Conclusion

Its fairly clear that when ATI introduced their PCI Express Radeon lines they didn't consider at that point on using them for multiple graphics board rendering and as such they have ended up with a few compromises, however they have tempered that by making the system somewhat flexible and using their capabilities that they implemented way back in R300 for high en visualisation use.

Obviously there was a desire to support their current X800 and X850 users such that their current graphics board could still be utilised in a dual board situation, however the only way they could be supported with identical boards would be to always have the transferred over the PCI Express bus in all situations, which would limit the performance - so instead they have come up with the master and slave concept that requires a new board with the capabilities for inputting the signal from the already existing board, which itself requires some more untidy leads dangling out the back of your PC case. Although this allows for some flexibility in terms of the boards supported, in many instances there will be wastage in terms of the processing capabilities due to the different board performances and no method of compensation. ATI have divided the support between the boards up in the manner they have in order to make it simple for the end user to understand, however in the case of the higher end X800 XT and and X800 XT Platinum Editions, particularly, they may benefit more from coupling with the X850 CrossFire Edition as they would the X800 CrossFire Edition as their specifications are closer - we'll have to see whether this is just a suggested mode of operation or a hard locked restriction in the drivers.

It also appears to be the case that each individual graphics chip may not be able to composite the images together of their own accord, which is why ATI have included a separate composting chip on the board. Although ATI's hardware since R300 has had specific capabilities for supporting multi-chip rendering it is quite normal for high end visualisation systems to have separate composing and display capabilities so it would not be an expected function of the graphics chip itself - although they would be able to support AFR without the engine, and possibly the scissor mode, the more advanced Supertiling and Super AA may not have been capable without the compositing device. Although it may be the case that ATI just didn't consider these applications for the X800 series when they set their design in stone, which would have been something in the order of at least two years ago now, its not necessarily a given that they will be supported natively in future chips as, for every chip sold, there will need to be an area of the die dedicated to this processing which is an additional cost with only a very small percentage of users actually using it.

Another potential issue is the need for carrying two ranges of graphics boards, especially since the CrossFire versions are going to be very low volume SKU's. We'll have to wait and see exactly how availability of these come about, especially in regions outside of the US which are solely reliant on partner boards and don't have ATI branded boards available. When we consider that ATI's next generation graphics chip, R520, has been seen as operable it would also suggest that these is gearing up for a launch within the next few months which would also put a question mark over the need for these boards - how many are likely to purchase a second $549 Shader Model 2 board now, which probably won't yield a 2x performance increase in many cases, when R520 looks to be available soon, must be on ATI and their vendors minds.

We also hope that ATI don't restrict these capabilities to their own platforms, or at least if they do they move to address the situation quickly and ensure support on alternative platform (and hopefully NVIDIA will do likewise with ATI's dual graphics chipsets, although there is less of a need for NVIDIA to do this as ATI are the ones with zero multi-board chipset market share right now). ATI pointed out that there are presently millions of graphics board users out there that could purchase a CrossFire Edition graphics board and get benefits from it, however we also pointed out that if it were limited to ATI platforms they would also need to change their platform in order to do it, whereas (although the actual numbers may be slim) there is an infinitely greater chance of there being X800/X850 users on an nForce4 SLI platform that could get away with just buying the second board.

There are some good positives to the solution that ATI have come up with. As we have discussed, at least this solution allows for the use of current X800/X850 boards already in service, and it doesn't have any prickly issues with running different vendor, or even slightly different performance boards. The different types of rendering modes are interesting, and may allow for increased compatibility - the fact that ATI are suggesting that when all the elements of CrossFire are in place the default mode will be to have multi-rendering enabled for all titles, and not just the ones that have been specifically tests or coded for, is good for the end user, although the compatibility and performance increases will obviously need to be tested once hardware is available. The extra FSAA modes available to the dual graphics rendering system are also a big plus for those titles that already see healthy performances even with high levels of FSAA and Anisotropic Filtering on current boards, or where things are very CPU limited - at least end users will be able to make some use of the extra graphics board by getting even better image quality out of them both together.

It's also of note that some of the physical elements to this present implementation are due to the fact that they are having to apply a system to boards that were quite evidently not designed to be used in a multiple graphics rendering environment and all of this is very much subject to change with future releases. Whilst we suspect that ATI R520's chip design would have been set in stone before they could make any serious thought to more support for desktop multi-graphics rendering at the silicon level (that may have to wait for future chips or architectures), they do have the opportunity to make implementation changes at the board level such that with the next generation of boards we may see an internal connector supported, as opposed to an external one.

On the whole, though, we are left scratching our head a little as to why ATI are taking this route right now. Obviously designing and developing the board support and platform support, as well as getting all the software ready, with a good level of compatibility, takes its time, however this would have made far more sense to be releasing in this form 3-6 months ago. Still, we keenly await to be able to test the system ourselves and we will update this article with the full review breakdown when we are able to do so.