Conclusion and thoughts
We've got through a glut of information and a barrage of benchmarks. Let's now recap and reflect on what exactly what the Radeon 9800 Pro is all about. Fundamentally, the card is an update of the popular R300 series. Running with faster VPU and memory speeds of 380MHz core and 680MHz memory respectively, it's innately faster than the 325/620 clocked Radeon 9700 Pro.
There's a little more to it than just a boost in clock speed alone. The card is physically larger with a molex-style connector, presumably to keep it fed with the required juice at all times. Looking at the performance exhibited by its main compeititor, the GeForce FX Ultra, a simple clock boost, a la GeForce3 - GeForce3 Ti 500, would have sufficed for ATi to unequivocally take back the performance crown. They've gone a little further on the hardware front to ensure the card is as future-proof as possible.
I think we're steadily seeing a shift towards an extensive use of vertex and pixel shaders. The ability to create far more realistic, immersive environments has been opened up with the present and forthcoming APIs. With 128-bit pixel precision and complex shaders that specify life-like images, we're looking to close the gap between cinema and gaming. To help us in this respect, the 9800 Pro adds a special F-buffer to allow, theoretically, a pixel shader of unlimited length. That contrasts well with the present 9700 Pro, whose pixel shader length is limited to 64 instructions. Instructions longer than that would have to be multi-passed through the pipeline, thereby causing huge bandwidth problems. This 'feature' will only show its worth when developers start tackling huge shaders. To match the FX Ultra, the 9800 Pro's vertex instructions are up from 1.024 to 65,280. Having waxed on lyrically about the F-Buffer, I'd trade it in for a second if the 9800 Pro omitted it but gave 400/400 clocks. It's a feature that won't be used for some time yet (OpenGL 2.0, as DX9 doesn't go that far)
It does appear that ATi's SMOOTHVISION (aniso and FSAA) has been improved when applied at high settings. Comparing a Radeon 9700 Pro clocked at the same frequencies, and with 4x FSAA and 8x AF thrown in, the 9800 Pro managed to pull out a 10% advantage. The only caveat is the different driver set being run on both cards, although non-SMOOTHVISION performance was identical. Further, the bandwidth-saving Hyper Z III sees a little refinement, especially when dealing with shadows. Two of the three main improvements won't be realised for some time yet, we must remember that.
It's both faster and has more features than the 9700 Pro, and generally matches the GeForce FX on the feature front. That alone will ensure that it sells by the proverbial bucket load. Our benchmarks have shown it to be the performance leader in the majority of benchmarks that we conducted. Excelling the key area of high resolutions and high settings of image quality (4x AA, 8x AF), it betters the 9700 Pro in almost every way. Evolutionary rather than revolutionary. After all, why change a winning formula ?. It should come in at roughly the same price point as the present 9700 Pro. That's good news for the consumer. With the promise of a 256MB version harnessing the power of DDR-II, the future looks rosy for ATi. Think of the ATi Radeon 9800 Pro as an enhanced version of the 9700 Pro. It does everything the latter card does, but with a little more speed and grace.
We'll await with baited breath to see if retail samples feature the same 380/680 clocks as this particular review sample. ATI's literature is a little contradictory at times, sometimes alluding to stock memory speeds of 700MHz, and at other times sticking with 680MHz. I personally challenge any of ATi's partners including, amongst others, Sapphire, Connect3D, Crucial and Powercolor to go a little beyond specification and supply the end-user with a Radeon 9800 Pro running with 400/700 clocks. It would make the fastest available card that little bit sweeter.
Highs
-
The fastest card in town. Around 15 - 25% faster than the 9700 Pro.
-
Doesn't take up a PCI connector
-
New hardware features bring it into line with the FX series
-
Excellent performance under heavy aniso. and FSAA loads
-
Small PCB, quietish cooler
-
The prospect of a faster core and 256MB of DDR-II memory is appealing.
-
Exceptional 2D quality
-
Should be available within a month at ~ Ā£300
Lows
-
Does the performance difference justify the price delta between R300 and R350?
-
Lack of Dual DVI on reference board - might be different between this and the partner boards.
-
Evolutionary rather than revolutionary. We'll have to wait for the R400 for that step forward.
Editors Note
This unit is avaliable at [OCS]