Written By : Jeff Nettleton and Trevor Haag
September 2004

Performance:::...

Test System

DFI Infinity 865PE
Intel Pentium 4 (Prescott) 3.2GHz
1024MB GeIL ULtra-X PC3200 Memory
Connect3D Radeon X800XT-PE
CATALYST 4.7, 4.8 and 4.9 Beta

The following driver settings were used at all times. Tests completed at 4xAA and 4xAF were done so by setting these parameters from within 3DMark05/03.

This was the initial 3DMark score when run at the default settings and CATALYST 4.7s.

And here's the breakdown.

Just as I was ready to wrap up the preview I got an email through from ATi that went like this (the bold areas are mine not theirs):

Driver
The new 8.07 Driver is available as a BETA version from our webpage here:

http://www.ati.com/support/infobase/4649.html

The above page includes a link to a driver hotfix which will open the download window. This driver is of version 8.07 and fixes a random lock-up problem in "Star Wars: Knights of the Old Republic" that occurred with CATALYST 4.9.

This driver also fixes a memory allocation problem for all of our 256 MB products, this fix will result in a significant performance boost on applications that can take advantage of the larger frame buffer.

One of these applications is 3DMark05, the scores on 256 MB boards will differ greatly between this and older drivers, so we strongly suggest that reviewers use this new version (please note however that this is NOT a 3DMark05 specific optimisation, it is a generic bug fix related to boards with 256 MB of memory).

So I did as they said and grabbed the beta driver. Naturally I wasn't just going to take anyone's word for it that these were just addressing a frame buffer issue and armed myself with about 25 screenshots from the 4.7s and 4.8s so I could compare and contrast later.

As it turns out I needn't have bothered. over an hour later and very single screenshot, over a hundred of them when you count the coloured mipmap versions, when combined in Photohop using the "difference" option gave me a solid black screen. In other words the screens rendered using the 4.9 beta drivers were, as ATi stated, identical to all the rest. You see, that's why I'm not a detective.

Actually, the lack of any obvious sneakiness was quite a major surprise considering the numbers I got...take a look:

That's a 30% plus increase in the 3DMark05 score! I'm guessing the new 4.9 build will be seeing a lot of benchmarking action from 256MB ATi users this week eh?

More worrying though is the fact that ATi had frame buffer allocation issues serious enough to cause such a massive performance hit in the first place.

One of the welcome features of 3DMark03 when compared to 2001 was that it was much less system dependant. Just to make sure that trend had continued I ran "03" and "05" both at a stock 3.2GHz (16x multiplier) and also at a slightly toned down 2.8GHz (14x multiplier).

Here's what I got:

It would appear that if anything, 3DMark05 is slightly less system dependant that "03" was which is good news.

How the scores are calculated:

Each of the three game tests generates an average frame-rate (frames rendered per second measurement) that is used to calculate the overall 3DMark score. The formula for calculating the overall 3DMark05 score is:

3DMark05 score = (Game Test 1 * Game Test 2 * Game Test 3)^0.33 * 250

This formula generates a geometric mean, weighting the game tests equally on the total score. Even though one game test may be heavier (run slower) than another test, both of these affect the total score equally.

The formula for calculating the CPU score is:

CPU score = (CPU Test 1 * CPU Test 2)^0.5 * 1500

Leon's Conclusion:::…

Futuremark once again did it, releasing a firm and solid benchmarking tool for the masses.
It hosts an array of settings that can be adjusted endlessly to compare systems with each other.
It is easy to use and very complete for professionals and novices alike.
This benchmark meets future requirements for benchmarking needs for upcoming video cards, settings the bar just a lot higher.

Aside from benchmarking, 3DMark is also a nice but of eye candy, provided you can get it to run fast enough.
Of course you need to have the right bit of kit for that, which I didn't really have.
A Geforce FX 5950 doesn't quite make it at 5 FPS. (hint to our sponsors :))

Wayne's Conclusion:::…

Much as I enjoy 3DMark it's always a reallity check to see that new hardware you've been bragging about brought to its knees in such an undignified fashion. Of course it has to be this way or the next generation of GPU/VPU would just breeze though it leaving behind a trail of semi-meaningless CPU limited results.

I was surpised to see no curvy females gracing my screen. We all know sex sells (just ask Dawn and Ruby), but Futuremark have been very politically correct. Not even a cave-busting troll-dolly this time around, and my pixel shaders are just crying out for a bit of flesh to render!

3DMark05 is a demanding and beautifully rendered benchmark that will keep our graphics cards on the verge of exhaustion for another few generations, and as always the demo is a tour-de-force of pure visual splendor.

If the current performance levels, or lack of, from something as relatively beefy as an X800XT-Platinum Edition are an indication, Futuremark are either not planning to upgrade for at least a couple of years of they've got high hopes for the continued pace of performance gains.