Connect3D Radeon 9700 Pro in the test: performance leap at high resolutions

Connect3D Radeon 9700 Pro in the test: performance leap at high resolutions

Table of contents
  1. 1 Introduction
  2. 2 The card
  3. Scope of delivery
  4. 3 Drivers
  5. 4 Specs and technology
    1. 5 FSAA and AF
  6. 6 Benchmarks
  7. Test system
  8. 7 Synthetic tests
    1. 3DMark2001 SE
    2. 8 3DMark Detail
    3. 9 Villagemark D3D
    4. 10 Villagemark Detail
    5. 11 TempleMark D3D
    6. TempleMark D3D detail
    7. 12 Codecreatures Bench
    8. 13 Codecreatures Detail
    9. 14 Aquamark D3D
    10. 15 Aquamark D3D detail
  9. 16 Game benchmarks
    1. Ultima IX D3D
    2. 17 Ultim a IX D3D Detail
    3. 18 Comanche4 D3D
    4. 19 Comanche4 D3D Detail
    5. 20 Dungeon Siege
    6. 21 Dungeon Siege Detail
    7. 22 Aquanox
    8. 23 Aquanox in detail
    9. 24 Jedi Knight II
    10. 25 Jedi Knight II Detail
    11. 26 Max Payne
    12. 27 Max Payne detail
    13. 28 Serious Sam SE D3D
    14. 29 SeSam SE D3D Detail
    15. 30 Serious Sam SE OGL
    16. 31 SeSam SE OGL Detail
    17. 32 UT2003 Demo FlyBy
    18. UT2003 Demo Botmatch
    19. 33 UT2003 FlyBy Detail
    20. UT2003 Botmatch Detail
    21. 34 Alice
    22. 35 Alice Detail
  10. 36 Conclusion

Specs and technology

Now let's take a quick look on the architecture and technical specifications of the Radeon9700pro. ATi itself of course provides extensive, but also very marketing-heavy information in order to induce interested parties to buy. If you want to take a look, you will find it here .

We have tried to filter out what is really relevant and give a small overview of the basic structure of ATi's flagship, which is also the Connect3D Radeon9700pro drives to give.

Key data of the Radeon9700pro at a glance
Radeon9700pro Radeon8500 GeForce4 Ti4600 Chip name R300 R200 nV25/nV28 Manufacturing process 0.15µ 0.15µ 0.15µ transistor count ~ 107million ~ 60million ~ 63million chip clock 325MHz 275MHz 300MHz rendering pipelines 8 4 4 TMU per pipeline 1 2 2 filling rate (billion pixels/s 2.6 GPix/s 1.1 GPix/s 1.2GPix/s fill rate (billion texels/s) 2.6 GTex/s 2.2 GTex/s 2.4 GTex/s fill rate (billion AA samples/s) 15.6 GSamp/s 1.1 GSamp/s 4.8 GSamp/s max.textures per pass 8 6 4 max.texture accesses per pass 16 11 4 memory size 128MB 64MB/128MB 128MB memory clock 310MHz 275MHz 325MHz RAM bus width 256Bit DDR 128Bit DDR 128Bit DDR RAM- Bandwidth (billion Byte/s) 19.84 GB/s 8.8 GB/s 10.4 GB/s vertex shader/TnL units 4 1/1 2 max. Vertex shader version 2.0 1.1 1.1 max.pixelshader version 2.0 1.4 1.3 max. vertex throughput ~ 325M vertices/s ~ 69M vertices/s ~ 136M vertices/s integrated RAMDACs 2x 400MHz 1x 400MHz 2x400MHz max.AGP mode AGP 8x AGP 4x AGP 4x/8x

Since we want to go into the DirectX9 functionality in a separate table, we will concentrate here for the time being on the possibilities of the existing hardware with current software.

As with the Radeon9000 and the recently introduced nV30, there was no further increase in TMUs per pipeline for the R300. The fill rate, which is indispensable for a high-end model, can be achieved by 'simply' doubling the entire render pipelines. This has the positive side effect that there is now a pixel fill rate that is more than twice as high as that of the closest competitors and that you can act much more flexibly with multitexturing thanks to the deep loopback functionality.

So that this enormous raw performance does not lie idle , ATi has also made a splash with the memory interface and has integrated a 256-bit DDR-RAM interface. The available memory bandwidth is almost doubled and its effective utilization is ensured by a quadruple memory controller (waiving it could have the same effect as with the Matrox Parhelia, which is grandiose flopped in terms of gaming performance).

The R300 also easily tops the 2 vertex shaders available on the GeForce4 Ti and doubles their number to 4, which results in the enormous (theoretical !!) transformation performance of more than 300 million corner points per second. As with the Radeon9000 (RV250), the R300 now also integrates two high-speed RAMDACs, thus ensuring that the multi-monitor feature is also available on all cards with two monitor outputs, unlike manycheap Radeon8500 cards of the last series. The AGP 8x functionality, which is now increasingly found, can be dismissed as a marketing feature for the time being, especially in the absence of mature mainboards that also support this standard.

Shader functionality
R300 nV25 vertex shaders Max.instructions (including loops) 65536 32 static instructions 256 128 Static flow control Yes No Dynamic flow control No No Texture access No No Pixel shader Floating point precision (max. precision) Yes (96Bit) No (32Bit) Flow control No No Multiple Render Targets Yes (4) No Max. Texture addressing instr. 32 4 Max. Instructions during runtime 160 (DX9 PS2.0: 96) 4

Like A few selected specifications clearly show that the shaders according to the DirectX8 standard that were formerly (also by us!) praised as 'freely programmable' units can only be used to a very limited extent, so that more and more former enthusiastic advocates of this technology are now seeing in DirectX9 what they thought they saw in DX7 before and then in DX8. In any case, the increased precision values ​​and the floating point suitability of the pixel pipelines are very advantageous, as this represents a not to be despised advance within the scope of the possibilities of the overall product, especially with complex material defects and light calculations. Due to the higher precision of the internal calculations and the more balanced distribution of the value range, especially in the light area, it is now possible to realize lighting and material surfaces with a much more 'real' effect than was the case recently. But here, too, you don't need to fool yourself, as the first extensions of theDX9 specification are in prospect and with DirectX10 at the latest everything 'old' will appear thoroughly out of date.

Further features of the Radeon9700pro:

  • Smartshader 2.0 This includes the extended shader options of the Radeon9700 mentioned above.
  • Smoothvision 2.0 Smoothvision now includes all options regarding full-screen antialiasing and anisotropic filtering. The former was completely converted to multisampling and really accelerated drastically by designing the hardware accordingly with 6 generated AA samples per pipeline. Instead, it now only smooths polygon edges and no more textures, as with nVidia. The anisotropic filter has been corrected so that it can finally combine the fast 16x AF with trilinear mipmap interpolation.
  • Videoshader Formerly known as Motion Compensation, iDCT (inverse discrete cosine transformation) Features for which the necessary additional hardware is now being replaced by the pixel shader.
  • Fullstream An innovation with the ATi using suitable plug-ins (currently only for the paid version of the Real-One Player) wants to get an increased image quality out of small pixelated-blocky video streams from the Internet by filtering them through suitable pixel shaders.
  • TruForm 2.0 The former is only the one TruForm comprehensive n-Patch support has now blossomed into DirectX9 Displacement Mapping , which should actually take care of the Legacy TruForm Support on the side. 'Should' because so far no driver (including the BETA DirectX9 driver) is able to display TruForm really at an acceptable speed.
  • Hyper-Z III That through all the Radeon generations available to date, achieved in itsthird incarnation a new level. Through lossless Z-buffer compression (ATi specifies a rate of up to 24: 1), a three-tier hierarchical Z-buffer and block-wise overwriting of the Z-buffer in combination with the four-fold sub-system, one tries to avoid unnecessary rendering of invisible pixels and the utilization of the 256-bit interface to a maximum.

On the next page: FSAA and AF