![[H]ardOCP](/images/blank.gif){kind=small}
- Date:
- Monday , January 06, 2003
- Author:
- Brent Justice
- Editor:
- Sean Quinn
- Google +1
SiS Xabre 600 Review
Announced last November, the Xabre 600, successor to the Xabre 400, is poised to compete in the mainstream market, offering full DirectX 8.1 support. We examine the performance and "image quality" of this latest GPU from SiS.
Test Setup:
Gigabyte 8IHXP, P4 2.53GHz, 512MB Kingston PC1066 RDRAM, Maxtor 40GB ATA/133, 3Com 3C905C-TXNM, Windows XP Pro SP1, DirectX 9, Intel Chipset INF 4.04.1007
ABIT KD7-RAID, Athlon XP 2400+ (Multiplier Unlocked), 512MB Corsair XMS PC3500, Maxtor 40GB ATA/133, 3Com 3C905C-TXNM, Windows XP Pro SP1, DirectX 9, Hyperion 4n1 4.45
SiS Xabre 600 64MB – Driver version 3.08
ATI Radeon 8500LE 128MB – Catalyst 3.0 drivers
There are a few interesting points to note in this review on how we tested. The newly released official DirectX 9 from Microsoft was installed and used for testing. In the main performance tests, the Pentium 4 system was used to compare the Xabre 600 to the 8500LE where noted in the graphs. The 8500LE was chosen because it also has 4 pixel pipelines with 2 texture mapping units per pipe and supports DX 8.1 as well. In some tests the 8500LE was tested at its default clock of 250MHz/250MHz as well as 275MHz/275MHz and then overclocked to the same speed as the Xabre 600 in our review at 315MHz/315MHz to directly compare clock for clock. That is not a typo either, our Xabre 600 came clocked at 315MHz/315MHz stock.
We also did CPU scaling tests with the Xabre 600 on our Athlon XP machine which is multiplier unlocked. This allowed us to keep the FSB at 133MHz and the memory at DDR333 and simply change the multiplier to attain different CPU speeds for scaling. All those testing parameters are also noted in the graphs.
We also did image quality comparisons where needed to show you how that compares on the SiS Xabre 600. All tests, which include the main performance test, the CPU scaling test, and the image quality tests will be put on one page under each game used for the review. They will not be in separate sections. Hopefully this will make viewing the results much easier to look at and navigate.
Drivers and IQ Settings:
The Xminator-II unified drivers provide a system tray icon that gives access to the 3D Wizard.
The 3D stereo tab allows for 3D glasses support in various games that are supported. On the Direct3D tab, you have the options to disable Vsync and enable 2X, 3X, or 4X AA as well as change from Performance to Balanced to Quality settings. In the OpenGL tab, you have the same quality setting slider and AA slider and Vsync options, and you can also force a 16bit z-buffer. There is even an overclocking tab included! Finally, a Xmart tab allows you to enable the XmartDrive and XmartVision.
Image quality is very interesting to look at on the Xabre 600, and we will show you some comparisons later on in the review. For now, we are going to take a look at the Quality slider that is in the D3D and OpenGL tabs. This slider changes a key in the registry called TexTurboMode. This TexTurboMode effects texture quality either negatively or positively.
By default, the Quality setting is set in the balanced position. This sets a value of 3 for TexTurboMode. When set to performance, this value is still 3. However, when moved over to the Quality setting, the value changes to 1. In our testing, we found that Quality (value 1 for TexTurboMode) does indeed have better texture quality. A value of 3 (performance or balanced) has very bad texture quality, but is faster in performance.
Overclocking:
Overclocking was successful using the built-in overclocking tab. I was able to overclock the core up from 315MHz to a stable 325MHz, giving us a 10MHz increase. The memory was successfully overclocked from 315MHz to a stable 330MHz, which is an increase of 15MHz. Not the best overclock, especially considering it is .13u. Overclocking results are included where indicated in the review.
3DMark2001SE
We're using the latest version of 3DMark2001SE (build 330) for our testing. We ran each test at default 3DMark2001SE settings in the benchmark. TT=1 refers to TexTurboMode at Quality (value of 1) setting. TT=3 refers to TexTurboMode at Balanced (default value 3) setting.
When comparing TexTurboMode 1 to TexTurboMode 3 on the Xabre 600, we see that there is definitely a performance difference between the two quality settings. There's around a 300 3DMark point difference on average. Looking at Radeon 8500LE performance at 250MHz/250MHz, it outperforms the Xabre 600 at every resolution. Moving up to 315MHz/315MHz, which is the same clock speed as the Xabre 600, the 8500LE takes a hefty lead.
I decided to break down the most popular tests to compare pixel and vertex shader performance in the SiS Xabre 600. The nature test actually uses quite a bit of alpha textures, making it a very bandwidth oriented test. The 8500LE shows a huge advantage here compared to the Xabre 600, even clocked at 250MHz/250MHz compared to the Xabre 600 at 315MHz/315MHz. The difference in performance between the TexTurboModes here are only a few FPS apart. Overall, the nature test runs very slowly on the Xabre 600.
In the pixel shader test, the Xabre 600 does extremely poorly. What you are seeing is not a typo. There was no difference in TexTurboMode 1 or 3 in the pixel shader test. The 8500LE literally runs away with the cake here having faster pixel shaders.
The same scenario we saw in the pixel shader test is present in the advanced pixel shader test. The Xabre 600 is getting creamed.
The vertex part of the Xabre 600 does much better then its pixel shaders. Performance is comparable to the 8500LE at 250MHz/250MHz. However, it does show here that a hardware vertex shader IS faster than the Xabre 600’s software vertex engine which relies on the CPU.
You can view and compare all of the detail test results of the Xabre 600 above in each resolution with TT=1 and TT=3.
CPU Scaling:
The CPU scaling tests were run on the Athlon XP system on a 133MHz FSB and DDR333 with a multiplier change to attain the CPU speeds. 3DMark was run with default settings at TexTurboMode = 1 for quality on the Xabre 600.
There is a 1414 3DMark difference going from 1.2GHz to 2.2GHz. This doesn’t actually seem like a lot, considering the Xabre 600 is supposed to scale well.
At 1280x1024, where the video card gets stressed even more, the difference from 1.2GHz to 2.2GHz is even less.
Here there is only a 121 3DMark difference from 1.2GHz to 2.2GHz.
What can be summarized here is that at lower resolutions the Xabre 600 scales better than at higher resolutions, although it still does not scale as would be thought. It's obviously hitting some kind of limit in the higher resolutions.
Overclocking:
The Xabre 600 actually showed very good gains in performance in 3DMark with the little overclock we had. It was 578 3DMarks faster at 1024x768. The increase gets smaller as the resolution increases, with a 392 3DMark gain at 1600x1200.