1. Foreword. The First Impression.
2. Cooling System. Basic Components.
3. Setting Up and Overclocking.
4. Synthetical Tests (3DMark 2001SE and 2005).
5. Direct3D Games (Far Cry, Serious Sam 2).
6. OpenGL Games (Quake 4, Prey).
7. Conclusions.

There are 3DMark 2001SE and 3DMark 2005 by MadOnion / Futuremark to serve the purpose of synthetical benchmarking. 3DMark 2001SE build 330 is almost 5 years old but still has some interesting tests to offer. First of all, those synthetical completely Fill Rate and High Polygon Count, though Nature is a good one as well. 3DMark 2005 is able to use pixel and vertex shaders v3.0, and they happen to be a must have for modern 3D gaming titles. One must note that shaders of 3DMark 2005 are written in HLSL and compiled at run-time rather than written in assembly language and pre-compiled, so they may lack optimisation. Unfortunately, many 3D software developers follow this popular trend. Both 3DMark 2001SE and 3DMark 2005 operate through Direct3D.
 
This test illustrates how fast a scene can be filled, i. e. texturing and rasterising speed. It constains almost no geometry data, so there is a very little to zero dependence on processor speed or video memory size.
 
Sapphire Radeon X1950 Pro has outperformed BFG GeForce 7800GS OC, though with the help of overclocking the latter has almost reached the leader in single-texturing and even outrun it in multitexturing. Sapphire Radeon X1600 Pro and Sapphire Radeon 9600XT are much slower and their results are close to each other. That's quite reasonable considering their architecture and clock speeds.
 
This test renders a scene with several objects moving dynamically and either 1 or 8 active lights. There is almost no texturing (background only), so it doesn't depend much on video memory bandwidth. It's a matter of vertex pipelines mostly to deliver high results in this test.
 
Sapphire Radeon X1950 Pro (8 vertex pipelines) is an undisputable leader, though BFG GeForce 7800GS OC (6 vertex pipelines) delivers not much worse performance especially if overclocked. Sapphire Radeon X1600 Pro (5 vertex pipelines) looks good, and Sapphire Radeon 9600XT (2 vertex pipelines) has nothing outstanding to show.
 
Well, Nature is considered fairly as the most beautiful game test of the 3DMark 2001SE suite. It employs vertex shaders v1.0 to produce a wind effect on grass and trees' leaves, also to calculate movements of the fisherman and butterflies. Pixel shaders v1.0 together with multitexturing are involved in generation of the water surface.
 
Sapphire Radeon X1950 Pro leads the run keeping BFG GeForce 7800GS OC not so far behind. It's clear to see that Sapphire Radeon X1600 Pro is over than 3 times slower than the leader, and Sapphire Radeon 9600XT isn't any better.
 
The other game tests of the 3DMark 2001SE suite (Car Chase, Dragothic and Lobby) don't use pixel shaders at all, so they're of a little value nowadays from the point of game design. However, they depend well on texturing and rasterising speed. Yes, somewhere back in the past it has been of key importance for 3D graphics hardware. Anyway, refer to the following tables.
 

3DMark 2001SE (1280x1024x32bpp), Low detail, average FPS
	Car Chase			Dragothic			Lobby
	BF NoAA	16xAF NoAA	16xAF 4xMSAA	BF NoAA	16xAF NoAA	16xAF 4xMSAA	BF NoAA	16xAF NoAA	16xAF 4xMSAA
Sapphire X1950 Pro	279	275	263	461	430	366	284	276	258
BFG 7800GS OC o/c	334	325	277	482	437	292	311	308	268
BFG 7800GS OC	325	307	246	425	380	256	309	303	246
Sapphire X1600 Pro o/c	262	167	119	235	196	134	228	144	104
Sapphire X1600 Pro	241	147	106	214	175	123	208	127	94
Sapphire 9600XT o/c	180	133	87	191	167	106	190	127	86
Sapphire 9600XT	159	114	73	160	140	85	156	107	71

The tests in low detail mode have revealed that both Sapphire Radeon X1950 Pro and BFG GeForce 7800GS OC deliver comparable results, but with the help of overclocking the second card has become the winner with the only exception of Dragothic rendered through 16x anisotropic filtering and 4x multisampled anti-aliasing. Sapphire Radeon X1600 Pro appears to be about 20% faster than Sapphire Radeon 9600XT.
 

3DMark 2001SE (1280x1024x32bpp), High detail, average FPS
	Car Chase			Dragothic			Lobby
	BF NoAA	16xAF NoAA	16xAF 4xMSAA	BF NoAA	16xAF NoAA	16xAF 4xMSAA	BF NoAA	16xAF NoAA	16xAF 4xMSAA
Sapphire X1950 Pro	95	94	93	224	222	210	130	129	125
BFG 7800GS OC o/c	109	109	109	244	230	190	145	145	140
BFG 7800GS OC	108	108	107	221	203	166	145	144	135
Sapphire X1600 Pro o/c	105	89	69	158	139	96	124	89	71
Sapphire X1600 Pro	104	81	61	144	117	89	117	80	64
Sapphire 9600XT o/c	91	71	50	121	106	71	108	79	57
Sapphire 9600XT	82	62	42	106	93	59	94	69	48

Although the tests in high detail mode are known for their processor dependent reputation, it's interesting still to take a look at their results. Sapphire X1950 Pro has delivered unexpectedly low performance in Car Chase, even slower than Sapphire Radeon X1600 Pro in case of bilinear filtering and no anti-aliasing. In general, BFG GeForce 7800GS OC has won this round. It seems that developers of ATI Catalyst have got many things to improve.
 
Having first conclusions made, let's continue with 3DMark 2005.
 
Return to Proxycon features advanced geometry (the hangar with many high-detailed objects including those moving dynamically) and advanced lighting (up to 8 sources). Pixel shaders are also used widely. The view is limited by walls of the hangar and corridor which is typical for most 3D action games.
 
Sapphire Radeon X1950 Pro has shown finally what it can do by outperforming BFG GeForce 7800GS OC for 52% to 88% in average FPS. Even the latter card with overclocking is far behind. Sapphire Radeon X1600 Pro looks not bad, but it has a long way to reach even BFG GeForce 7800GS OC.
 
Firefly Forest depends on vertex performance mostly because of trees and grass: branches of the first move with the wind, and the second just needs to be generated dynamically. In a matter of fact, there are only 2 lighting sources: the moon which is a statical one and the flying light which is a dynamical one.
 
Sapphire Radeon X1950 Pro is on rampage in this test. As you can see, it delivers better minimal FPS than average FPS by the primary competitor even if overclocked. Sapphire Radeon X1600 Pro looks not bad still, but nothing more.
 
Canyon Flight employs advanced very much pixel shaders to generate walls of the canyon and the water surface. There are different fog effects to take place as well. Geometry of objects is moderate, the sun serves as the only lighting source.
 
Once again, Sapphire Radeon X1950 Pro is far ahead outperforming BFG GeForce 7800GS OC for about 70% in average FPS. Once again, Sapphire Radeon X1600 Pro has nothing to boast with.
 
Overall, Sapphire Radeon X1950 Pro is the only and undisputable leader in 3DMark 2005. BFG GeForce 7800GS OC has possessed the second place, but its results are closer to Sapphire Radeon X1600 Pro which holds the third place than to the winner. As it has been expected, Sapphire Radeon 9600XT has rendered many good slide-shows. In general, 3DMark 2005 depends very much on pixel shader performance, therefore it shouldn't be a miracle that Sapphire Radeon X1950 Pro has been able to prevail so hard. Considering 1024x768 with trilinear filtering and no anti-aliasing enabled, it has scored about 9500 marks. BFG GeForce 7800GS OC has managed to get about 5800 marks before the overclocking and about 6900 after. Sapphire Radeon X1600 Pro has been tough enough to obtain about 4100 marks before the overclocking and about 4500 after, Sapphire Radeon 9600XT — about 2000 and 2400 respectively. With bilinear filtering which is set by default ("Optimal") the scores would be approx. 2% higher. Just in case, you can calculate scores for any particular mode using average FPS from the diagrams above and the following formula from the 3DMark 2005 help: (test1 x test2 x test3)^1/3 x 250