|
Written by Mavke
|
|
Saturday, 02 October 2004 |
Seems we forgot the review at Bit-Tech on 3DMark05. 3DMark03 was something of a mixed bag on its release some 18 months ago - designed to usher in a new era of spanky DirectX 9.0 graphics, it was soon discovered that in fact several of the tests were somewhat DirectX 7 and 8 biased. The relative newness of DirectX 9.0 hardware was blamed - it would have alienated too many users who were still running older generations. As a result, it landed in no-man's land - many people persisted with 3DMark2001 as it was an old stalwart of measuring system performance, but many people including ourselves moved away from using 3DMark - with the invention of "real-world" measures like FRAPS, we could measure a card's performance using the things that mattered the most: Games!
 3DMark05 Review
Fast forward to 2004, and after this Spring's release of the latest generation cards from both nVidia and ATI, even the 3DMark fans admitted that 3DMark03 was no longer the stressful workout it once was - with scores heading towards 20,000 (and 3DMark2001 scores nearing 40,000!), it was time for something new. Thankfully, the abundance of DirectX 9.0-compatible cards out there has allowed Futuremark to ditch the legacy components of their test suite and focus more on the latest, cutting edge features such as nVidia's Shader Model 3.0 and ATI's Shader Model 2.0b HLSL's (higher level shader languages).
We've covered an awful lot of technology here, which should give a good insight in to the technology behind 3DMark05. However, our feelings about this benchmark are slightly mixed. We're starting to believe that moving away from time demos and synthetic benchmarking is a step in the right direction. After all, 3DMark is a synthetic benchmark that ATI and nVidia appear to concentrate on more so than the majority of game titles, because being number one is so crucial for their sales. Really speaking, they should be focusing their funds and manpower on fixing driver bugs and increasing performance in the top game titles. From our perspective, there are some useful feature tests included, but apart from that we are not overly impressed by the latest 3DMark release - this is probably due to the fact that we were spoilt with eye candy in the recent release of Doom 3.
We were expecting something quite different to what has been presented, and quite frankly, the poor image quality, texture shimmering and over-optimisation that is already apparent has got to stop. 3D graphics performance should not just be about a score or a frame rate, image quality is paramount in ensuring that the user is experiencing a fantastic overall game experience in this day and age. This is where we were expecting the default test to use a certain degree of Anti-Aliasing and Anisotropic Filtering. We understand that Futuremark have to enable the benchmark to run on as many people's computers as possible, and we cannot fault them for choosing to stick with the 1024x768 resolution that has been used over the past 3 versions of 3DMark. However, the advanced shaders and lighting effects that are used could be done with slightly less stress placed on the geometry pipeline in order to sacrifice complexity for something that is actually pleasing to look at.
The fact that the application doesn't scale well with resolution is an indication that our theories that the performance is limited by the frame buffer required to render the huge geometry instructions - we can expect a 512MB board to receive a huge performance boost, as there will no longer be the requirement for texture thrashing to occur across the AGP/PCI-Express interface. This brings worries to the performance of the HSI bridge chip, which could, quite likely reduce the performance of the GeForce 6800 series across the PCI-Express interface. There have been no tests to prove that texture thrashing has a telling effect on the HSI bridge chip at the moment - maybe this is the first of many?
Where does this leave us? Well, to be quite honest, we're not quite sure at the moment. 3DMark05 has a comprehensive set of Feature Tests that are useful to us. We're very sceptical about using 3DMark as an actual benchmark, as both the developers; along with both ATI and nVidia seem to feel that the score is far more important than the actual image quality. We cannot deny that 3DMark05 severely stresses even today's bleeding edge graphics hardware, but we have to ask whether there is any use in it when there's nothing to be experienced in terms of image quality. It's evident that the default test does not provide us with any great level of image quality, in fact we were more than disappointed at lack of image quality in the form of Anti-Aliasing and Anisotropic Filtering, which should really be enabled in the default test. An idea that we've come up with, would be for the feature tests to run with Anti-Aliasing and Anisotropic Filtering to be forced off in the feature tests, unless specified otherwise by the user, while the game tests were ran with Anti-Aliasing and Anisotropic Filtering enabled in a higher level than what is the case in the current default test - it would give a more realistic bearing on today's game performance.
|
|
Last Updated ( Saturday, 02 October 2004 )
|
EVGA 9600GT (SC) PCIe 512MB - Rev 01
