DriverHeaven ATI 9800 Pro 256 mb review

ATI 9800 Pro 256 mb review

The card is slightly larger than the 128mb version in order to incorporate the extra ram chips on either side. Other than that it has the same look as the other 9800 series cards from ATI, red PCB, silver ramsinks and a large silver heatsink on the core. The feature set of the card is the same as the other models however the speed of the 256mb version is slightly different. The 256mb version uses DDR-2 memory clocked at 350mhz where as the 128mb versions memory is clocked at 340mhz using DDR-1.

click for high res image

Test Setup:

Hardware:
ATI Radeon 9800 pro 256mb
Hercules 3D Prophet Radeon 9800 pro 128mb
AMD Athlon XP 2600+ (333fsb) at 2088mhz
Leadtek K7NCR18DM N-force 2 Motherboard
2x256mb DDR 333 (Twinmos)
IBM Deskstar 120gxp 40gb 7200rpm ATA100 hard drive.
TDK CDRW161040X
Hitachi DVD-ROM GD-7000
Alps Floppy Drive
Q-tec Dual Fan Gold 550w PSU
Compaq V710 Monitor

Software:
Windows XP Pro SP1
Direct X 9.0a
Catalyst 3.7 (beta)
Nforce drivers v2.03
3d Mark 2001SE
3d Mark 2003 build 330
GL Excess 1.2b
Codecreatures 1.0
Unreal Tournament 2003 build 2225

The test system was built from scratch for each card. A format of the hard drive was performed (NTFS) and then Windows XP pro was installed. Following the completion of the install the N-force drivers were installed. The only updates applied were SP1 and Direct X 9.0a. Following a reboot the Catalyst 3.7 (beta) drivers were installed and all graphics options were set to best image quality. Next the benchmarking tools were installed and finally the hard drive was de-fragmented.

Good Benchmarking Practice:
Each benchmark was performed 3 times and the middle result for each resolution/setting is shown in the tables which follow. After changing any Anti-Aliasing (AA) or Anisotropic Filtering (AF) setting and before a benchmark was run the system was rebooted.

3dmark2001 SE is a very popular benchmark that has been broadly used to evaluate Microsoft's Direct X 8.0 and 8.1 graphics hardware. This test is primarily composed of fixed-function DX7-style rendering, with a selection of DX8-style pixel shaders included on a few specific tests.

256mb 9800p	*Image Settings*	0xAA 0xAF	2xAA 8xAF	4xAA 16xAF
Resolution
1024x768		15235 marks	12225 marks	10996 marks
1280x960		13743 marks	10434 marks	9140 marks
1600x1200		11957 marks	8381 marks	6847 marks

128mb 9800p	*Image Settings*	0xAA 0xAF	2xAA 8xAF	4xAA 16xAF
Resolution
1024x768		15080 marks	12253 marks	10811 marks
1280x960		13660 marks	10400 marks	8940 marks
1600x1200		11872 marks	8228 marks	6706 marks

A fine set of results from the 256mb card. Comparing the above results against the 128mb version we see that the 256mb card is ahead in but one of the tests though the difference overall is small (between 1% and 2%).

click image to visit Futuremark

3dmark03 is the newest benchmark from Futuremark, and the only benchmark in common use that is designed specifically for Direct X 9.0 graphics hardware.

This test is especially challenging in that it exerts a heavy pixel shader and vertex shader load on the graphics adapter. Today's popular game titles still tend to use a fixed-function DX7 style of rendering, with a few optional shaders thrown in for special effects. While the current crop of game-based benchmarks can be made to create a heavy fillrate load (by increasing resolution and adding AA and/or AF), they do not stress the portions of the graphics processor that will become critical in next-generation titles - the shader engines.

3dmark03 contains multiple scenes that are completely shader-driven and as such is highly valuable as a forward-looking benchmark test.

ATI have stated that the RADEON 9800 PRO offers full-time precision Pixel Shaders. Nvidia may be able to produce shader performance that is close to the RADEON 9800 PRO by falling back to a half precision (16bit) floating point mode. 16bit precision in the Pixel Shader calculations may improve performance in some cases, but at the cost of image quality.

256mb 9800p	*Image Settings*	0xAA 0xAF	2xAA 8xAF	4xAA 16xAF	6xAA 16xAF
Resolution
1024x768		5497 marks	3760 marks	2787 marks
1280x960		4318 marks	2861 marks	2120 marks
1600x1200		3246 marks	2125 marks	1567 marks	1136 marks

128mb 9800p	*Image Settings*	0xAA 0xAF	2xAA 8xAF	4xAA 16xAF	6xAA 16xAF
Resolution
1024x768		5419 marks	3684 marks	2712 marks
1280x960		4291 marks	2860 marks	2053 marks
1600x1200		3219 marks	2053 marks	1492 marks	998 marks

At lower resolutions and settings the two cards are very similar in their performance. When moving up through the resolutions we can see the effect of the extra 128mb on the ATI card taking effect. For example at 1024x768 0xAA 0xAF the difference is 1.4%, at 1600x1200 6xAA and 16xAF the difference in performance increases to 12.1%

Next: Benchmarking in Codecreatures & GL Excess