Unfortunately, the Radeon 9600 XT is not faster than the Radeon 9700 Pro. Although the Radeon 9600 XT has a higher clock speed, the Radeon 9700 Pro has more pixel pipelines, 8 vs 4, and it also has a higher memory interface, 256bit vs 128bit.
do manufacturers like sapphire and such have to stick to the same memory interface? is that a part of the gpu or can they mix and match, use a 256bit interface?
The memory interface is part of the GPU; the bandwidth can be decreased but not increased... i.e. 256bit can be halved to 128, but a GPU with 128 cannot be upgraded to 256 unless it was already a disabled 256 card, like the old 9500NPs.
Since newer video boards often follow designs similar to PC architecture, I'll make the following guess:
The GPU must be designed to interface with a 256bit bus. This has been proven in the hardware mod that first turn 9500 cards into 9700 Pros. We had to change transistors on top of the chip (Before the quick advent of RivaTuner + other methods) to enable the 256bit bus on the hardware level. The board must also have the appropriate number of traces between the GPU and the memory to support a 256bit bus.
If these two conditions are met, maybe someone could make a 9600XT out of the 9800XT PCB by disabling the 256bit bus (A la 9500 from a cripled 9700 Pro PCB), but I think ATI may have learned its lessons, and the cost between manufacturing two PCBs or cripling one to support two cards needs to be balanced.
//EDIT: Damn, ghoodsum beat me while I was typing.
Yea, it is called the 9800XT.
It puts up some serious numbers, but the 95 and 97 pro's are going to be plenty of card for some time to come (a couple of months anyway).
Comments
do manufacturers like sapphire and such have to stick to the same memory interface? is that a part of the gpu or can they mix and match, use a 256bit interface?
The GPU must be designed to interface with a 256bit bus. This has been proven in the hardware mod that first turn 9500 cards into 9700 Pros. We had to change transistors on top of the chip (Before the quick advent of RivaTuner + other methods) to enable the 256bit bus on the hardware level. The board must also have the appropriate number of traces between the GPU and the memory to support a 256bit bus.
If these two conditions are met, maybe someone could make a 9600XT out of the 9800XT PCB by disabling the 256bit bus (A la 9500 from a cripled 9700 Pro PCB), but I think ATI may have learned its lessons, and the cost between manufacturing two PCBs or cripling one to support two cards needs to be balanced.
//EDIT: Damn, ghoodsum beat me while I was typing.
Radeon 9700 Pro:
FILL RATE: 325 MHz x 8 Pipelines x 1 TMU per Pipeline = 2.6 Gigapixels/sec
MEM BAND: (256-bit x 310 MHz x 2 DDR) / 8 = 19.8 GB/sec
Radeon 9600XT:
FILL RATE: 500 MHz x 4 Pipelines x 1 TMU per Pipeline = 2.0 Gigapixels/sec
MEM BAND: (128-bit x 300 MHz x 2 DDR) / 8 = 9.6 GB/sec
Here you can definately see that the 9700 Pro outperforms the 9600XT.
Now 500 MHz core on a 9600XT with a 256-bit memory-controller... there's something to think about
It puts up some serious numbers, but the 95 and 97 pro's are going to be plenty of card for some time to come (a couple of months anyway).