MSI GeForce FX 5700 Ultra/NonUltra
Test Results: Synthetic benchmarks
Following the tradition, we start with our substantially amended benchmarking package.
ShaderMark v2.0
Deja-vu? The results of this synthetic benchmark that analyses the execution speed of various shaders may have made a dry mouth in this regard. It's been long known that version 2.0 shaders are executed on NVIDIA GeForce FX quite poorly and hardly can help much but for the close cooperation with game developers who would optimize their engine code specially for the NVIDIA architecture, which is being successfully done. The results are self-explanatory if we recall our recent tests with FarCry which are part of NVIDIA initiative dubbed "The way it`s meant to be played". We can see that boards on NVIDIA chips lagged behind ATI cards quite a bit. Anyway, repeating it again that such an approach is extensive by its nature, and NVIDIA can radically improve its positions only through release of a new chip.
For now we can merely ascertain sometimes a three-fold lag of NVIDIA chips at that test relative to ATI chips of similar positioning.
D3D RightMark: Geometry Processing Speed
This test allows assessing the speed at which the geometry is processed by the accelerator. The test allows to choose the following illumination models (computed at the vertex level):
- Ambient Lighting — simplest constant illumination
- 1 Diffuse Light
- 2 Diffuse Lights
- 3 Diffuse Lights
- 1 Diffuse + Specular Light
- 2 Diffuse + Specular Lights
- 3 Diffuse + Specular Lights
We used the most advanced mode with three diffuse-specular light sources in combination with three different operating modes: the traditional TCL (Fixed-Function Pipeline), version 1.1 vertex shaders and version 1.1 pixel shaders, version 2.0 vertex shaders and pixel shaders, as well as the simplest mode of ambient illumination is also in combination with the three operating modes: the traditional TCL (Fixed-Function Pipeline), vertex shaders 1.1 and pixel shaders 1.1, vertex shaders 2.0 and pixel shaders 2.0. In the case of Ambient illumination (the simplest constant illumination) and transformation we reach the practical limit for the card's bandwidth at processing the triangles.
 Let's look at the test results. So, the first case deals with the simplest mode using ambient lighting in combination with three operation modes: the traditional TCL (Fixed-Function Pipeline), vertex shaders 1.1 and pixel shaders 1.1, vertex shaders 2.0 and pixel shaders 2.0. Despite the greater number of vertex processors in NVIDIA chips (in NVIDIA GeForce FX 5700 and NVIDIA GeForce FX 5700 Ultra they are 3, while in ATI Radeon 9600 Pro and ATI Radeon 9600 Pro they are merely 2), the processing of geometry in the case of the simplest lightning source the NVIDIA chip does better. The results do not differ during the switching from hardware T&L emulation to the use of version 1.1 and 2.0 shaders. At that, ATI Radeon 9600 XT is an undisputable leader. All is natural in that the results produced in the ambient lighting mode depend to greater extent on the clock speed of the graphic chip which is the highest of all the accelerator presented herein (i.e. in ATI Radeon 9600XT). The NVIDIA GeForce FX 5700 chip is definitely not bright in terms of geometry and thus takes the last place.
 Now we are moving from the simplest lightning mode towards the most advanced model with three diffuse-specular light sources. Here the picture radically changes in various modes and requires additional explanation. In the Fixed-Function Pipeline mode, we see a considerable leadership of NVIDIA boards. This is quite a natural victory of NVIDIA chips in that ATI chips simply have no hardware support for T&L emulation, but NVIDIA chips offer the implementation of hardware units responsible for light source computation.
With the transition to version 1.1 vertex shaders, the results for NVIDIA show a significant two-fold drop in their peak values. And finally here comes the most exciting thing - version 2.0 shaders: ATI chips take a lead! It is quite natural in that execution of any functions on the base of version 2.0 pixel and vertex shaders on NVIDIA chips runs strikingly slower than with DX 8.1 version shaders. At that, flaws in the NVIDIA GeForce FX architecture make themselves felt.
Pixel Filling
This test performs a number of various tasks, but we were mostly interested in the possibility of measuring the performance of frame buffer filling. We used two schemes in this test: both with a 256x256 texture, and without such.
The values of frame-buffer fill-rate produced in this test with the ATI chip do not coincide a bit with the theoretical maximum announced by ATI, but with NVIDIA everything is almost practically correct for the case of the Texelrate (Color+Z) mode.
In this test, the 4x1 NVIDIA GeForce FX 5700 Ultra as well as NVIDIA GeForce FX 5700 architecture are evident, which shows, as was noted above, that the results fully match this configuration of the number of pixel processors and texture units jointly with the chip's clock speed. The peak values differ from the theoretically admissible by merely 40-50 mln pixels per second, which in view of the produced absolute values can be regarded as a minor error. However, in this test ATI chips do not reveal their theoretically announced pixel fillrate bandwidth. For example, in the ATI Radeon 9600XT chip, declared was 500õ4 = 2.0 Gpixels/s, but in fact we get merely 1.1 Gpixels/s, which is clearly twice as less than the theoretical maximum. We can't tell the cause of such behavior of ATI chips, however, it's not the first time already, since during tests of ATI Radeon 9800XT we produced exactly the same queer results.
 In the case of adding just one more bilinear texture sampling, the peak values of pixel fillrate dropped, but the general picture remained as before.
Pixel Shading
This test in the D3D RightMark benchmarking package allows to estimate the performance of executing various pixel shaders of the second version. In this test, the geometry has been substantially simplified to minimize the dependence of results of the test on the geometric performance of the chip and verify the operation of pixel pipelines only. We've brought in the operational modes for both 16-bit and 32-bit floating-point precision (switching between the precision modes is topical for NVIDIA chips only).
NVIDIA chips proved shattered at performing version 2.0 shaders under 32-bit floating-point precision. Yes, we can make reservation for that the ATI chip was running that time at its customary 24-bit precision, but let's look at the fps values produced with 16-bit floating-point precision. In this case, the NVIDIA chip is in a vantage point, since it uses less precision rate than the ATI chip (whose absolute values, as expected, do not change - it runs at its invariable 24-bit).
Yes, the fps values in NVIDIA chip went up expectedly, but in the end it is again a fiasco. By the results of the test, NVIDIA chip executes pixel programs twice as slower than ATI chips do.
Point Sprites
This test is aimed at revealing the accelerator speed at displaying point sprites. Let's enumerate the adjustable test parameters:
- Resolution
- Windowed and full-screen mode
- Testing times (statistics accumulation) in seconds
- Vertex shader software emulation mode
- Operation mode:
- Vertex Shaders 1.1 and Fixed Function Blend Stages
- Vertex Shaders 2.0 and Fixed Function Blend Stages
- Animation mode:
- Lightning mode:
In the test settings, we used 2 diffuse light sources and enabled animation. We also investigated the dependence of execution speed on displaying point sprites on the version of vertex shader used.
In the total, NVIDIA chips are definite outsiders at that. See for yourselves: NVIDIA GeForce FX 5700 Ultra shows results precisely on par with ATI Radeon 9600Pro with which NVIDIA GeForce FX 5700 should have competed, but it lagged in this test quite substantially (the low clock speeds must have played a bad trick at that).
Hidden Surface Removal
This test allows to estimate the efficiency of removal of hidden points and primitives by the accelerator. The randomly generated scene will then be displayed in one of the three selected modes:
- sorted, in the front to back order
- sorted, in the back to front order
- unsorted
Also, we can investigate the dependence of the efficiency of hidden points and primitives removal efficiency on the version of vertex shader used (1.1 or 2.0).
The NVIDIA GeForce FX 5700 Ultra chip is beyond competition. The NVIDIA GeForce FX 5700 also ranks pretty good at that, however, in this test our settings play in favor of NVIDIA chips - we perform texture sampling, which can't give us 100% adequate results when estimating the efficiency of the chip's HSR unit. In the nearest future, this part of our analysis of accelerator operation will be complemented with one more HSR efficiency estimation mode in order to produce a clearer idea in this regard.
The version of vertex programs hasn't brought any changes to the alignment of forces.
Codecreatures
The well-known Codecreatures benchmark shows a definite leadership of MSI boards built on NVIDIA chips. NVIDIA optimizations added to the good characteristics of the chips NVIDIA GeForce FX 5700 and NVIDIA GeForce FX 5700 Ultra give a not bad result at all - the NVIDIA GeForce FX 5700 Ultra proves to be an indisputable leader, and NVIDIA GeForce FX 5700 shows results on par(!) with ATI Radeon 9600XT.
Tests run with the antialiasing hasn't brought anything new into the alignment of forces among the boards.
But in view of the implementation specifics observed in the AF algorithms used for NVIDIA chips, the alignment of forces changes a bit, and ATI Radeon 9600XT finally overtakes the MSI's board built on the NVIDIA GeForce FX 5700 chip.
3DMark 2003 v340
We are traditionally tampering with 3DMark :). There were many times when got reasons to doubt the evidence of this benchmark, so we'd better not judge once again who was right or wrong at that benchmark, but will simply present the results for this comprehensive package just for general information only :).
 |
Top Stories: |
 |
 |
|
MoBo:
| |
|
|
VGA Card:
|
|
|
|
CPU & Memory:
|
|
