Conclusion

Just by looking at the architecture of X1900 it is obvious where its major benefits were going to lie in relation to ATI's previous generations of parts. In some ways this harks back to Radeon 9700, which had a fairly revolutionary shader architecture for its time, but which ATI had leant on for over 3 years without significantly altering the dynamics until RV530 and R580. Looking at current applications, though, we can generally see performance increases over X1800 between 15%, even for the less shader bound titles, up to the mid 30% range for the more shader bound cases. ATI points out that with a mere 20% increase in transistors over X1800 they can triple one aspect of performance, and with titles that are available currently that die size increase corresponds exactly in performance. Adding in the high resolution results we see that the performance difference between the X1900 XTX and X1800 XT grows to a fairly large degree, from the 30% range for the less shader bound Quake 4 and towards 50% for F.E.A.R., highlighting the difference the shader performance is making in these cases.

As the pure shader tests demonstrate, it's where there is high pixel shader utilisation that X1900 will begin to see the major benefits, and that is realised to a limited extent with the more shader bound Splinter Cell: Chaos Theory and F.E.A.R. applications. To realise somewhat closer to the true benefits of X1900's architecture applications need to rely less on texturing and more on math and branching capabilities. To ensure this happens, ATI have to evangelise the use of Fetch4 to developers to maximise texture performance with shadow maps (something that many are already doing for NVIDIA's boards) and heavily promote pixel shader utilisation and branching (although many developers are expressing an interest in this, but it has to happen soon for it to become of meaningful use within X1900's lifetime) - in all, there appears to be a heavy onus on ATI's developer relations here. There is an argument to suggest that the more traditional elements of the pipeline hardly need speeding up much more, as we can already achieve in excess of 100 FPS in titles such as Doom 3, and it's shaders that are going to be one of the major performance drains in upcoming titles.

There are many aspects of the R5xx architecture that have moved things on effectively without needing to push developers down particular paths too strenuously. There are a series of improvements to the architecture, such as the new dispatch engine, and the memory bus and efficiency enhancements, that are paying off immediate - the efficiency of the memory bus changes to the ROPs are certainly still paying off where FSAA rendering is concerned. These differences are highlighted when comparing X1900 to the last generation refresh with the X1900's performances often two times greater, or more, then the X850 XT PE's.

These performances can be further enhanced with the X1900 Crossfire system. While the Crossfire system in this review was probably constrained by the CPU and resolutions we were able to run, it's clear that the platform can scale well, and for those situations that it can't, still offer much higher quality levels, without too much of an overhead. Systems such as these come at a price, but for those that can afford it will really come into its own with newer, very high resolution panels that are becoming available. When we add in the results with such a panel it the performance of a Crossfire really becomes more apparent, with resolutions such as 2560x1600 with 4x FSAA and 8x AF enabled and performances close to 100 FPS in Quake 4 being reached - these configurations make such high resolutions achievable and attainable with high image quality options enabled as well.

Pricing is of course the issue for most with boards such as these. With the increase in die size required for R580 the price of the board has increased correspondingly such that ATI are now topping $649 recommended prices (which don't necessarily bear any resemblance to retail prices initially). The curiosity here is the massive price disparity between the X1900 XTX and XT, standing at a $100, and yet the performances are only about 4% apart - clearly the XT is a better buy, especially if you are considering a Crossfire platform and the performance increase will scale more inline with the XT rather than the XTX. Another curiosity is the inclusion of 1.1ns RAM on the XTX board when the part isn't using 1.1ns memory speeds - given they are within 1.2ns memory speeds, and hence aren't charging for the benefits of 1.1ns memory, its crossed our mind a few times whether ATI may be skimming off higher performance boards for a later date.

ATI's decisions for the R580 chip that powers the X1900 have been interesting, to say the least, even a little bold in some ways. Even though we've looked at the X1900 fairly thoroughly in this article we've possibly yet to see it pay off fully. As shown by the pure pixel shader tests, but not fully realised by the games, this really is an architecture that has more to come, but only if games are going to going to go in the way that ATI are predicting, and quickly. For now, though, its clear that Radeon X1900 XTX offers a performance increase over X1800 XT in all situations, but it also has the price premium to match.