ABit Siluro GF4 Ti4200
Overclocking
Expectedly, overclocking the OTES video card has given absolutely predictable results which generally justify using such cooling system, It's worth mentioning that overclocking is a masterly painstaking job requiring a lot of patience and efforts. We succeeded in overclocking the card to 315 MHz on the chip and to 300 MHz (600 MHz DDR) on the memory. It was not possible to push the memory clock speed up to over 600 MHz, - if we add even 1 MHz more, then we get artifacts specific to excessive memory overclocking. This is no surprise, since, as we noted above, the memory cooling was not organized at all. It's more interesting for the case with the chip. The GPU easily overclocked to 300 MHz and thus overstepped the threshold of GeForceTi4600 chip speed, moreover the GPU was running stably at 315 MHz. This is approximately the speed to which an average Ti4600 can be overclocked.
Between 315 MHz and 330 MHz(!) all 3D tests ran flawlessly, but artifacts we observed, which is natural because the higher the speed the more unstable is the functioning of programs. As per our observations, there is no dependence between the overclocking of GPU and memory. That is, no matter what speed the GPU is running at - the memory can be always overclocked to 600 MHz, which is the stable factual limit of overclocking achieved on our specimen of the board, at the most. Repeating again that the board's factory settings were overclocked to the Ti4400 level by default, and the card was running very stably even with the lowered turbine rotational speed in the cooling system.
Tests
To test the performance of ABit GeForce 4 Ti4200 with the OTES cooling system we assembled these two test beds:
Test bed No 1 (Hi-End system):
- Processor Intel Pentium 4 2.8 GHz;
- Motherboard Intel 850EMV (i850E + 512 Mb RDRAM PC800);
- Sound card Creative Sound Blaster Audigy;
- HDD Seagate Barracuda ATA IV 7200 rpm.
Test bed No 2 (Mid-Range system):
- AMD Athlon XP 1600+ processor
- Motherboard EPOX 8K3A (the KT333 chipset, memory 256 Mb DDR PC2100);
- Integrated audio codec;
- HDD - IBM IC35AVVA07 7200 rpm.
We tested the OTES performance in the following benchmarks and applications:
- Return to Castle Wolfenstein - Checkpoint demo (OpenGL, Quake 3 core);
- Unreal Tournament 2003 Demo (Direct3D stress test benchmark);
- DooM III E3 Demo (extremely resource-hungry overkill benchmark with shaders, bump-mapping, OpenGL and lots of other stuff to be topical by the end of 2003);
- Codecreatures demo (complex DirectX 8 benchmark);
- 3DMark 2001SE (synthetic Direct3D benchmark)
In the tests, we produced the following results:
1. Return to Castle Wolfenstein
Upon a closer look at the two produced graphs we can see that at resolutions up to 1280x1024 the board is not falling behind GeForce 4 Ti4600. Considering the much lower price this is a really good news.
It's curious to note that with a weaker processor the board consistently lags behind Ti4600, and at lower resolutions the result is practically identical to that for GeForce 3. We can safely assert that first that the Ti4200 with OTES is meant for resolutions 1024x768 onwards or for games with FSAA disabled. Now look at the processor dependence of the card by the example of RTCW:
Note that the difference in performance is essential right up to the maximum resolutions. Return to Castle Wolfenstein is in more extent a processor dependent game, and with Athlon 1600+ everything is restricted by the processor's low performance. If we install a processor far more powerful than Pentium 4 2.8 GHz, the results at low resolutions should be much higher. Reminding you that all the measurements were carried out at the maximum quality and 32-bit color. Also, in view of this processor dependence we can assert that for those who have low-power processors it does make sense to use FSAA. In using it the bottleneck will be the accelerator itself, and the performance difference between Pentium 4 2.8 and Athlon 1600+ will be minimized. Now, on to Unreal Tournament 2003 benchmarks.
2. Unreal Tournament 2003
For the tests we used an official demo version of the game. The full version of UT2003 that sells on the Russian market has some issues with testing. So far they haven't been solved yet and we'll be using the demo.
With Pentium 4 2.8 GHz the situation is similar to that observed for RTCW. The differences at low and medium speeds are minimal, and they come into play only at 1280x1024 and 1600x1200. To be fair, it's worth noting that the difference is that big - the 7 FPS difference at the maximum resolution is not worth the price difference.
On a less powerful system the trend stays the same. Now let's analyze the processor dependence:
Already at 1280x1024 the results become identical. The owners of low-power processor still can resort to FSAA which in most cases will be free at low resolutions. UT2003 is definitely a less processor dependent game and puts a much heavier load on the GPU part than in RTCW. On to the Codescreatures benchmark.
3. DooM III E3 Demo
Recently, CDs with DooM III Lite have hit the retail shelves in Moscow. We strongly recommend not to buy this "game". In fact, that version of DooM III was stolen from E3 and is absolutely unsuitable for neither the multiplayer nor single-player gaming. The version 0.02 demonstrated at the expo was meant for merely displaying the potentials of the core developed by John Carmack's team, and no more. Even as a benchmark it can be used with a great reserve. The testing techniques as well as the moral and ethical issues will be covered elsewhere in a separate review.
Of course, one should see Doom III in play at least once to understand what first-person shooters will look like in 2004 and - take it on trust - we'll certainly need very fast PCs to handle them. Frankly, even with Pentium 4 2.8 GHz and overclocked GeForce 4 Ti4600 the demo is running at a snail's pace. It's useless running Doom3 on Athlon 1600 or something less powerful than GeForce 3 - at best, the maximum speed won't be faster than 10 FPS, and at worst - less than 1 FPS. The alpha-version is very critical to the RAM capacity. 512 MB is the minimum amount necessary for running tests. The processor must be at least 2 GHz. What's in the upshot? We tried only one test bed of the two. But the graphics... Wow!!.. it's really fantastic, something unseen before, so let's talk about the figures:
Basically, we see a quite standard thing - GeForce 4 Ti4600 with 128 MB of memory is ahead of the pack in all the tests, including those at the minimum resolution. The game is very demanding to the gaming hardware and we can safely assert that in our case it was not the processor (P4 2.8 GHz) but the graphic accelerator that had restrictions in all the resolutions.
Tests in high resolutions exhibit a similar trend. The gap between 4200/64 and 4600/128 is not great. Playing Doom III at resolutions higher than 800x600 even on top-end configurations makes no sense. For comparisons, we recorded the second demo. The first (demo002.demo) runs in the first level, with another (demo006.demo) in the second. The alpha-version of Doom III is made up of three levels, with the complexity and geometric richness gradually rising from level to level. The first level is very easy - no complex algorithms are used for computing the object physics and not so much of the architecture is involved. The most demanding is the third level, and currently we haven't yet prepared an adequate demo to be used for testing. Now on to demo006.demo:
On the whole, the trend is the same but for the slightly reduced gap between 4200 and 4600. Presumably, for Doom III the shaders handling will be more important rather than the raw performance of the video card. Soon we'll present an additional review analyzing the Doom III phenomenon in greater detail. In the end, here are the test results produced with demo006.demo at high resolutions:
As you see, the results are predictable and haven't brought any surprises. The results on the graph are self-evident without any comments. If you remember, Quake 3 was released in the times when Voodoo3 was just about to be batch-produced but we are still using it and derived games as benchmarks up till now. Playing Doom 3 and derived games will be comfortable no earlier than late 2003 or even in mid-2004 when graphic accelerators acquire 256 or more megabytes of video memory with the speeds and potentials going several generations far ahead of GeForce 4.
Codecreatures
Codescreatures on the agenda. As the tests have shown, for this game the larger video memory capacity is more important than the raw chip speed and VRAM:
From the graph we see that the 128 MB GeForce 4 shot ahead leaving the pack far behind, with the 64 MB OTES in some tests is very close to GeForce 3.
They are more closer in tests carried out with Athlon 1600+. All the advantages of Ti4200 vs. GeForce 3 have come to naught in here.
In the upshot, the total results of benchmarks don't look that impressive. The card's scores are almost 1.5 times less than for Ti4600.
3DMark 2001SE
This is one more standard benchmark which we use for examining the performance of 3D accelerators. This time, we are not bringing in the scores for each test segment. The general trend is vividly illustrated by the total results:
We are not citing separate graphs for each tests since their dynamics is generally identical to the total results.
 |
Top Stories: |
 |
 |
|
MoBo:
| |
|
|
VGA Card:
|
|
|
|
CPU & Memory:
|
|
