As we have expected, after a short lull new graphics cards started springing up like mushrooms after a good rain.Many permanent Internet haunters have already seen numerous reviews on Riva TNT2-based cards by various manufacturers.
Authoritative testers have already checked super-cool graphics cards made by Diamond Multimedia, and even Reference Cards from nVidia, however, these cards are not overloaded with any additional services.
About a month ago ASUSTeK Computer announced a new graphics cards family ASUS V3800 based on Riva TNT2, which has a TV-in/out, supports LCD-monitors and goes together with stereoscopic glasses.
By a lucky chance we managed to get the fullest package set of all possible (the only thing it lacks is LCD-support).
To begin with we offer you a comparatively full specs list of the long-awaited Riva TNT2 chipset:
|Nvidia Riva TNT||ATI Rage128||Nvidia Riva TNT2||3dfx Voodoo3|
|API support||Direct3D, OpenGL||Direct3D, OpenGL||Direct3D, OpenGL||Direct3D, Glide/OpenGL|
|Core frequency, MHz||90||100||125-183||143-183|
|Memory frequency, MHz||110||110||125-183||143-183|
|Max supported memory, MB||16||32||32||16|
|3D Truecolor (32bit) rendering||yes||yes||yes||no|
|3D maximal resolution|
|- Highcolor (16bit)||1600x1200||1600x1200||1600x1200||1600x1200|
|- Truecolor (32bit)||1600x1200||1600x1200||1600x1200||no|
|The number of rendering pipelines||2||2||2||2|
|Fillrate, mln pixels per second||180||200||300||143|
|Processing power, mln polygones per second||6||4||8||6|
|Larger (1024x1024) texture support||yes||yes||yes||no|
|- AGP 2x||yes||yes||yes||no|
|- AGP 4x||no||no||yes||no|
|Z buffer accuracy||24||32||24||16|
|Per pixel MIP-mapping||yes||yes||yes||yes|
|- Single pass||no||yes||yes||yes|
|OpenGL support||ICD||ICD||ICD||Miniport (wrapper)|
Besides, it should be pointed out that the stated core and memory frequencies are conventional because nVidia announced the launching of several Riva TNT2 chipset types, which differ only by their frequencies (varying from 125 to 183MHz). The sample considered in the present review operates at 140MHz core frequency and 150MHz memory frequency.
Further on we will speak about V3800 graphics card. Below you can see two pictures of the card and of the memory chip.
The graphics card possesses 32MB SGRAM 7 ns memory, AGP 4x, chip cooler (rather known among ASUS V3400TNT owners), BIOS microchip, TV-in/out, stereo-glasses support and some other corresponding logic.
The set seems too rich for a single graphics card, which resulted into a great number of connectors on metal side.
The connectors location (from top to bottom):
- Stereo-glasses in-connector;
- TV-in (there is only one input port for S-Video but S-Video converter is also included);
- Monitor connector;
- S-Video TV-out;
This graphics card is almost as packed as a combine. It is supplied in a box together with the following items:
- Stereo-glasses VR100;
- CD-disk with drivers;
- A set of TV-in/out connecting cables.
And now about the benchmarks. Before going over to testing system description we should mention the fact that some mainboards appeared unprepared for AGP 4x graphics cards. For instance, Chaintech 6BTM (440BX) (BIOS dated back to 25 February, 1999) didn't manage to function properly with ASUS V3800. And as for ASUS P2B family, everything worked perfectly well. That's why all benchmarks were carried out on ASUS P2B-B.
The testing system was configured in the following way:
- ASUS P2B-B (440BX) mainboard;
- Intel Pentium II 450 and Intel Pentium III 450 CPUs;
- 128 MB PC100 system memory;
- Nokia 447Xav (17") monitor;
- Windows 98.
Every benchmark starts with the installation. Unfortunately, we have to admit that the software on the enclosed CD-disk turned out a bit unfinished. For example, after inserting the disk into the CD-ROM drive you get the initial menu program. However, all it does is just to demonstrate what we can do (install the drivers or Live3800), and then nothing else works. That's why we had to install the drivers manually through an inf-file. Further on everything went on smoothly and we managed to install the drivers successfully. As a result the display properties panel acquired several additional pages.
Among the common properties typical of Riva TNT2 (such as the number of MIP-levels, filtering types, Vsync, etc.) there also appeared some newer ones. First of all, this is enabling stereo regime, i.e. enabling stereoscopic glasses. This regime allows to regulate optical settings, which serve to adapt the glasses for each single case, so that it could be comfortable and convenient for the end-user to use them. Besides, one of these pages contains some info about this particular graphics card and its drivers.
Another important thing before we dwell on the benchmark's results is the best choice of some other graphics cards, which will be used for comparison. As for 2D-graphics, it seems most reasonable to compare our sample with all the latest novelties. As for 3D-graphics here we will first of all dwell on comparison with Riva TNT (Creative Graphics Blaster Riva TNT), then goes the former leader in 32-bit color - a card based on ATI Rage 128 (ATI Rage Fury), and finally the newcomer we have recently tested - 3dfx Voodoo3 2000 (of course, here it will be compared only in terms of 16-bit color in 3D).
So, let's get started! As usual first of all we are going to discuss 2D-graphics. The charts below show the speed values achieved while running WinBench99:
The results are very illustrative and speak for themselves: nothing has changed compared to Riva TNT. And a close visual study of the graphical image proves it. at the resolution equal to 1024x768 everything is perfect, and as it comes to 1024x1024 the picture turns a bit washed out. This observation makes us conclude that the 2D-regime of our graphics card is intended purely for gaming purposes, it suits mostly for those with 15" and 17" monitors (and if someone prefers 1024x768 resolution on his 19" monitor then this card will suit him as well).
Make your stakes, ladies and gentlemen! Making stakes means excitement and thrill, and who isn't excited about 3D-graphics quality and speed now? So, now we will talk about 3D-graphics for a while.
We have to mention right away that having carried out the first tests with ASUSTeK Computer drivers we noticed that there is hardly any gain in speed, any tangible gain in speed we've been expecting to see. And having installed nVidia 1.20 Reference drivers (Detonator) we could observe a rather evident speed-up. That's why these drivers were chosen for all further benchmarks (with the exception of stereo-glasses, of course). Besides, this graphics card BIOS version was 2.05.04.
And now a few remarks concerning overclocking. Unfortunately, Riva TNT2 chipset can't boast any utility, which allows to change video processor core frequency and memory frequency. By now it is only PowerStrip (2.41.04) that supports Riva TNT2. However, it turns out absolutely impossible to set the chipset frequency over 140MHz (there were even a couple of attempts to adjust pstrip.cfg though all in vain). As for the memory frequency PowerStrip allows to change it but as soon as it exceeds the nominal frequency by at least 7MHz (157MHz, for instance) the graphics card loses its stability and starts stumbling. That's why all the nominal frequencies were left equal to 150MHz memory and 140MHz core.
First we performed the tests with Intel Pentium II 450 CPU at various resolutions for 16-bit and 32-bit color. Here the following benchmarks were utilized:
- 3DMark 99 MAX (DirectX);
- Incoming (DirectX);
- Quake2 massive1.dm2 demo (OpenGL).
To estimate the graphics card performance in 32-bit color we also post the chart showing the performance decrease while shifting from 16- to 32-bit color depth in 3D-graphics. The results were obtained from the game Incoming. The higher is the value the sharper is the performance decrease while the color depth is getting greater.
Then we ran 3DMark 99 MAX with resolution 1024x768 and 16-bit color depth for different Intel Pentium II processors:450, 300 and 233MHz. The results are displayed below:
What conclusion can be drawn from these figures? Well, undoubtedly Riva TNT2 based graphics card retains the leadership in all Direct3D benchmarks. In OpenGL (Quake2) it manages to be the first only with 32-bit color, and with 16-bit color depth the victory is gained by Voodoo3 2000 (note that it takes place at the almost equal graphics processor and memory frequency values!). However, we suppose that additional drivers improvement will help reduce this gap significantly.
But what is really pleasing it's the fact that the speed drop, which accompanies the transfer to 32-bit color depth, was significantly reduced. However, Riva TNT2 can't catch up with ATI Rage 128, which managed to achieve impressive results in this test, but nevertheless the mere fact gladdens your soul.
And again we are inclined to repeat the same conclusion as for Voodoo3 2000: at lower resolutions the performance gain is either absolutely absent or very insignificant if compared to other graphics cards. However, in the scalability tests (showing the influence of CPU performance on graphics cards performance) Riva TNT2 proved to be the fastest at the resolution 1024x768 for various processor frequencies (except 450MHz). As the CPU frequency went down the performance decreased not so rapidly as by Riva's main rivals, which actually lets not very cool processor owners do themselves proud.
It can seem quite natural that almost equal results shown by absolutely different cards at 640x480 & 800x600 make you suppose that the benchmark aims at different things. It looks as if it were not the graphics card speed, which is measured, but a CPU speed, and hence an evident conclusion can be easily drawn: the processor has totally used up its reserves while the graphics card still has a certain potential.
Unfortunately, the today's market can boast no faster CPUs, which can make use of all the graphics card potential resources. And that's why we may definitely conclude that buying such a cool 3D accelerator makes sense only if the working resolution is at least 1024x768 (though the 2D feature of this particular card doesn't allow us to recommend higher regimes). Another essential requirement is an at least 15-17" monitor and in case you have a 15"-monitor it should be of very high quality. It will for sure be a great pity if you spend the whole bunch of money for a super-graphics-card and get the same speed at lower resolutions as the 3D-accelerators already available in the market.
Well, it looks as if we'll have to make similar conclusions pretty often while testing a new super graphics accelerator.
Now let's separately mention Riva TNT2 with Intel Pentium III 450. You should bear in mind that there aren't so many real applications, which take advantage of Pentium III SSE technology. That is why the performance increase of the graphics card can be seen only in 3DMark 99 MAX:
|3DMark 99 MAX||4611||4604||4015||2602||4611||4171||3083||1708|
Frankly speaking, it is a rather good increase. The only thing that deserves our complains is that hardware engineers are working too fast and the software developers are simply falling behind with the corresponding soft-support for all the novelties and improvements.
The modern technologies keep on working wonders. The common reality turns into a virtual world, which can absorb you up to your neck and won't let you out. This metamorphosis has become so easily achievable due to powerful graphics cards, which get more and more skillful in simulating the 3D-world. This romantic prelude signals that it's time for 3D-graphics quality discussion. In our case, i.e. by Riva TNT2, it seems to have remained practically on the same level compared to Riva TNT, that's why we made up our mind to save you time and to leave out all the numerous screenshots proving this point since they are almost 100%-alike. Those of you, who have the Riva TNT-based graphics cards, know what we are talking about, and the rest can look through some reviews on Riva TNT with screenshots showing the quality of this card performance.
And as for 3DMark 99 MAX we managed to notice certain concrete highs and lows of Riva TNT2 chipset. One of the screenshots shows a slight blurring on the latticed floor, which is absent on the reference image:
nVidia Riva TNT2
The same effect also occurred by Riva TNT.
The results of subpixel correction quality benchmark appeared to be quite surprising. Riva TNT2 results were much closer to the reference than those of Riva TNT despite the fact that both graphics cards were utilizing one the same drivers!
nVidia Riva TNT
nVidia Riva TNT2
However, a certain improvement is still evident.
So, what will the conclusion sound like as far as Riva TNT2 3D-graphics quality is concerned? In fact, it remained the same as by Riva TNT. As for us, we are pretty satisfied with Riva TNT quality, and consider this graphics card to be the best of all the other gaming cards. That's why Riva TNT2 has also won our heart.
And if we also take into account a significant speed increase, the Riva TNT2-based graphics card will undoubtedly occupy the first place among all known video-accelerators (though there some compulsory requirements: a fast CPU and a large monitor).
And now we'll pass over to stereo-glasses VR100, which go together with the graphics card ASUS V3800. As it was predicted the glasses are based on the H3D model and are produced according to Metabyte technology. The principal difference between VR100 and Wicked3D eyeSCREAM is a wire connection to the graphics card. This is how these stereoscopic glasses look:
Well, as you can see VR100 is far too massive compared to eyeSCREAM. Moreover, it lacks a very convenient feature of adjusting the size of the glasses for one's particular eyes and face (however, it is not a grave drawback since VR100 glasses are provided with larger lenses, which should fit most people). But as for the bridge it seems rather unpractical. The soft pad bridge, which theoretically should lean against the nose bridge, is located too deeply, so it is only a piece of hard plastic that rubs your poor nose. And sometimes the eyeglasses intended to slip down from my nose at all.
And now a few words about their practical value. You enable the stereo-glasses by putting a tick opposite the corresponding item on one of the settings pages. To our great disappointment we failed to find any tests, which could allow us to set the glasses for a particular user because as we know different people perceive the stereo-effect in different ways. It's all very specific, that's why a live setting feature could be of great use (as with eyeSCREAM, for instance). When the stereo is on, the monitor refresh rate doubles if the resolutions are cut down (for example, 800x600 is cut down to 800x300). So, you have to bear it in mind and if you set the refresh rate equal to 75Hz while the resolution is 800x600 your monitor (if it's not very cool) may not work properly because the system will require 150Hz. However, we would like to point out that stereo regime doesn't support any frequencies exceeding 75Hz. To illustrate the following statement we can turn to eyeSCREAM, which graphics card frequency settings are closely connected with enabling the stereo regime and hence no unpredictable combinations may occur.
Speaking about drawbacks, which are most likely to be connected with imperfect drivers, we may mention total absence of stereo in OpenGL as well as in Direct3D when the resolution is set to 1024x768 and up.
Visually the stereo provided by VR100 is absolutely identical to what we can get from Wicked3D eyeSCREAM.
And finally a few words about the TV-in/out. The main functions and the provided quality of this feature are very much like those of ASUS V3400TNT/TV. TV-out is still operating only at 800x600 or lower resolutions after restarting the system if it is connected to the TV. TV-in is controlled by Live3800 utility, which hardly differs from the previous version Live3400. And that seems to be all here.
By 20 April this graphics card supplied within the above described package costs about $260. Taking into account that it goes together with the stereoscopic glasses the card itself may cost about $200. That's why this graphics card with super-performance may be an excellent choice for those of you who failed to buy a good gaming graphics card (the first Riva TNT-based cards also cost about $200). However, this corresponds only to those who have really fast CPUs, at least a Pentium II (Celeron) 400. Riva TNT2-based graphics card combines both: high speed and good quality.But as for software ASUS drivers still need improvement. Besides, similar devices are about to appear rather soon.Their main peculiarity will deal with operating at higher frequencies compared to the ones mentioned above. So, what we'd recommend you is to have patience and wait for another Riva TNT2 Ultra based graphics board to come out.