|
• AMD Athlon XP3200+ 'Barton' Processor, 2200MHz, 512KB L2 cache, 11 x 200MHz, Socket A
• Asus A7N8X Deluxe 2.0 motherboard, nForce2 Ultra 400 chipset, Socket A, 1003 BIOS
• 2 x 256MB Corsair XMS3200 memory, CL2, 6-2-2, DDR400
• MSI GeForce FX 5900-VTD256
• MSI GeForce FX 5900-TD128 (MS-8929)
• Hercules ATI Radeon 9800 Pro
• ATI Radeon 9800 Pro 256MB reference
• Windows XP Professional Build 2600.xpclient.010817-1148 w SP1
• ATI CATALYST 3.4
• NVIDIA Detonator FX 44.03 reference drivers
• NVIDIA nForce 2.41 platform drivers
• DirectX 9.0 Runtime
• 3DMark 2001SE v330
• UT2003 Retail (Build 2225)
• Comanche 4 Demo
• Serious Sam 2 Demo
• Quake3 v1.30
No midrange or low end finery here, just four of the fastest graphics cards on the planet. The MSI FX5900 and Hercules' Radeon 9800 Pro are the 128MB representatives, while I managed to drag ATI's 256MB reference board out of the closet for the fat framebuffer comparo with ATI's finest.
While there are newer beta drivers out for NVIDIA's GPU's, along with 2 new generations of CATALYST around for ATI's boards, I stuck with 44.03's and CATALYST 3.4 for consistency with older results.
If I'm doing my sums right, that's well over £1000 worth of graphics hardware, expensive as well as quick.
For the testing, 4xAA on the GeForce FX boards means 4x. Not Quincunx (soft 4x with 2x performance, a nice looking mode in my opinion) or 4XS which is a Direct3D only mode, just plain old 4x. The quality slider was at Quality at all times, here's a shot of the Intellisample control panel for setting it. To set 4xAA and 8xAF, only the two sliders were adjusted. Texture sharpening was off for all tests.
Driver Quibbles
I think it's time to say something about the recent furore over NVIDIA's (and indeed ATI's) drivers. I'd pondered a co-written editorial with Tarinder, or some front page ramblings a-la Tech Report, but a footnote here seems apt.
To put it bluntly, NVIDIA's 44.03 Detonators are a mess. Applying all kinds of 'optimisations' to pretty much all the standard benchmarks we run. The net effect, utter skewing of our recent results, since 44.03 is the driver we put faith in for recent looks at GeForce FX. Not good, and I'm definitely not happy about it. As you'll see in this review, the 256MB Ultra FX5900's results are out of place on a couple of the tests and it's almost entirely down to the driver.
So, what to do? Well, I've left the skewed results in this review, commenting on them when necessary, so as not to give illusions of performance that really isn't deserved.
But for future reviews (maybe not the next couple from me) we'll be moving to a new test suite at HEXUS. One that does its level best to avoid being detected by drivers, which is how NVIDIA (and ATI) recognise an application being run and apply their 'optimisations' to it. We don't want that thanks, we want the settings we chose, not the ones a driver maker thinks we should have.
Also, when we've managed to procure a test 256MB Radeon 9800 Pro from ATI, we'll run all the currently available consumer cards through their paces on the new test suite, maybe even with an updated driver from NVIDIA, since that's the only card we are missing. That will give us a fairer look at current performance than we have at the moment.
Anyway, until then, keep an eye on 44.03 based test results, they aren't exactly accurate.
At least NVIDIA hve owned up recently and promised to fix it. Good show, make sure you follow through on that guys.
I must stress that this isn't MSI's fault, it just happens that it's an MSI card that I'm looking at today.
|
