Xabre 400 Roundup - August 2002
Now that we've got a better idea of our 'guinea-pigs' reviewed, it's high time we tested and found out their potentials. Before moving on to the results, let me take a few minutes of your time to tell what and how we tested.
For comparisons, we used all the five Xabre400 video cards received for testing - a reference card from Elitegroup, one from Vinix, one from PowerColor and from Gigabyte. All those cards were tested at the normal and overclocked modes. In the nearest future a new chip, Xabre200, will come into play, differing from Xabrea400 in the clock speed only: it will run at 200/166(333) MHz. So, to get an idea of its performance, we pushed the clock speed of the reference board Xabre400 down to that of Xabre200 and produced a 64-MB analog of a still non-existent video card.
Since we did our tests on a motherboard based on the VIA KT333 chipset, all the Xabre video cards switched into the AGP2x mode automatically. Frankly, the performance difference for these video cards at AGP2x, 4x and 8x will be negligible, considering that each of the cards had 64 Mb onboard. Since the flaw comes from some engineers at SiS, and VIA chips sell like cakes, we won't accentuate that fact - mind you, the tests were performed in real-life conditions, on a real computer, and Xabre videocards have got to be prepared to face these like any other kind of hardware.
As far as the installation is concerned, most video cards had no trouble installing under Windows XP. Only PowerColor EvilXabre400 was reluctant to install and wouldn't work in a mode other than VGA. Perhaps that was caused by installation of extra 64 Mb, maybe the right drivers are not yet ready, but I would call the drivers almost a miracle, and I wouldn't want to replace them with newer ones for that video card.
For comparisons, we used two mid-end nVidia-based video cards of two competitors. Those were cards based on GeForce4 MX440 and GeForce4 Ti4200. The first one had 64 MB DDR SGRAM and ran at 270/400 MHz. The second one had 128 MB DDR SDRAM and ran at 250/444 MHz. Well, let's start.
Tests
- Processor AMD AthlonXP 1800+;
- Motherboard: Jetway V333DA;
- Memory: 512 Mb DDR PC2700;
- Video cards: Xabre reference board; ECS AG400; Gigabyte SP64D-H, GeForce4 MX440, GeForce4 Ti4200 128Mb;
- HDD - Quantum 30Gb AS+ 7200 RPM;
- Monitor - LG Flatron 795FT Plus;
- Drivers - Detonator 29.80;
- SiS 3. 03. 51;
- OS - Windows XP.
2D
In the first test we analyzed the 2D graphics and gauged the performance. The issue of 3D quality has already been covered. Unfortunately, nothing good was said about that, - but in the 2D the situation is reverse. The 2D image quality in the reference board from SiS and Gigabyte SP64D-H is simply fantastic, no soaping was observed even at high resolutions. Can't be compared with competitors. SiS did a really great job of that - that's what the 400 MHz RAMDAC means! The layout of the boards is excellent. But Elitegroup spoiled all with the awful design of AG400, which resulted in that the video card failed to run at resolutions higher than 1024x768 at 85Hz. The "soaping" will also spoil the scene. The same was observed for the Vinix board. Visually, the SiS reference board was the best performer, with the Gigabyte counterpart following quite close, then follow GeForce4-based video cards (at virtually the same score),then PowerColor EvilXabre made a big leap from the counterparts,whereas the ECS AG400and Vinix VX-3340 lagged far behind with the worst scores. With this awful quality, Vinix VX-3340 anyway has an advantage over the ECS AG400, namely: via a DVI-VGA adapter a monitor can be plugged to it, which will improve the quality slightly, but it will be restricted to the resoultion 1280x1024 at 32-bit color depth. And one problem is there: such an adapter is not shipped with the package, so we would have to look for it elsewhere.
We tested the 2D speed using the PCMark2002 benchmark. The 2D speed is fairly high and enough for most users, so we tested video cards at the rated modes at rated speeds. Here are the results, with the best highlighted:
| 2D Tests in PCMark2002 |
| Video card / Test |
Xabre200 |
Xabre400 |
ECS AG400 |
Gigabyte SP64D-H |
Vinix VX-3340 |
PowerColor EvilXabre400 |
GeForce4 MX440 |
GeForce4 Ti4200 |
| Lines |
14936.7 |
15031.3 |
14872.3 |
14878.9 |
14990.6 |
14991.4 |
17768.6 |
17832.4 |
| Ellipses |
5450.48 |
5452.2 |
5447.7 |
5338.12 |
5420.3 |
5410.2 |
5996.7 |
6011.46 |
| Arcs |
10976.5 |
10999.6 |
10988.1 |
10959.3 |
10990.2 |
10961.3 |
12334.1 |
12375.1 |
| Beziers |
6819.5 |
6819.7 |
6820.7 |
6808.1 |
6813.3 |
6820.3 |
7800.16 |
7795.45 |
| Curves |
1521.6 |
1535.3 |
1535.58 |
1535.25 |
1535.35 |
1533.3 |
1648.3 |
1647.64 |
| Filled Ellipses |
15082.5 |
15083.6 |
15076.7 |
15054.7 |
15080.6 |
15076.6 |
23583.7 |
25558.9 |
| Gradient Boxes |
41.99 |
42.82 |
42.86 |
42.73 |
42.81 |
42.85 |
34.76 |
32.74 |
| Gradient Polygons |
11.72 |
11.93 |
11.95 |
11.92 |
11.91 |
11.93 |
10.29 |
9.68 |
| Filled Pies |
9170.7 |
9190.1 |
9186.4 |
9173.3 |
9170.2 |
9188 |
17422.4 |
17366.8 |
| Images |
473.69 |
477.44 |
477.28 |
477.84 |
477.36 |
477.46 |
645.44 |
642.507 |
| Cached Bitmaps |
1818.7 |
1819.1 |
1818.9 |
1819.8 |
1819.6 |
1819.6 |
5538.0 |
5618.18 |
| Strings |
221405 |
221650 |
213685 |
219304 |
215318 |
219467 |
208827 |
206985 |
As you can see, Xabre is losing in some tests and leads in others. Of course, it's not quite honest to compare the 2D core of Xabre400 with the powerful GeForce4, but it's the reality. Frankly, we expected things about it to be even worse, but Xabre is beginning to justify its hopes, although it is still early to make conclusions - the best is still to be seen.
Direct3D
In Direct3D there is no better tool like 3DMark2001SE benchmark to measure the video card clock speed. We wanted to test Xabre in CodeCreatures Bencmark Pro where even GeForce4 Ti4600 sucks, but the benchmark failed to run with the SiS gear. If such problems arise, Xabre is not fully compatible with DirectX 8. 1 then. Well, let's make an allowance for the raw state of drivers and get round to analyzing the 3DMark2001 results. We first start with the official benchmark scores for these video cards.
| 3DMark2001SE |
| Video card / Test |
Xabre200 |
Xabre400 |
ECS AG400 |
Gigabyte SP64D-H |
Vinix VX-3340 |
PowerColor EvilXabre400 |
GeForce4 MX440 |
GeForce4 Ti4200 |
| 3DMark Result |
5680 |
6983 |
6722 |
6823 |
6784 |
6760 |
5520 |
8871 |
| Game1 Low Detail |
95.7 |
132.9 |
126.5 |
126.9 |
126.7 |
125.0 |
95.5 |
139.1 |
| Game1 High Detail |
42.1 |
43.7 |
43.2 |
43.5 |
43.3 |
43.5 |
43.8 |
50.0 |
| Game2 Low Detail |
81.3 |
110.7 |
100.5 |
106.6 |
103.1 |
101.3 |
88.1 |
147.3 |
| Game2 High Detail |
48.0 |
53.9 |
53.7 |
54.8 |
54.2 |
54.2 |
46.0 |
90.3 |
| Game3 Low Detail |
90.1 |
108.7 |
106.2 |
106.0 |
106.3 |
104.2 |
91.6 |
127.1 |
| Game3 High Detail |
43.6 |
51.4 |
50.6 |
51.0 |
50.9 |
49.8 |
48.6 |
60.2 |
| Game4 Nature |
16.8 |
23.9 |
22.1 |
22.2 |
22.2 |
22.1 |
Not Supported |
36.2 |
| FillRate (Singletexturing) |
322.0 |
488. |
451.0 |
450.5 |
450.7 |
450.7 |
394.7 |
758.2 |
| FillRate (Multitextuing) |
1202.4 |
1806.2 |
1665.1 |
1660.3 |
1666.3 |
1655.9 |
507.6 |
1774.0 |
| High Polygon Count (1 light) |
17.8 |
27.7 |
25.9 |
26.0 |
26.0 |
25.7 |
26.3 |
41.3 |
| High Polygon Count (8 lights) |
5.2 |
8.4 |
7.8 |
7.8 |
7.8 |
7.8 |
6.5 |
10.4 |
| Bump Mapping Environment |
30.9 |
46.4 |
43.0 |
43.1 |
43.1 |
43.1 |
Not Supported |
138.6 |
| Bump Mapping DOT3 |
69.7 |
96.1 |
87.8 |
90.1 |
87.9 |
91.2 |
60.5 |
106.9 |
| Vertex Shader Speed |
44.6 |
45.2 |
46.4 |
46.2 |
46.3 |
45.4 |
48.6 |
85 |
| Pixel Shader Speed |
20.1 |
30.7 |
28.6 |
28.6 |
28.6 |
29.0 |
Not Supported |
96.9 |
| Advanced Pixel Shader |
13.3 |
19.7 |
18.2 |
18.5 |
18.3 |
18.5 |
Not Supported |
72.3 |
| Point Sprites Speed |
5.5 |
8.0 |
7.5 |
7.5 |
7.5 |
7.6 |
9.1 |
20.5 |
As is seen from the results, Xabre performed no worse than GeForce4 MX440 at theoretical tests, and at fillrates with multitexturing it scores even better than GeForce4 Ti4200. At gaming benchmarks, Xabre is superior to MX440, but we'll talk about that below. In the final scores, Xabre easily wins over MX440, albeit still far behind Ti4200.
Now move on to the gaming performance tests. In the tests, the video cards ran at the rated and overclocked modes.
It's amazing to see that Xabre video cards were almost on par with GeForce4 MX440. Sure, Xabre wins. By the way, Xabre has a specific trait. When overclocked, the video cards exhibit lower speeds than observed at the rated frequencies. This was seen for Gigabyte SP64D-H at high resolutions. No mistake here - indeed, the card did show lower speeds than at the rated frequencies. And note - only at high resolutions. By the way, it's seen from the very first tests that the 128-Mb card has advantages over PowerColor at high resolutions.
Game 2 is an easier test and here we see how hard it is for competitors to reach the scores achievedby the GeForce4 Ti4200 core. If we disregard the 1600x1200 resolution, Xabre defeats GeForce4 MX440 at other resolutions, and the 128-Mb version of Xabre400 performed on par even at high resolutions!
The trend stays: while GeForce4 Ti4200 is far ahead of the pack, the Xabre cards press on GeForce4 MX440. There is not much of a noticeable advantage, but it is explicit and stays at all resolutions except 1280x1024, where the low-end GeForce is second only to overclocked Xabre cards. In this test, the Vinix board made quite a good showing, although it is losing to the PowerColor stuff.
GeForce4 MX sucks in Game4, but Xabre is alive, which shows its advantage. No use comparing a video card equipped with software-driven vertex shaders to the advanced core of CeForce4 Ti having two shader blocks running in parallel. In this test, PowerColor acts on behalf of the whole Xabre line and takes all the hard job on itself. This is especially seen at the resolution 1600x1200. I worked that test through twice at the EvilXabre400 - no mistake here: there is a three-fold speed increase compared to 64-Mb video cards.
Dungeon Siege
Since the CodeCreatures benchmark failed to run at Xabre, we tested it on the Dungeon Siege game. The procedure of gauging the speed was quite simple - you simply switch on the FPS counter at the same scene in the game and register its average value. Of course, that was not Timedemo, but since the results were correct for equal conditions, the results can be regarded as valid. More important is that Dungeon Siege is a real application, not a synthetic benchmark. The resolution used was 1024x768, 32-bit color depth, with all the image settings set to maximum quality.
GeForce4 Ti4200 is the leader in the test as usual, but Xabre failed to catch up with the low-end GeForce4 MX440. Xabre cards shared the scores and gave in the performance crown to the PowerColor counterpart. Amazingly, overclocking gave little use to the ECS and Gigabyte boards.
OpenGL
To test the OpenGL performance, we used Quake III Arena. All the settings were set to maximum quality, and the test was run at the 32-bit color depth.
Real cool speeds start from several hundreds of FPS. The gap between GeForce4 Ti4200 here is not so wide as before. MX440 wins at the lowest resolution only. In all the other ways, the trend persists: a sure leadership of the reference card at lower resolutions, with the Gigabyte video card at middle resolutions, and PowerColor taking a lead at high resolutions.
Villagemark
The Xabre specifications lack any information on the video memory access optimization. So the only way to find out about Xabre operation at high Overdraw rates is starting up STMicro's VillageMark benchmark. If the video card handles all hidden surfaces, it won't withstand speed tests in either Direct3D or OpenGL.
As is seen from the results, Xabre400 is optimized for handling such scenes and applies some technique for cutting off invisible surfaces. And at Direct3D it does that even better than GeForce4 Ti4200. MX440 sucks!
 |
Top Stories: |
 |
 |
 |
MoBo:


|  |
 |
 |
VGA Card:


|
 |
 |
 |
CPU & Memory:

|
|