FutureMark : Bench Giant Interview
David Legrand le 09 février 2006 (10 685 lectures)
PCi : 3DMark06 now uses SM 3.0, HDR, and the rendering of more complex scenes. Why didn’t you comply earlier with DirectX 9 specifications, supported by NVIDIA's Geforce 6 cards, yet released more than one year ago ?
TS : The idea of 3DMark is to be forward looking (i.e. feature workloads that games appearing in the future will employ) by about 1 to 2 years. With this in mind, when we specced and prepared 3DMark05, it was the farthest we could go – even then some folks said that we are crazy when we launched 3DMark05, they said it is far too demanding and no games would anytime soon utilize those kinds of workloads. Sure enough, games of today are now starting to be at the level of 3DMark05 :)
3DMark 2005 en pleine action
PCi : Why and which features in particular on this new version you think make the product a must in today's available benchmarking tools ?
TS : Without a doubt it is the HDR and the threaded CPU tests. HDR enables effects in real-time 3D graphics that were impossible before it and the difference is mind blowing. We have posted comparison screenshots of 3DMark05 vs 3DMark06 and the difference is just like night and day!
NR : HDR, threaded CPU tests and SM3.0 are the key features of 3DMark06. Of course the benchmark as a whole is a package no reviewer should miss since 3DMark06 also comes with great feature tests, an image quality tool, a filtering & anti-aliasing quality tool and much more. 3DMark06 is not only about performance measurement (though it is its primary function), but also about rendering quality.
HL2 Lost Coast avec AA & AF, sans HDR complet puis avec
PCi : ATI's Radeon X8xx and X1k scores were higher than NVIDIA's products in previous 3DMark05 (except SLI obviously). This tends to be the reverse using the new 3DMark06, could explain why ?
NR : We usually don’t comment on various IHV’s performance differences since we are not in the position to do so. We have posted 3DMark06’s Whitepaper online which includes a lot of in-depth information how 3DMark06 works. While 3DMark06 uses some improved content from 3DMark05, the two benchmarks have very different workloads. 3DMark06 has more dynamic shadowing and lights in the SM2.0 graphics tests than 3DMark05 and the new HDR/SM3.0 tests are fully loaded with SM3.0 features and true HDR rendering.
PCi : ATI seems to think that you don't take enough into account AA and AF filtering capabilities of each card in assessing the overall score, what's your response to this criticism ?
NR : I personally haven’t heard/seen any such criticism, but there are a couple of reasons why we don’t have Anisotropic filtering and/or Anti-Aliasing on as default in 3DMark06. The first reason is that neither of the two features are requirements in the DX9 specifications. The second reason is that there is no defacto quality level for Anisotropic filtering, or Anti-Aliasing which would make qualifying and approving drivers very difficult.
As soon as both these image quality enhancing features make it into the DX specifications as required features we don’t see any reason not to use both of them as a default setting in 3DMark. Note, however, that one can enable those features in the benchmark settings.
PCi : ATI’s strategy seems to be pushing pixel shaders as a key component of the architecture with its new Radeon X1900 models. What do you think about this choice, and what are 3DMark's software developers' opinion ?
NR : Again, we are not in a position to comment on any IHV’s architectural decisions.
PCi : Previous 3DMark versions suffered from the release of specific, enhanced drivers, allegedly released by ATI and NVIDIA. Are you considering to do anything to prevent this with 3DMark06 ?
TS : We are actively working with all industry companies and encouraging everyone to comply with our driver optimization guidelines. We inspect all new WHQL certified official drivers against the guidelines. This is to protect the integrity of benchmark results. Beware of using a benchmark that does not have this control and beware of using non approved drivers.
NR : It is important for media professionals and all serious about getting trusted benchmark scores to use Futuremark Approved drivers when benchmarking with our products to make sure that the results are comparable and valid. We will continue to approve drivers for 3DMark03, 3DMark05 and 3DMark06 and recommend media people (and users) to only use such drivers in their reviews.
TS : The idea of 3DMark is to be forward looking (i.e. feature workloads that games appearing in the future will employ) by about 1 to 2 years. With this in mind, when we specced and prepared 3DMark05, it was the farthest we could go – even then some folks said that we are crazy when we launched 3DMark05, they said it is far too demanding and no games would anytime soon utilize those kinds of workloads. Sure enough, games of today are now starting to be at the level of 3DMark05 :)
3DMark 2005 en pleine action
PCi : Why and which features in particular on this new version you think make the product a must in today's available benchmarking tools ?
TS : Without a doubt it is the HDR and the threaded CPU tests. HDR enables effects in real-time 3D graphics that were impossible before it and the difference is mind blowing. We have posted comparison screenshots of 3DMark05 vs 3DMark06 and the difference is just like night and day!
NR : HDR, threaded CPU tests and SM3.0 are the key features of 3DMark06. Of course the benchmark as a whole is a package no reviewer should miss since 3DMark06 also comes with great feature tests, an image quality tool, a filtering & anti-aliasing quality tool and much more. 3DMark06 is not only about performance measurement (though it is its primary function), but also about rendering quality.
HL2 Lost Coast avec AA & AF, sans HDR complet puis avec
PCi : ATI's Radeon X8xx and X1k scores were higher than NVIDIA's products in previous 3DMark05 (except SLI obviously). This tends to be the reverse using the new 3DMark06, could explain why ?
NR : We usually don’t comment on various IHV’s performance differences since we are not in the position to do so. We have posted 3DMark06’s Whitepaper online which includes a lot of in-depth information how 3DMark06 works. While 3DMark06 uses some improved content from 3DMark05, the two benchmarks have very different workloads. 3DMark06 has more dynamic shadowing and lights in the SM2.0 graphics tests than 3DMark05 and the new HDR/SM3.0 tests are fully loaded with SM3.0 features and true HDR rendering.
PCi : ATI seems to think that you don't take enough into account AA and AF filtering capabilities of each card in assessing the overall score, what's your response to this criticism ?
NR : I personally haven’t heard/seen any such criticism, but there are a couple of reasons why we don’t have Anisotropic filtering and/or Anti-Aliasing on as default in 3DMark06. The first reason is that neither of the two features are requirements in the DX9 specifications. The second reason is that there is no defacto quality level for Anisotropic filtering, or Anti-Aliasing which would make qualifying and approving drivers very difficult.
As soon as both these image quality enhancing features make it into the DX specifications as required features we don’t see any reason not to use both of them as a default setting in 3DMark. Note, however, that one can enable those features in the benchmark settings.
PCi : ATI’s strategy seems to be pushing pixel shaders as a key component of the architecture with its new Radeon X1900 models. What do you think about this choice, and what are 3DMark's software developers' opinion ?
NR : Again, we are not in a position to comment on any IHV’s architectural decisions.
PCi : Previous 3DMark versions suffered from the release of specific, enhanced drivers, allegedly released by ATI and NVIDIA. Are you considering to do anything to prevent this with 3DMark06 ?
TS : We are actively working with all industry companies and encouraging everyone to comply with our driver optimization guidelines. We inspect all new WHQL certified official drivers against the guidelines. This is to protect the integrity of benchmark results. Beware of using a benchmark that does not have this control and beware of using non approved drivers.
NR : It is important for media professionals and all serious about getting trusted benchmark scores to use Futuremark Approved drivers when benchmarking with our products to make sure that the results are comparable and valid. We will continue to approve drivers for 3DMark03, 3DMark05 and 3DMark06 and recommend media people (and users) to only use such drivers in their reviews.
Sommaire
- 1. Presentation, the society, relationships
- 2. 3DMark 2006 : Part 1
- 3. 3DMark 2006 : Part 2
