New 3D Chips – Banshee, G200, RIVA TNT And Savage3D – THG.RU

Introduction

It’s now a few weeks after I published the preview of 3Dfx’s Voodoo Banshee, NVIDIA’s RIVA TNT and S3’s Svage3D. Since then I received new and better hardware of two of those as well as a final Millennium G200 from Matrox. Whilst Matrox is already shipping the G200 now and Hercules is planning on shipping their Savage3D card this week, Banshee will stay off the shelves for about another 4 weeks longer. It will take even longer for the RIVA TNT, let alone 3Dlabs’ Permedia 3 and Rendition’s Verite 3000 series. For now the G200, Banshee and Savage3D are the three new 3D chips that are closest to its final revision so that it makes sense to publish a review rather than a preview of them now.

No News In The 3D Benchmarking Scene

Whilst there has been a lot of motion in the 3D hardware scene in the last few months, the 3D software scene was mainly blessed with Microsoft’s recently released DirectX6 and Epic’s Unreal game. The benchmarking scene looks even worse, still only leaving the good old game benchmarks to the poor reviewer, maybe with the addition of Ritual’s new Sin demo. We have learned that Ziff Davis’ 3D Winbench 98 is a great CPU benchmark, but not very useful for 3D card benchmarking. In lack of own ideas they came out with a thing called `Game Gauge’, a benchmark that assembles all the game benchmarks that Tom’s Hardware Guide was using for the last 6 months, keeping us stuck in the same situation as before. I would like to thank Ziff Davis for the appreciation, showing that consequences were taken after my article about 3D Winbench 98. Those consequences were again based on research from my website, which makes me feel very glad.

The problem we currently have in regards of 3D benchmarking is that there aren’t any games that can take real advantage of DirectX 6 yet. DirectX 6 is for the first time offering a very useful geometry engine to the game developers, so that they don’t need to use stuff like e.g. Glide or their own engine anymore. DirectX 6 also offers a whole lot of new 3D features, which are not yet used in any new game, particularly not in any game that would be suitable for benchmarking. This means that the cards cannot be tested under conditions they may meet in a few months from now (August 98), which should be considered as a serious pity.

Texture Compression

S3’s Savage3D is the first 3D chip that is using texture compression. This compression technique, called S3TC, was even implemented into DirectX 6, so that future games can take advantage of it. S3 supplied a registry switch that forces even current games into running with texture compression, but this is a very unpleasant solution, because game developers cannot control which textures will be compressed and which won’t. Texture compression is a procedure that incorporates a texture quality loss, which may not be welcome in several situations. I have dedicated a complete section to this issue, including several voices from the industry that detest this `auto-compression’ feature.

Texture compression will become a pretty common thing for many 3D-chip developers and we can soon expect a lot of new texture compression techniques besides S3TC. The idea behind it is to reduce the memory bandwidth used for texture transfers, which makes the majority of the data transfers between the chip and memory. Future high resolution textures and high 3D screen resolutions will increase the need of memory bandwidths far beyond the numbers we currently know, so that many developers consider texture compression as a very good way around the bandwidth problem. The downside of texture compression is the likely quality loss and the time compression of static textures takes before the game starts as well as the CPU power needed to compress dynamic textures on the fly. I will get into that when talking about Savage3D.

2D Performance Close To Maximum

The three tested chips are all combined 2D/3D solutions, thus including a 2D core in addition to the 3D core. 2D graphics performance has become more and more less interesting, since all major chip manufacturers are getting closer and closer to the theoretical maximum in 2D performance. The three tested chips are no exception to this rule, their 2D performance is almost equally high.

3Dfx Voodoo Banshee

3Dfx’ first combined 2D/3D solution comes with typical 3Dfx-features. It supports the well known and widely used Glide engine, 3Dfx’ proprietary 3D engine that is very easy to program and using only pretty little CPU power. This makes Banshee the only 2D/3D chip that can take advantage of all the game titles using Glide. Banshee comes in a PCI as well as AGP version, but the AGP version is only able to take advantage of AGP’s faster DMA bandwidth. Banshee is not at all able to use any of the real AGP features like AGP texturing, AGP 1x or 2x. This can be a problem when running games with particularly large textures especially at high resolutions, because the local memory won’t be able to contain all those textures so that they have to be swapped from main memory via DMA transfers, which isn’t by far as fast as AGP texturing. So far have I have yet to see any games that do that, which gives Banshee’s lack of this feature an only theoretical significance.

The 3D core is pretty close to what we know from Voodoo2, however with two major differences. Banshee has only got one texture unit as opposed to two texture units in Voodoo2. This means a disadvantage for Banshee in games that are using a feature called `multi texturing’. Unreal and Quake 2 are two examples, where over the texture there’s an additional lighting map, which has to be processed as well. Banshee needs to do two passes for those two jobs, Voodoo2 can do both at the same time. This means that Unreal and Quake 2 are running faster on a Voodoo2. On the other hand is Banshee running at a higher core clock than Voodoo2, resulting in a higher fill rate, which makes games that don’t use `multi-texturing’ run faster on Banshee than on a single Voodoo2.

The 3D quality of Banshee is identical to Voodoo2, meaning that it doesn’t have a more than 16 bit deep Z-buffer and cannot do real 32-bit color rendering either. Other chips are better than that today, but Voodoo2 and Banshee are still looking pretty good.

Matrox G200

The G200 from Matrox was developed as a pretty fast 3D chip that offers very good 3D image quality. It comes with an excellent 2D core, something that we are used from Matrox. The 3D core is not as fast as Voodoo2, but in most cases it’s offering a significantly better image quality. 32-bit color rendering and a 32 bit Z-buffer are the ones responsible for that besides a lot of other nice features. Currently the G200 boards Millennium G200 and Mystique G200 are shipping with a pretty well performing D3D wrapper for OpenGL games like Quake, Quake 2 or Sin, but Matrox will hopefully soon supply an OpenGL ICD that is supposed to further improve OpenGL performance. The Millennium G200 is a very attractive solution for high end users that want to have a perfect 2D desktop screen at highest resolutions whilst having high quality expectations at 3D games. However, of the three chips tested here it scores the lowest frame rate numbers, which are still good enough for high quality game play in many cases.

NVIDIA RIVA TNT

When NVIDIA announced RIVA TNT at WINHEC this year, everybody was amazed about the huge numbers they were talking about. 250 Mtexels/s texel fill rate and 8 million triangles/s are numbers far beyond what any current 3D chip is able to provide. As time went by NVIDIA had to learn that those numbers will not be achieved this year anymore, because the huge 8 million transistors chip is getting too hot at 125 MHz clock as long as it’s produced in 0.35 micron technology. Thus the claims from WINHEC won’t be fulfilled before the shrunk to 0.25 micron has taken place, and nobody expects this to happen before 1999. Instead of this, NVIDIA is now talking of 180 Mtexels/s and 6 million triangles, due to a chip clock reduced down to 90 MHz.

The specs of TNT are still impressive enough though. TNT stands for `Twin Texel’ and is supposed to point out that TNT has got two rendering pipelines that can work in parallel, equivalent to the two texelfx2 units in the Voodoo2 chipset. Each pipeline can produce one pixel per clock. Games like Quake2 or Unreal and many future games are using more than just one texture map. Quake 2 does e.g. have a texture map and then a lighting map as well, which both have to be taken into consideration when drawing a pixel. One pipeline has to first render the pixel using the texture map, then another time using the lighting map. The same is valid for bump mapping, where there is a texture map and an environment map that has to be put on top of each other. Two parallel pipelines can do both stages at the same time, thus increasing rendering performance. Voodoo (only on Quantum3D boards) and Voodoo2 were the first main stream 3D chipsets that could do this parallel rendering, TNT is now the next chip that provides this important feature. ATI’s Rage128 and 3Dlabs’ Permedia3 will follow soon.

6 million triangles sound impressive as well, but since there’s no clear definition of triangle rates, it could be that this number is a bit on the high side. A high triangle rate is very important too nowadays, since only high polygon counts can make a scene look really detailed. E.g. a face of a player in Quake2 could look a lot better if it would consist of more polygons.

NVIDIA is also very proud on TNT’s 2D engine and my results will show you that they’ve got all reasons to be. Other features like DVD support and such are not quite that important for the retail market and count mainly for OEMs really. ATI and S3 are making a big deal out of the missing `motion compensation’ feature of TNT and Banshee, but this is only important to people who want to watch DVD on their computer monitor and who are using a weak CPU at the same time.

TNT was supposed to be the Voodoo2-killer in the first place, but it looks as if this will not quite be reached, since NVIDIA had to retract the originally claimed numbers. The claim that TNT will reach the performance of two Voodoo2 boards in SLI configuration is certainly more wishful thinking than anything else. Nevertheless does TNT have several serious benefits over Voodoo2 and even Voodoo2 SLI. First of all there is the 2x AGP interface, which can prove quite helpful for games that are using very large textures. Secondly can TNT display games at resolutions up to 1600×1200, Voodoo2 is limited to 800×600 and Voodoo2 SLI is limited to 1024×768. TNT has got a 24 bit Z-buffer that provides more accuracy than the 16 bit Z-buffer of Voodoo2 and TNT is able to do 32 bit color rendering … something that Voodoo2 is also not able to do.

What we should expect from TNT is very good 3D performance, close to Voodoo2 performance, excellent image quality, excellent 2D performance and quality (internal 250 MHz RAMDAC) and high resolution support in 3D. Most of those expectations will become satisfied….

S3 Savage3D

`Voodoo2 performance at S3 prices’ is the slogan for the Savage3D and after I first opposed to that opinion I’ve now got to seriously consider it. However, there still are quite a few unanswered questions. First of all it’s true, Savage3D is performing very well in the benchmarks. It surpasses Voodoo2 and scores pretty much the same as Banshee does. In games that use multi-texturing the Voodoo2 can show its muscles though, since the Savage3D doesn’t have a second texture unit as well as Banshee. All in all the benchmark results are very close to Banshee, sometimes a bit faster, sometimes a bit slower. From that point of view there is all reasons to congratulate S3 for their new chip.

However, there are a whole lot of issues that remain unsolved or need to be discussed still though.

S3 Savage3D – Quake 2

The bug I found in Quake 2 was a major thing to me as a Quake 2 player. Several maps in Quake 2 have some kind of window screens, where you can look through and see if your opponent is there on the other side of that glass screen. Well-known maps with that are e.g. `The Frag Pipe’ and `Sudden Death’. When you see your opponent in the other room, you can jump on a button and flood this room with lava, thus killing your opponent who loses a frag this way. The first two drivers I got from S3 wouldn’t let you look through the screen, the windows appeared as opaque white boxes with some textures on it. The latest driver I received partly solves this problem, but you’ve still got a hard time recognizing anything behind the glass.

Unfortunately, when rushing out the new driver, the S3 testing team overlooked a new bug, which makes the game even less playable now. You haven’t got a cross hair anymore. Now I may not be Thresh, but I think that most Q2 players will have a serious problem without a cross hair.

S3 will certainly be able to sort out this bug as well, but it’s another proof that the Savage3D drivers are not mature yet. I really wonder what driver Hercules is planning to ship with their Savage3D cards this week.

Last but not least there is a serious gamma problem with Savage3D in Quake 2. The light seems to flicker, depending on which direction you look at. Most of the time it’s horribly dark, but then suddenly everything gets brighter for a moment.

Download 640×480 BMP file (900kB)

No crosshair and hardly transparent screens – that’s Quake 2 with the Savage3D.

Download 640×480 BMP file (900kB)

Crosshair is where it belongs and you can look through the screen just fine – Quake 2 with the G200.

S3 Savage3D – Quake 2 Map `NEWS3′

This Quake 2 map was developed by S3 to show the advantage of texture compression used by Savage3D. It uses a huge amount of textures, so that only cards that can do at least AGP texturing are able to run it at a reasonable speed. This means that cards with 3Dfx chips cannot really be used very well with this map, Matrox’ G200 with its AGP 2x support is looking a lot better than Voodoo2 and Banshee, but Savage3D easily beats them all. The only problem with Savage 3D is that you need to enable the registry setting `AA’, standing for `auto-AGP’. That wouldn’t be too bad, but e.g. Turok runs slower with this setting enabled. I wonder how this issue is supposed to be solved. You cannot be asked to change registry settings for each game.

S3 Savage3D – AGP

S3 claims that Savage3D has full AGP 2x support. You know how highly I rate this feature, but if something is used for marketing I may be allowed to check it as well. Final Reality has a nice AGP benchmark built in, which lets you check the AGP transfer performance of an AGP board. Banshee naturally sucks at this feature, because it hasn’t got a real AGP implementation altogether. G200 with its AGP 2x support scores a fine 76 fps whilst Savage3D only scores 23 fps. This doesn’t look like a good AGP 2x implementation at all. After enabling the `AC’ or `auto compression’ registry setting, which forces all used textures to be compressed, the result quadruples to 96. This is not surprising because the textures get compressed 4:1, which is almost exactly the increase we are getting here. S3 told me that Savage3D has definitely got an AGP 2x implementation, but the results I’m seeing speak a different language. Intel’s performance tool IBASES shows exactly the same result, G200 scores more than 6 times as high as Savage3D.

S3 Savage3D – Sin

Ritual’s upcoming new game that’s based on the Quake 2 engine should normally run with the Q2 wrapper as well. However the first two drivers wouldn’t let me run Sin in OpenGL mode at all and the new driver I’ve got run’s Sin horribly slow for a short while before the game simply crashes. I may ask what Hercules is going to tell their customers who want to play Sin after they bought a Savage3D board next week.

S3 Savage3D – Chip Clock

When playing a game I came across several triangle drawing errors. This effect is well known to all Voodoo users who overclocked their cards. It seems as if the Savage3D chip with its 125 MHz clock frequency is close to being overclocked. This issue will most likely be solved in the future and the Hercules cards that are supposed to ship this week are only running at 110 MHz and thus won’t have this problem.

S3 Savage3D – Texture Compression and Benchmarking

Whilst all the current D3D games only run with texture compression when forced to do so by the `AC’ registry setting, Quake 2 is always using texture compression because the OpenGL/D3D wrapper enables it permanently. This leads to quite a few unusual situations. Before you are running a map for the first time, all the static textures of this map are compressed and stored onto the hard drive in the subdirectory `S3tex’. The compression of all those textures can take up to 5 minutes, particularly in case of big 64 player maps. You also have to be aware that quite a huge amount of hard drive space is needed for that. Once the map has been ran once, this compression doesn’t take place anymore and the map starts straight away. Now there aren’t only static textures in games but also so called dynamic textures, which are produced all of the time, depending where you are walking around in the sewer and what you are looking at. Those textures are compressed on the fly while you play the game. The effect of this is that a benchmark always gets faster after each run, because more and more of those dynamical textures are somewhere in the memory cache of the system. A benchmark runs always the same route through a sewer, so the dynamic textures are always the same. The problem is, that in actual game play you hardly ever walk the exact same route twice. Thus different new dynamic textures are produced all of the time. The result of this situation is that actual game play does never run as smoothly as the frame rates scored after several benchmark runs would make you expect. I have yet to still test this issue, but I had the feeling that in actual Q2 game play the card occurred a lot slower than the high frame rates would suggest. It is very difficult to quantify this though.

I would like to give an example on how the frame rate results change when running the demo several times: 64-65-69-71 fps in the 1^st, 2^nd, 3^rd and 4^th run of the mon2.dm2 demo in Quake 2 at 640×480. This does not happen with any other graphics card and is clearly explained by the dynamic textures issue.

S3 Savage3D – Questionable Registry Settings

I have never come across as many different registry settings as in the Savage3D driver. Some of them occur a bit strange and can impact benchmarking by a considerable amount. The number one topic is the `AC’ setting. It forces texture compression for any game that’s run. Texture compression can degrade the image quality, which is why game developer are strongly against this feature, as you can read in those 8 voices from the 3D game developing industry. Enabling `AC’ leads to better benchmark results of course and how many reviewers take the time to look at what the game benchmark looks like? Another setting is `AT’, which means `auto-mipmapping’. This feature is well known from NVIDIA’s RIVA 128 driver and widely disapproved. It causes strange artifacts and thus also degrades image quality. S3 states that they are not planning on enabling these features in any of their drivers, but one may wonder what those settings are there for in the first place. It gives an easy tool to OEMs for scoring very well in benchmarks, while degrading image quality. Last but not least is there a setting called `FD’, which improves the performance in Incoming. I haven’t got a clue what it does though.

I hope you can appreciate that there are two reasons for being so critical at the Savage3D. The one reason is that the first Savage3D boards are supposed to ship soon. This seems too early to me. The other reason is that a few claims of S3 don’t seem to be met. The AGP performance is below par and the issue with the `auto’ registry settings could be used as a tool to only reach high benchmark results. Last but not least is there a possible divergence between benchmark results and actual game play, also generated by texture compression.

Benchmark Setup

The benchmarks were run in the following configuration:

• Asus P2B motherboard w/ Intel 440BX chipset
  
  Adaptec 2940UW SCSI host adapter
  
  IBM DGVS 09U SCSI hard drive
  
  64 MB LGS PC-100 SDRAM DIMM
  
  Intel Pentium II 400 and Celeron 266 CPU
  
  Windows 98 operating system
  
  DirectX 6 final release

• 3Dfx Voodoo Banshee driver:
  
  4.10.01.0090-1.00
  
  “Flip on Vsync disabled”

• Matrox Millennium G200 driver:
  
  4.10.01.4140
  
  “Flip on Vsync disabled”

• NVIDIA RIVA TNT reference driver:
  
  4.10.01.0032
  
  “Flip on Vsync disabled”
  
  Chip clock 95 MHz
  
  Memory Clock 112.5 MHz
  
  Used RAM Type: SGRAM

• Savage3D driver:
  
  4.10.01.4000-6.00.20B
  
  Registry settings:
  
  AA on
  
  AC off
  
  AM off
  
  AT off
  
  VO off
  
  WE on
  
  WV off
  
  “Flip on Vsync disabled, automatic texture compression disabled”

• All games run in 16 bit color mode.
• Winstone 98 ran at 1024×768, 85 Hz refresh rate.

3D Game Benchmark Results – Forsaken Mark

Forsaken shows that G200 is almost at the same speed range as Voodoo2, and this at a better quality. Banshee can easily outscore a single Voodoo2 card. Savage3D is very close to Banshee’s performance, in the high resolutions the Savage3D has got a slight edge over Banshee. TNT can’t show off too much in forsaken, it’s pretty close to the performance of Banshee and Savage3D.

3D Game Benchmark Results – Turok Mark

Banshee is looking pretty good at Turok, regardless if running Direct3D or Glide. Savage3D doesn’t score that well in Turok as long as the `AA” setting is enabled, but I decided that a registry setting is either enabled for all test runs or for none, and Savage3D runs the mon2.dm2 demo in slow motion without `AA’ enabled. TNT scores very well in Turok and comes pretty close after Voodoo2 SLI, as long as you are using a powerful Pentium II and not a cacheless Celeron or even less.

3D Game Benchmark Results – Incoming Gameindex

Incoming shows quite a few interesting things. First of all it seems obvious, that 33 fps are as fast as the G200 can display Incoming at 1024×768, regardless which CPU is used. A Pentium II 400 scores exactly the same as a Celeron 266. For the other cards it seems obvious, that 51-55 fps are the magic border when using a Celeron 266, since all cards are scoring almost the same. Things look a lot different in a PII 400 system. Banshee scores quite a bit better than single Voodoo2 and Savage3D can surpass Banshee as long as the `FD’ registry setting is enabled. I wish I knew what this setting means. TNT scores extremely high with a PII 400 at 640×480, but Savage3D is catching up with it at 1024×768. When using a Celeron 266 it’s almost opposite, TNT looks best at 1024×768.

3D Game Benchmark Results – Quake 2 Demo 1

Q2 shows a very different picture to the previous game benchmarks. TNT looks very good indeed, scoring higher than Voodoo2 at 800×600. Now I would almost have called TNT the `Voodoo2-Killer’ if the performance of TNT wouldn’t drop extremely when using a slower CPU as the Celeron 266. With this CPU TNT is certainly the wrong choice for Quake2, unless you only care about 1024×768 resolution rather than frame rates above 40 fps.

Running demo1.dm2 as a benchmark in Quake 2 brings up one serious question in case of Savage3D. The scored fps raise continuously with the number of test runs, probably because the dynamic textures needed in the run are slowly but surely all available from the cache memory. When we play Quake 2 we hardly ever run the same way twice, so that the results scored with Savage3D are probably falsely high. Savage3D has certainly an obvious lead over Banshee in high resolutions as already seen in the other games. However, the Quake 2 image quality produced by Savage3D is the worst of the tested cards as well. Matrox runs Quake 2 on the G200 with a D3D wrapper for the time being. Once the full OpenGL ICD it available, G200 should score quite a bit better. I am expecting numbers as seen in Incoming. Nevertheless the image quality displayed by G200 is worlds apart from all the other cards in the test.

3D Game Benchmark Results – Quake 2 Massive1

The results speak for themselves. Voodoo2 still rules Quake 2, but Savage3D is coming pretty close. Banshee does not look as bad as some people suspected. 36 fps at 800×600 with a Celeron should be just fine.

TNT is again looking excellent with a P2 400, with a Celeron 266 it can only show off at 1024×768. Massive1 at 29 fps is not so bad though and playing Quake 2 at 1024×768 is anyway the best you can do, unless you want to get even higher.

3D Game Benchmark Results – Quake 2 mon2.dm2

S3’s own `mon2.dm2′ demo, using their own `newS3′ map is using a whole lot of really large textures, making the scenery look very detailed and pretty. It would have been a shame if Savage3D would not score well in it, but this would almost have happened. In the latest driver I received, the `AA” setting needs to be enabled to run mon2.dm2 at a decent speed. Anyway, in mon2.dm2 the Savage3D scores higher and higher the more test runs you do. I decided to put down the result of the second run. Matrox’ G200 looks very good at mon2.dm2 as well, obviously due to its AGP 2x interface. It reaches about 66% of the Savage3D scores and this without texture compression. The coolest thing are the TNT results however. TNT scores better than Savage3D although TNT does not use texture compression!!! Only with a Celeron 266 TNT looks a little slower than Savage3D, which is another proof for the CPU dependency of TNT in Quake 2. 3Dfx products have yet to learn what AGP means, so the mon2.dm2 is a serious threat to them. 22 fps scored by a $500 Voodoo2 SLI combo looks a bit sad, doesn’t it? Banshee looks even worse, its `AGP-interface’ only consists of the mechanical connector for the AGP slot.

3D Game Benchmark Results – Sin Demo killer.dm2

The first thing that catches your eye in the Sin results is the obvious inability of Savage3D to run it. That is another proof for the `maturity’ of the OpenGL D3D wrapper. The next surprising thing is that Banshee scores just the same as Voodoo2, it seems as if Voodoo2 cannot take advantage of its second TMU although Sin is using the Quake 2 engine. The TNT results are another proof that for some reason the Sin demo doesn’t seem to use multi texturing. TNT scores worse than Banshee, although it should certainly score better if it could take advantage of its twin texel engine. The Celeron 266 results of TNT are pretty bad again, just as in Quake 2. G200 scores ok, but I guess that everything under 30 fps isn’t really any good for playing Sin.

Office Performance Benchmark Results – Winstone 98

As you can see, the 2D scores of the three cards are almost identical at 16 bit color depth. Switching to 32-bit color has got no impact on the RIVA TNT, the largest impact on Savage3D. However, the results are still too close to speak of serious differences. All of the four chips are damn fast in 2D and RIVA TNT is the fastest, although with only an extremely tiny lead.

Benchmark Results – AGP Performance

The AGP performance of Banshee is simply non existent, since it is no proper AGP implementation. Savage3D looks pretty bad here too, which makes you wonder if it has a real 2x AGP implementation. G200 looks very good and TNT seems to have an excellent AGP implementation indeed.

Image Quality Quake 2

Download 640×480 BMP file (900kB)

That’s what `The Edge’ looks like with Voodoo2. Ugly banding of the sky, the fine structures of the building in the background look a bit rough, there’s no trilinear filtering.

Download 640×480 BMP file (900kB)

Matrox’ G200 paints a beautiful sky and the trilinear filtering gives the building in the background a smooth and realistic look.

Download 640×480 BMP file (900kB)

The sky drew by Savage3D looks as if the driver is still buggy. The structure of the building in the back ground looks similar to the V2 picture, seemingly no trilinear filtering used here either.

Conclusion

Conclusion, Continued

One thing is for sure, the days when 3Dfx’ Voodoo2 was the only real thing for 3D games are going to be over soon. Two Voodoo2 cards in SLI configuration will stay on top for a longer time, but you have to spend a serious amount of money for it too. Getting a Voodoo2 right now is mainly interesting for Quake 2 players and people who are planning on upgrading to a second Voodoo2 card later.

Voodoo Banshee shows that it certainly isn’t just a castrated Voodoo2 product. The 3D performance in games that use only single texturing is significantly higher than the performance achieved with one Voodoo2 card. Quake 2 does of course run slower on Banshee than on even a single Voodoo2, but it still doesn’t run slow. Compared to Savage3D, Banshee may be a bit slower in most of the benchmarks, but Banshee is running without any trouble. The benchmark results scored with Banshee do directly translate into game play performance, something that’s at least questionable with Savage3D. Banshee is significantly faster than the G200, but it cannot possibly live up to the G200’s 3D-image quality. Last but not least, Banshee’s 2D engine seems to be the best currently available, it runs at 32 bit color almost exactly as fast as at 16 bit. However, the 2D performance of the other two chips is not significantly less. AMD K6-2 owners may also realize that you currently need a 3Dfx Voodoo2 compatible card to take advantage of 3DNow! In Quake 2.

The Matrox Millennium G200 painted a positive picture here in my lab. It is not the fastest 3D chip of the three, as a matter of fact it’s the slowest one, but except for its average Quake 2 performance it’s still scoring pretty well. This 3D scores are achieved at the highest currently available 3D image quality, which may be of importance to many people. We shouldn’t forget that at the time when the G200 came out, it was the fastest 2D/3D solution available, still far ahead of products like NVIDIA RIVA 128/ZX, Intel i740 and such. G200 offers the second best 2D performance in this test and displays the best looking desktop picture.

NVIDIA’s RIVA TNT is certainly the most exciting product in this test and it made many of NVIDIA’s recent claims come true. The 6 million triangles/s claim however seems wishful thinking rather than actual truth, as you can see when looking at the results at 640×480 and compare them with the Voodoo2 results. At this low resolution the frame rate is not fill rate, but polygon rate limited and that’s exactly where the Voodoo2 looks better. 3Dfx only claims 3.3 million triangles per second for Voodoo2 though. Nevertheless TNT’s 3D performance is very good and the excellent image quality makes this chip a pleasure to play with. It was very impressive to see that TNT seems to have the best AGP interface to date, beating S3’s Savage3D in their own demo and that although TNT doesn’t use texture compression. There is a big downside for many of you however, since TNT does not like to be driven by a slow CPU at all. Owners of cacheless Celerons and all owners of non-3DNow! Socket7 CPUs should really get a different 3D chip, e.g. the Savage3D or upgrade to a 100 MHz Pentium II. Even the support for 3Dnow! is not yet existing and we can only hope that it will be fast and available soon.

The other serious issue is the chip clock. NVIDIA leaves the chip and memory clock speed to the card manufacturer and it could easily be that some chips don’t even do the 90 MHz, let alone the 95 MHz chip clock that my test board was running at. This will either result in crashes or in slower 3D performance. Make sure that you don’t get a slow TNT board. I will make a utility available to you, which can check and set the chip and memory clock.

The 2D performance of the RIVA TNT is excellent as well and so is the picture quality. This chip is certainly the hottest chip for 100 MHz FSB Pentium II users as long as you get a card with a fast running TNT chip on it.

S3’s Savage3D has cost me a lot of nerves in the last 2 weeks. Although I am sure that the chip has a great potential, I am also sure that it will take quite a while until it has matured. The drivers are not really stable, you cannot really play Quake 2, unless you paint a crosshair in the middle of your monitor, Sin doesn’t run properly at all, there are some strange lines produced when running Winstone98, ….

If Hercules should indeed start shipping their first Savage3D card this week, they are facing a whole lot of trouble. I wonder if this is worth the OEM deal for Xmas. First of all Hercules will have to ship those cards with a nice Q2 crosshair sticker for your screen.

What I like least about Savage3D however is the bad feeling that those benchmark numbers may not translate into real game play.

S3 has never reacted to this report and when you look at their website you will find that they are ignoring this review, whilst linking to all the positive Savage3D reviews on the web. S3 did not give me a chance of re-testing their board with improved drivers since this review was posted either. I have certainly criticized a lot of products and companies in the last two years, but I never experienced this ostrich kind of behavior of S3 with any other manufacturer. You should consider that when dealing with this company in any way.