I'll be honest, I am not a big advocate of Crossfire. Having battled for the last year with the drivers on a X1900XT Crossfire based system the thought of having to deal with this review was akin to having my teeth pulled out by a dentist who found out I was sleeping with his wife.

My colleague Stuart "Veridian3" Davidson has been singing the praises of Crossfire for the last year and he seems very happy with the overall package. Fortunately for ATI I didn't handle the launch review as I still think their claims have not rang true on many levels. When I was at the original Crossfire press launch I was told that the ATI solution would be superior to SLI as they don't rely on a driver profile to support dual card rendering for your favourite game. One year on, I think most of us know that was perhaps a slight exaggeration on ATI's behalf. Anyone who had a copy of Ghost Recon Advanced Warfighter on launch day and renamed the EXE to an AFR supported title such as "3DMark03.exe" instantly received twice the frame rates. Obviously not quite just a convenient coincidence. We have given up trying to convince ATI to let the end user toggle dual card rendering modes via the driver control panel, much to our disappointment.

The purpose of this review is to compare an X1900 Crossfire setup with the replacement configuration, the not so creatively named X1950 Crossfire. I am not going to bore everyone with detailed specifications of both sets of cards, they have been out for quite some time and you can always peruse our prior reviews to catch up with the technology if you are fresh to the hardware. ( X1950XTX ) ( X1900 and X1900 CF )

All the hardware today has been generously supplied by our good friends at Connect 3d and Sapphire, they sent us a range of X1900XTX and X1950XTX cards to produce this article.

ATI's CrossFire platform allows the end user to mix graphics cards clocked differently and with various memory configurations. This means we are able to use Radeon X1900XTX, X1900XT or even lower clocked Radeon series graphics cards with the master board. It isn't widely known but you can mix and match cards from different ranges as well, as I did in an experiment before writing this article, using a X1900XTX master and X1950XTX slave. When swapping out the master card for the newer X1950 series however I ended up with a black screen when booting in windows, requiring me to reboot into VGA safe mode and deinstalling the driver, so it would be safe to assume that there may very well be a few issues if you go down this road.

Earlier I detailed links to prior reviews with background information on the technology, however im aware some of you just want a basic overview of the technology.

	ATi Radeon X1950XTX	ATI Radeon X1900XTX
GPU Codename	R580	R580
Manufacturing Technology	TSMC 0.09 micron Low-k	TSMC 0.09 micron Low-k
Number Of Transistors	384 million	384 million
Clock Frequency	650mhz	650mhz
Graphics memory controller	256bit GDDR3 SDRAM	256bit GDDR3 SDRAM
Graphics memory clock frequency	2000mhz	1550mhz
Memory bus peak bandwidth	64 gb/s	49.6 gb/s
Graphics memory size	512mb	512mb
Interface	PCI Express x16	PCI Express x16
Pixel Processors, pixel shaders
Shader Model	3	3
Number of pixel shader units	48	48
Static loops and branching	yes	yes
Dynamic loops and branching	yes	yes
Multiple Render Targets	yes	yes
Floating Point Render Target	yes	yes
Number of texture mapping units (TMU)	16	16
Number of Z compare units	16	16
Texture filtering algorithms	bi-linear, tri-linear, anisotropic, tri-linear and anisotropic, area-anisotropic	bi-linear, tri-linear, anisotropic, tri-linear and anisotropic, area-anisotropic
Maximum level of anisotropy	16x	16x
Vertex processors, vertex shaders
Shader Model	3	3
Number of vertex processors	8	8
Static Loops and branching	yes	yes
Dynamic loops and branching	yes	yes
Reading textures from the vertex shader	no	no
Tesselation	no	no
Full screen anti aliasing
FSAA algorithm	Rotated-grid, multisampling, temporal anti aliasing, adaptive antialiasing	Rotated-grid, multisampling, temporal anti aliasing, adaptive antialiasing
Number of samples	2,4,6	2,4,6

As you can see the new model doesn't differ from the older one in anything except the memory bandwidth, however the difference of 14.4gb/s is substantial and should not be ignored. It is pointless to test this hardware at a low resolution as the memory bandwidth will not be stressed, obviously it is also clear that high end expensive dual card solutions are purchased to run at the highest resolutions. Therefore we will be limiting resolutions to 1600x1200 and 1920x1200 with eye candy maxed out. Also later in the article I will be testing the cards with 8aa and 14aa.

OCZ sent us a 2 GB kit of PC2 6400 (15-3-4-CL4) Special Ops Urban Edition memory and we are using it in this test system, a full review will be forthcoming at a later date. It may be difficult to see in the picture below, but the memory sticks are finished with a rather unusual camouflage style heatspreader.

The X1900XT/X card was criticized when it was released for the noise levels from the undersized fan, to test, I measured dBA with a Velleman DVM1326 digital sound level meter from a range of 7cm's next to the cooler. This obviously is a (extremely) worst case scenario as you would have the cards inside a closed case and would be probably 1-2 meters away from the chassis. Ambient noise was around 34dBA.

	2d	3d
X1900XTX	50	59
X1950XTX	43	50

Next we measured noise levels with the side door open from a range of about half a meter.

	2d	3d
X1900XTX	45	53
X1950XTX	40	48

There is a vast difference in noise levels between the cards and we would recommend that anyone using a X1900XT/X move to a third party cooling unit to lower these levels. Under full load the noise from the X1900XT/X is extremely irritating, whereas the X1950XTX is virtually inaudible under all but the heaviest load. For those interested the power consumption of both cards under full load is as follows. X1900XTX 120watt and X1950XTX 125 watt.