1