RADEON 9800XT vs GeForce FX 5950 - Head to Head #36

After a brief flirtation with synchronous product launch and availability, the graphics industry has tumbled back into the realm of paper launches. It took over a month from ATI's announcement of the RADEON 9800XT for cards to enter Australia, and it has been even longer for the mid-range 9600XT. NVIDIA did things a little quicker, managing a two week turnaround on the GeForce FX 5950 and only a little longer for the 5700 Ultra.

Now that the delays have ended, we can finally line these two new flagship cards up against each other to see how performance differs in this final refresh of the NV3x and R3xx series of video cards. It now looks like ATI will announce its R420 and R423 products and NVIDIA its NV40 sometime during the first quarter of 2004.

It will be with the introduction of these products that the market will undergo another shift. This generation has not been the best for NVIDIA, who gambled on a custom hardware implementation of DirectX 9.0 and got shafted by ATI's pure DirectX 9.0 shader design. However, these companies are not run by idiots and with DirectX 9.0 relatively stable it would be naive to expect to see NVIDIA make the same mistakes again.

At the moment ATI has the better shading unit in the high end. NVIDIA's require developers to use more custom code and mixed shader versions to get it flying at full speed, and when shown the much easier option of pure DirectX 9.0 a la the RADEON they will naturally prefer it.

But with the next generation it is a reasonably safe bet that shader performance will level out between companies and the market will become a lot more interesting. NVIDIA has clawed back a chunk of the raw performance ground lost with the GeForce FX 5800 Ultra, and once the major architectural changes of a new generation come online they should be able to match ATI's performance (ATI has publicly talked about a target of double the performance of the RADEON 9800XT for the R420) in a much more equitable way than now.

To gauge the performance of the RADEON 9800XT and GeForce FX 5950 Ultra we have tested them extensively with our new benchmarks. In the left corner we have Powercolor's RADEON 9800XT, already standing a little taller thanks to the inclusion of Half-Life 2 (when it is ever actually released, probably March or April next year) via a redemption voucher. In the right we have Gigabyte, newly returned to the NVIDIA fold with its GeForce FX 5950 Ultra card.

Both cards have been tested using our Athlon 64 3200+ testbench, with an ASUS K8V motherboard and 512MB RAM. The tests were run using Windows XP with SP1 and DirectX 9.0b. The RADEON 9800XT was tested with CATALYST 3.9 drivers while the GeForce FX used NVIDIA's ForceWare 52.16 drivers.

John Gillooly


Aestheticisms
The 9800XT is one of the weightier cards ATI have released - this is due to the swanky new all-copper heatsink, which is made up of several plates of square-folded copper. On the other side of the fence, the FX5950 Ultra isn't quite as heavy as it has inherited the same aluminium heatsink design from its predecessors - the cooling system we like to call the lawn blower design. This means the same steal-an-expansion-slot problem still remains, killing off any hope of fitting it in the majority of barebones machines without some kind of major surgery. Though admittedly that'd look rather swanky.

The 9800XT was damn quiet but the FX5950 Ultra was the noisiest with its slowest fan speed matching close to that of the highest fan speed on the gloriously peaceful 9800XT. The 9800XT was so quiet it's hard not to be fooled into thinking it uses passive cooling - it barely ever went above its slowest speed even when under full load. With a slower, yet larger fan, the temperature of the 9800XT was cooler than the FX5950 Ultra, with the FX5950 Ultra at times edging on uncomfortable to the touch. Because of the advantages of reduction in both the heat and noise, it would seem this copper design should be used more often on video cards.


The gruntmasters
In the past, ATI's new beasts have completely stripped the competition nude. NVIDIA tried to evade this by creating a card that literally blew our clothes off, but that wasn't impressive enough. Nor was the noise. Over the months, NVIDIA has slowly been creeping closer to the insane performance of the RADEON 9800 PROs, but never enough to fully squish and defeat them. In the last couple of months, they managed to steal a place or two at the top of Framerate but only by small skips and crawls. So the ultimate question is has NVIDIA finally done it? Which new beast is superior - the GeForce FX5950 Ultra or RADEON 9800XT?

Peering at the results, you can see pretty much wherever non-prettied-up raw frames are concerned, the FX5950 Ultra really does a fine job shooting ahead, albeit by a small margin. It stole sides from both theoretical and gaming tests. We're impressed with the results, as this shows NVIDIA have finally got off their arses and done something about the huge lack in performance. There is still one test they can't yet quite muster, and that's with 3DMark2001SE. There is a hint of performance though, which shouldn't be boogered at - it appears 4x antialiasing and 4x anisotropic filtering is its sweet spot. This is one of the first instances we've seen a GeForce card beat a RADEON by so much using a decent amount of filtering.

You may notice in 3DMark2001 the FX5900 Ultra is missing the 8x test results. Even though it should have worked fine (8x antialiasing and anisotropic filtering are both supported on this card for both primary graphics APIs - Open GL and Direct3D), for an unknown reason the card promptly ran with antialiasing turned off at this setting, regardless of how much monkey dancing and screaming was done (you really don't want to see us in action). Not wanting to skew the existing results, we kept it out of the charts.

With all its controversy, 3DMark2003 Pro is making an appearance, as we believe it is getting to a more mature state. At the time of writing, release 340 had just been pushed to the public and this one keeps NVIDIA's 52.16 Forceware (the drivers are no longer known as Detonators) at bay from any cheats (particularly shading) they may attempt to use. Which, might we mention again, is really frucking stupid of NVIDIA. The FX5950 Ultra card still pushed ahead in all tests here, so why NVIDIA stubbornly continues to cheat with its drivers is beyond us.

NVIDIA's line also pushed ahead in the demanding Aquamark3 and Codecreatures as they did also in Unreal Tournament 2003 - showing a slight edge we haven't seen in quite a while.

We've also introduced Call of Duty as a benchmark, as it's a game based on a modified version of the Quake 3 engine. And as you probably know, Q3A has damn good inbuilt benchmarking abilities, but it barely manages to stress the video anymore. It has aged and is now onl

• Purchase your copy now - only $8.95 including postage
• Subscribe to our digital version - only $49.95 a year