Introduction
The most comprehensive 3D card performance evaluation 22 cards tested with four benchmarks and four different CPUs.
Drivers and Cards
This is the list of drivers and graphic chips that were used for benchmarking
ATI XPERT@PLAY | 4.10.2278 | ATI Rage Pro |
Asus V264GTE | 4.10.2278 | ATI Rage Pro |
Asus V3000 | 4.10.01.0205 | NVidia RIVA 128 |
Canopus Pure3D | 1.30.04 | 3Dfx Voodoo |
Canopus Total3D 128V | 4.10.01.0206-1.01.10 | NVidia RIVA 128 |
Diamond Fire GL 1000 Pro | 4.10.01.2248 | 3DLabs Permedia 2 |
Diamond Monster 3D | 4.10.01.0013 | 3Dfx Voodoo |
Diamond Stealth II S220 | 4.10.01.0101 | Rendition Verite 2100 |
Diamond Viper V330 | 4.10.01.015 | NVidia RIVA 128 |
Elitegroup 3DVision | 4.10.01.0905 | SiS 6326 |
Elsa Victory Erazor | 4.10.01.0202-0003 | NVidia RIVA 128 |
Elsa Winner/Office 2000 | 4.10.01.0204-0004 | 3DLabs Permedia 2 |
Hercules Dynamite 3D/GL | 4.10.01.3103-0286 | 3DLabs Permedia 2 |
Hercules Stingray 128/3D | 4.10.01.2063 | 3Dfx Voodoo Rush / Alliance AT3D |
Hercules Thriller 3D | 4.10.01.3465 | Rendition Verite 2200 |
Intergraph Intense 3D Voodoo | 4.10.01.2063 | 3Dfx Voodoo Rush / Alliance AT3D |
Quantum3D Obsidian 100SB 4440/4400 | 4.10.01.0036-0164 | 3Dfx Voodoo |
Videologic Apocalypse 3Dx | 4.10.01.0012-4.1.0.12 | NEC PowerVR |
The Videologic Apocalypse was tested together with a Matrox Millennium. Voodoo and Voodoo Rush cards were tested at a refresh rate of 60 Hz at 640×480, all other cards were tested at 120 Hz refresh rate at 640×480.
Turok TMARK
Turok is a pretty nice 3D 1st person shooter game with a lot of pretty 3D effects and really beautiful level design. For the benchmark you only need to download the demo version, which is playable as well. To run the benchmark you use ‘-benchmark’ as command line option and if you’ve got several different 3D cards in your system add the command line option ‘-alldrivers’ as well. This will let you choose which 3D engine you like to use. Turok supports Direct3D for all cards, Glide for cards with the 3Dfx Voodoo and Voodoo Rush chipset, a PowerVR engine for Videologic’s Apocalypse cards and Matrox’ M3D as well as a special engine for ATI’s Rage Pro chip, as found on the Xpert@.. cards and the new All in Wonder Rage Pro card. The beauty of the benchmark is that it runs for quite a while, showing you the actual frame rate, the average frame rate since you started the benchmark and after running through all the levels it also gives you the overall frame rate, called ‘TMARK’. Due to the long run time of this benchmark, the results are very stable and reliable. Running it twice will in most cases give you the exact same frame rate.
For running Turok TMARK you need only the free demo version. You can download it here directly from Acclaim’s website. As already said, start it with command line option ‘-benchmark -alldrivers’. Then you can choose the 3D device you want to use and can adjust the performance settings. I ran all tests with all performance settings ticked, except in case of Permedia 2 cards, where ‘render overlap’ has to be switched off if you want to see a picture. PowerVR was set to ‘normal’, Glide ran with ‘flip at vsync’ switched off and the Rage Pro engine ran with ‘fog’ enabled.
I used this benchmark to show the Direct3D as well as the special engine performance of the tested cards. It should give you a pretty realistic view of how the cards perform in real world games. I also think that Turok will become pretty successful, so that it will be important how well a card runs this game anyhow. 3Dfx used this benchmark at Comdex to show the performance of their upcoming Voodoo2 chip.
I ran these tests on a system with a Pentium II 300, using the Abit LX6, to see how well the cards perform on a high end platform. To have a view at the other end of the CPU performance spectrum I used a system with Cyrix/IBM 6x86MX PR200, which I consider as a pretty bad 3D gaming CPU. This systems used the FIC PA-2012. You can see that the results differ considerably, in some cases even shockingly.
Turok TMARK Penitum II 300 MHZ
Looking at the Pentium II 300 results shows pretty well that in quite a few cases not the CPU, but the 3D adapter was the limiting factor for the 3D performance. Especially Quantum3D’s brute 3D accelerating force shows how much you can get out of a PII 300 system as long as you are using the right 3D adapter. The Glide engine is obviously quite a bit faster than the Direct3D engine for 3Dfx cards, visible at results with Quantum3D’s monster card as well as the normal Voodoo and the Voodoo Rush cards. Glide rules this benchmark. You can see that Direct3D is still ruled by NVidia’s RIVA 128 chip, the five fastest Direct3D performers are all using this chip. Right after that there come the normal Voodoo cards at Direct3D. Then there is the newcomer ‘Thriller 3D’ with Rendition’s Verite 2200 chip, overclocked to 69.8 MHz, followed by the same card at default chip clock (62.6 MHz). You can see that Diamond’s Stealth II S220 with Rendition’s Verite 2100 is just as fast as the Thriller 3D as long as you clock it at the same speed. In the Stealth’s case this means overclocking to 62.6, because the default clock is significantly less. My Stealth II couldn’t get clocked faster than that. The cool thing is that the cheap ($99) Stealth II S220 is even faster than Voodoo Rush cards running with the Glide engine as long as you overclock it. The Voodoo Rush cards are a lot more expensive than the Stealth II though. Behind the two Voodoo rush cards you can find three cards with 3DLabs’ Permedia 2 chip. The ELSA Winner Office lead is mainly due to the fact that it doesn’t display smoke properly, you can clearly see some rasterization, which you cannot in case of the Diamond Fire GL 1000 Pro. The last third is lead on by Videologic’s Apocalypse 3Dx, next is a Voodoo Rush card at Direct3D and then there comes the biggest disappointment of this test, the ATI Rage Pro cards. Although Turok is using a special Rage Pro engine, the performance as well as the image quality are pretty sad. Don’t even bother running the Rage Pro under the Direct3D engine, it looks really horrible. At the very end you’ll find Intergraph’s Voodoo Rush card under Direct3D and then a completely unknown card with the SiS chip 6326. This card is certainly very slow, but the image quality is just fine. You wonder how it comes that this unknown card can display Turok well under Direct3D while the ATI Rage Pro cards can’t. Hercules’ Dynamite 3D/GL is unfortunatley unable to run Turok. Unlike the other cards with 3DLabs Permedia 2 chip it has got some serious driver problems. The last driver (build 254) would run Turok, but without performing any bilinear filtering, which looks pretty ridiculous, the latest driver (286) only shows some blue screen with an arrow somewhere in it. I guess Hercules will have to ask Diamond or Elsa how they got their drivers to work properly.
Turok TMARK Pentium MMX 200
Running Turok on a Pentium MMX 200 shows still the different classes, although the lead of the Obsidian isn’t that obvious anymore. The top is ruled by 3Dfx Voodoo, next is NVidia’s RIVA 128, then Rendition’s V2x00, followed by a mix of 3Dfx Voodoo Rush and 3DLabs Permedia 2. As you will see when looking at all charts, the Apocalypse 3Dx with its PowerVr chip is on the continuos way down the chart while CPU 3D performance gets less.
Turok TMARK AMD K6 233
AMD’s K6 at 233 MHz doesn’t change the above picture much, the lead of the Obsidian melts slowly away, but Voodoo is still No. 1. Rendition’s V2x00 and 3Dfx’ Voodoo Rush are on the continuous way up the chart as the 3D power of the CPU decreases.
Turok TMARK – Cyrix/IBM 6x86MX PR200
Running TMARK with a low end 3D CPU shows the world a lot different. The biggest surprise is Quantum3D’s Obsidian 100SB. This amazing high tech card is let down completely by the 6x86MX. A card that can reach highest frame rates requires one important thing: a CPU that feeds it with data fast enough. In case of the 6x86MX the Obsidian can be as fast as it wants, the CPU just wouldn’t let it run. Due to the fact that the Obsidian was not overclocked as the other two Voodoo cards were, it performs even worse than a Pure3D or a Monster3D which doesn’t even cost 1/8 of it. It is reassuring to realize that someone that spends over $700 or even $1600 for a 3D accelerator will certainly not buy a 6x86MX CPU.
Going through the results shows that Voodoo and it’s Glide engine still rules, then there are the Direct3D kings with NVidia RIVA chip again, but after that it’s different to above. Hercules’ Stingray 128/3D with Voodoo Rush chip shows that it’s a good choice for a low end 3D CPU, offering >25 fps and hence playability. However the overclocked Thriller3D is right behind it and may be even a better choice as we’ll see further down. After that there are the Voodoo cards under Direct3D and in between the two cards with Rendition’s Verite 2×00 chip, both at the same chip speed of 62.6 MHz and with the same performance. The PowerVR chip performs worse than above, something that’s quite known, and please let’s just not talk about ATI’s Rage Pro … it’s just a shame. The SiS 6326 card was running at something like 10s per frame, so that I canceled the test.
Image Quality – Turok TMARK
As if you had expected it, Turok looks best with 3Dfx Voodoo and Voodoo Rush under the Glide engine. The only chip that comes close in image quality is Rendition’s Verite 2×00, as found in Diamond’s Stealth II S220 and Hercules new Thriller 3D. After that it’s quite surprising to see how well Turok looks with Diamond’s Fire GL 1000 Pro, who have obviously the best Permedia 2 driver currently available. Elsa’s Winner/Office 2000 does almost look the same, but it shows some ugly rasterization of the smoke that occurs when shooting somebody. NVidia’s RIVA offers a decent image quality, but it’s not as beautyful as Turok could look, the PowerVR has some serious problem with the sky and its transparency. The worst image quality is offered by the ATI Rage Pro cards, and that even though Turok has a special engine for it. The biggest surprise is the SiS6326 card, which shows a very pretty image quality
Conclusion – Turok TMARK
We’ve got to face it, Voodoo still rules. When a game comes with Glide support (and which game dares not to nowadays?) it performs and looks best on Voodoo, regardless if you’ve got a fast or a slow CPU. This will probably only change a little bit in the future, instead of ‘Voodoo rules’ is will be ‘Voodoo 2 rules’. Games that indeed only support Direct3D perform best on 3D accelerators with NVidia’s RIVA 128 chip, closely followed by Rendition’s Verite 2×00 family. If higher image quality is worth the few less fps to you, consider buying a Thriller3D, the Verite looks definitely better than the RIVA. For owners of systems with weak 3D performing CPUs the Hercules Stingray 128/3D may be worth considering as well, because its Voodoo Rush performs pretty well in those systems and it offers Voodoo top image quality. However those systems would benefit just as well from a Diamond Stealth II S220, which is considerably cheaper. Cards with 3DLabs Permedia 2 chip aren’t really that great for gaming and are only interesting for people that have their second or third priority on 3D gaming. The PowerVR is clearly outperformed by other solutions in Direct3D, so that one rather goes for a 2D/3D card than for this only 3D add-on solution. Last but not least there’s the ATI Rage Pro. I would say that you also shouldn’t have a high priority on games when you want to buy this card. Some games run fine and fast on it, but it seems to have some serious Direct3D problems, as this test clearly showed.
GLQuake BIGASS1 Benchmark
Some of you may wonder what this nasty word ‘bigass1’ is about, but unfortunately this is the official name of a special downloadable .dem file for Quake. The special thing about this demo is that it runs much longer than the well known ‘demo1-3’, which are implemented into Quake and it is much more complex. The bigass1 demo shows a deathmatch between about 16 players and it gets the last out of your Quake configuration. The big advantage of this demo over the demo1-3 ones is that it’s MUCH more realistic. The frame rate it returns as result is much closer to your realworld Quake performance, especially when you’re playing on the net. Now before you look at the results please make yourself aware that demo1-3 does NOT show the real truth. Run bigass1 on your own Quake configuration and don’t be surprised if the frame rate makes you sad. This is the REAL Quake frame rate, it’s not for fps-horney quakies, that can’t play, but impose with their astronomical timedemo frame rate everywhere. This benchmark comparison will show, that the Quake performance of 3D accelerators is possibly quite different to what you’ve read elsewhere and even here in the previous review before.
Bigass1 is one multiplayer DM demo of many, which you can find at Scary’s Quakeholio – Deathmatch Demos. It’s a 1.7 MB .dem file, which you simply put into your id1 directory. To run this demo as a benchmark you just type ‘timedemo bigass1’ instead of the usual ‘timedemo demo1-3’.
The NVidia RIVA 128 cards were ran with NVidia’s Alpha 2 OpenGL driver. Of course am I hoping as well, that NVidia will eventually come up with a faster and better looking driver and the fact that it’s currently not even a beta, shows that we can probably expect a lot still.
The benchmark was ran on a high end Pentium II 300 system with Abit’s LX6 motherboard and also on a low end 6x86MX PR200 system with FIC PA-2012 motherboard. Both systems were equipped with 64 MB SDRAM, Quake ran without any tweaks, no autocfg file, no status bar, 640×480.
GLQuake BIGASS1 Benchmark – Pentium II 300 MHz
The results of the high end system start off as in the Turok benchmark. Obsidian rulez !!! An amazing 60 fps with the 2 TMU version of the 100SB in bigass1 is actually equivalent to 94 fps in demo2. You can see that the cheaper version 100SB 4400 with only 16 instead of 24 MB is still worth the investment, 51.4 fps are still way ahead of the competition. After that the field is lead by the normal Voodoo cards, what else? The PowerVR in Videologic’s Apocalypse 3Dx is actually performing quite impressively here at place 5, but let’s not forget that the PowerVR GL engine does not support the lighting effects in Quake. Right behind the PowerVR there’s Rendition’s Verite 2×00 family again, first the overclocked Thriller 3D, then the default Thriller 3D and just as fast the overclocked Stealth II. It’s impressive that the Verite 2×00 could even beat the two Voodoo rush cards, which pretty much close the field if we are talking of playability. 20 fps should be the limit. The RIVA cards as well as the Permedia 2 cards are showing some considerable performance, but I wouldn’t want to Quake death matches on the net with these cards. I actually included the Elsa Winner/Office only due to a little trick, because the Win95 driver comes without any OpenGL support. I installed the Fire GL 1000 Pro driver first and then installed the Elsa driver on top of it. This way the Winner can use the Permedia 2 OpenGL driver of the Fire GL 1000 Pro.
GLQuake BIGASS1 Benchmark – Pentium MMX 200
The picture hasn’t changed much, but evrything is a little closer together. The Apocalypse 3D takes a dive from place 5 to place 10 and the Voodoo Rush cards mark pretty much where’s the limit of playability. The most interesting thing in this chart is the small difference of the score of many cards compared to their performance in a Pentium II 300 system. The Hercules Thriller 3D only moves from 23.7 to 24 fps, the Intergraph Intense 3D Voodoo moves from 18.9 to 20.2. This should make you clear how little a faster CPU gives you when playing Quake on these cards.
GLQuake BIGASS1 Benchmark – AMD K6 233
The AMD K6 may be performing very well in Direct3D, but in GLQuake it takes a hit. The decrease in frame rate compared to an Intel Pentium MMX system is very significant when you look at the CPU sensitive cards like the Obsidian 100SB. However, an owner of a Hercules Thriller 3D, a Diamond Stealth II S220, a Hercules Stingray 128/3D or an Intergraph Intense 3D Voodoo couldn’t care less about owning an AMD or Intel Socket 7 CPU. The results are close to identical. This should be very interesting to those who want a powerful CPU for their business system, where the AMD easily beats the Pentium MMX. It does NOT always have to be an Intel CPU for good 3D gaming.
GLQuake BIGASS1 Benchmark – Cyrix/IBM 6x86MX PR200
Playing GLQuake on a system with 6x86MX PR 200 is quite a commitment to make. The CPU does not have enough power to supply the powerful Obsidian card with enough data so that it even falls back behind the normal (but overclocked to 60 MHz) Voodoo cards. The performance is clearly limited by the CPU power. If you really want to play GLQuake with a low end system, don’t go for anything else than either Voodoo, Voodoo Rush or Rendition Verite 2×00, anything else pretty much sucks, I gotta say it this way. The worst experience was benchmarking RIVA PCI cards with bigass1.
Please notice the almost identical result of the overclocked and the non-overclocked Thriller 3D. The low FPU geometry setup power of the 6x86MX keeps the Thriller 3D from showing a better 3D performance, just as the Obsidian. The overclocking of the Stealth to 62.6 MHz still generates some 3D performance increase, but more than that doesn’t improve it anymore.
Image Quality – GLQuake BIGASS1 Benchmark
Again I have to say that the only real thing is Voodoo, but Rendition Verite 2×00 cards look almost as good. Permedia 2 cards offer a pretty decent image quality, although the Dynamite 3D/GL shows some strange rasterization. NVidia’s alpha 2 release of the OpenGL engine is not as dark as before anymore, but it still isn’t up to compete with Verite or Voodoo yet. The PowerVR does not support the lighting in Quake, which makes it look really dull.
Conclusion – GLQuake BIGASS1 Benchmark
If you’re a real Quake freak and if you’ve got the cash, go for the Quantum3D Obsidian 100SB to get Voodoo2 performance now. This card rocks big time as long as if you’ve got at least a K6 or Pentium MMX 200. Otherwise my doctor recommends Voodoo. If you want a cheap 2D/3D solution that still does pretty well, go and get the Diamond Stealth II S220 and overclock it to 62.6 MHz. This will make it as fast as an unoverclocked Hercules Thriller3D which costs double, but which has got its temptations too. It’s faster and offers a lot more features.
Quake II Timedemo Demo1 Benchmark
Unfortunately there isn’t any ‘bigass1’ for Quake 2 yet, so I had to go for the implemented demo1. However the results are pretty close to the bigass1 results in GLQuake, so that it seems as if the implemented benchmark is better than demo1-3 in Quake. It’s actually the next proof for the high authenticy of ‘bigass1’ against ‘demo1-3’ in Quake.
To run the benchmark in Quake II you have to type a little bit different commands than what you are used from Quake. First type ‘timedemo 1’ to switch on the benchmarking mode, then type ‘map demox.dm2’, where ‘x’ stands for the number of the demo you want to run. I ran ‘demo1’.
The benchmark was ran on a high end Pentium II 300 system with Abit’s LX6 motherboard and also on a low end 6x86MX PR200 system with FIC PA-2012 motherboard. Both systems were equipped with 64 MB SDRAM, Quake ran without any tweaks, no autocfg file, no status bar, 640×480.
Quake II Timedemo Demo1 Benchmark, Continued
The results are pretty much the same as in the Quake Bigass1 Benchmark, with the only difference that cards with Rendition’s Verite 2×00 chip are performing even better here. You can see here also how with decreasing CPU 3D power the Rendition V2x00 and the Voodoo Rush cards are moving up and the Apocalypse 3Dx is moving down the chart. There shouldn’t be any doubt about it that Quake II is pretty much unplayable with a 6x86MX unless you can put up with less than 20 fps or run it at a lower resolution. The conclusions for Quake II are the same as in the bigass1 benchmark. Take Voodoo or get Verite 2×00, consider Voodoo Rush for less powerful systems and leave the rest, basta.
PCPlayer D3D Benchmark – Pentium II 300 MHz
I don’t want to say that this is an exactly highly sophisticated benchmark, but it shows quite nicely the basic Direct3D performance. I am running this test with stars (only 2D) and anti-aliasing off, the latter switched off because the Rendition Verite 2×00 chips are the only ones that can do anti-aliasing and it would give cards with this chip a performance disadvantage in this benchmark.
You can see that it takes a double pack Voodoo in SLI technique, as offered in the Quantum3D Obsidian 100SB, to outperform NVidia’s RIVA in Direct3D, the Thriller 3D from Hercules and the overclocked Stealth II S220 from Diamond are even faster than the ATI Rage Pro cards, which at least don’t have any problems in this benchmark. Hercules’ Dynamite 3D/GL is a lot faster than the other Permedia 2 cards, I wonder if this is because it can’t run Turok and has therefore not all the required D3D functions included in its driver, which makes it less complex and therefore faster.
PCPlayer D3D Benchmark – Pentium MMX 200
This benchmark is pretty useless for a Pentium MMX 200, since it hardly shows any differences anymore between most of the cards. The Obsidian 100SB is still outstanding and the Voodoo Rush as well as the PowerVR card(s) are performing worst.
PCPlayer D3D Benchmark – AMD K6 233
The K6 performs better than the Pentium MMX in this benchmark and it shows the well known order. Obsidian, V2x00, RIVA, …
PCPlayer D3D Benchmark – Cyrix/IBM 6x86MX PR200
Can you see that the overclocking of the Thriller 3D doesn’t make it faster anymore? Surprised? This is another case of low FPU power limiting the 3D performance of a system. Even overclocking much higher wouldn’t give any performance increase anymore, look where the Obsidian 100SB is located. In low end systems the Verite 2×00 is a very good choice, for Direct3D the RIVA is a good choice too and for real games that run under Glide or OpenGL get the Voodoo.
CPU Scaling of 3D Graphic Chips
After having a close look at the benchmark results of the different graphic cards we learned that you cannot predict the 3D performance of a 3D accelerator in your own system by just transferring the 3D performance results of a system that’s faster or slower than yours. This means that if you only get 3D performance results of a high end Pentium II system, you may think that the slowest card in this comparison is also the slowest card in your K6 system, which we now know is seriously wrong.
The reason to this strange occurrence is that 3D chips don’t necessarily scale linearly with CPU performance. A 3D graphic chip has its very own maximal polygon and pixel fill rate, which simply limits its 3D performance. On the other side, a CPU has got its own maximal 3D geometry setup performance, mainly determined by its FPU, which limits the amount of geometry data it can send to the graphic chip. One of these two determine the frame rate scored in a 3D benchmark. If the CPU is lacking of 3D geometry setup performance (pretty much equal FPU performance), the graphic chip has to wait for geometry data from the CPU, so that the frame rate is determined by the CPU performance. However, if the graphic chip is at its limit of pixel fill rate or polygon rate, the CPU can try sending geometry data to the graphic chip as fast as it wants, the graphic chip can’t produce a higher frame rate anymore. In this case the frame rate is not determined by the CPU at all, it’s the graphic chip that limits it.
Now what does this mean to us?
After publishing my first 3D benchmark results with a Pentium II 300 and a 6x86MX PR200, I received a very rude mail from a guy who accused me of lying, because in his Pentium MMX 166 system, his Voodoo Rush card scored only about 1 fps less than what I had published as the result of the same card running in a Pentium II 300 system. He could not believe that with a CPU which is 3 to 4 times as fast as his, the frame rate would only increase by a few percent. His argumentation was certainly sensible, but he forgot that it could be, that the Voodoo Rush chip is either scaling very slowly or maybe already at its performance limit in a small Pentium MMX 166 system.
The same wrong estimation could happen if you look at how the Apocalypse 3Dx scored in the system with the Pentium II 300. It reaches 5th place, scoring a very decent 77% of Voodoo performance. However, if you have got a system with e.g. a quite respectable 6x86MX 200, you should not expect the PowerVR chip to score 5th in this system as well. In this system it drops down to 12th place, only offering 58% of Voodoo performance anymore.
If you want to buy a 3D graphic card, you got to realize the 3D geometry performance of your system. Currently the IDT WinChip is offering the worst, the next one is the 6x86MX, then the K6, followed by the Pentium MMX and the best 3D geometry performance in X86 systems you’ll of course find in Pentium Pro/Pentium II systems. Then you have to know if you want to play Quake, Quake II or any other game that’s using this engine. The Quake/Quake II engine is the most FPU power hungry 3D engine, Turok e.g. is not as particular about the FPU performance and many other Direct3D games like e.g. Forsaken or Incoming aren’t either. After you are aware of these two parameters, you want to take a look at the ‘CPU Scaling of 3D Chips’ charts below.
Now if you have got a powerful 3D system with a Pentium MMX or Pentium II, you want to go for a 3D chip that reaches high absolute results in frame rate as well as scales well over the CPUs. The number one chip to achieve this is currently the RIVA 128 from NVidia, which obviously is far from its maximal 3D performance even in a system with Pentium II 300. Videologic’s PowerVR chip is another one that scales well, but doesn’t score as high, and the same is valid for 3DLabs’ Permedia 2. 3Dfx Voodoo only scales excellent if you buy it in form of a Quantum3D Obsidian 100SB, the normal Voodoo cards reach their absolute 3D performance limit somewhere between Pentium MMX 233 and Pentium II 300. However, the absolute 3D performance of the Voodoo is what makes it still the best choice for a high end system as well. The soon released Voodoo 2 will solve this problem and scale much better.
The owner of a mid range to low end system still wants a 3D chip that scores high absolute, but it doesn’t matter if the chip scales well, as long as the performance is good in systems with lower 3D geometry performance of the CPU. The number one choices in this case are besides the 3Dfx Voodoo, which always is an excellent choice, the 3Dfx Voodoo Rush and the Rendition Verite 2×00. In my benchmark the Voodoo Rush chip is the one that scales least. It increases the Turok frame rate by unbelievable 11% from a system with a 6x86MX PR200 to a Pentium II 300. The Verite 2×00 scales better, but it reaches its limit somewhere around the Pentium MMX 233 and the Pentium II 233.
OK, so far about the theory, have a look at the results yourself.
CPU Scaling of 3D Graphic Chips – 3Dfx Voodoo
A normal Voodoo card like the excellent Pure3D from Canopus only shows a pretty small increase in Turok frame rate from a system with a Pentium MMX 200 to a system with a Pentium II 300. In GLQuake and Quake 2 the story is a lot different, there are still enough reserves within the Voodoo to make owners of the new Deschutes still happy.
Things are a little bit different with Quantum3D’s Killer board Obsidian 100SB. It uses double the chips of a normal Voodoo card and lot of more RAM, the 100SB 4440 uses even another couple of Texelfx chips more. This card is still craving for CPU performance and the Pentium II 300 makes it glow but not shine yet. The Obsidian has one big let down, which is its price and the Voodoo 2 that is turning up on the horizon. If you don’t mind around 700 bucks for a 4400 or even more for a 4440, go and get this card in your system, it rocks big time.
CPU Scaling of 3D Graphic Chips – 3Dfx Voodoo Rush
The Voodoo Rush is some kind of phenomenon to me. It offers a pretty good frame rate in Turok and a decent but not great frame rate in Quake and the CPU that’s used in the system doesn’t seem to scratch the Voodoo Rush at all or at least hardly. Pentium II and even Pentium MMX owners should avoid cards with that chipset unless they want to castrate their system. Owners of the 6x86MX or the K6 166 should have a close look at these cards though. After all they offer Voodoo image quality.
CPU Scaling of 3D Graphic Chips – 3DLabs Permedia 2
The Permedia 2 scales …. well. Well, the area where it scales is not very interesting though. 3D Winbench 98 can give you as high frame rates of that chip as it wants, this is altogether no chip for 3D gamers. Period! Of course, in case you’ve got a Pentium II 800 you may want to consider it …
CPU Scaling of 3D Graphic Chips – ATI Rage Pro
Besides the fact that the Rage Pro isn’t very lucky with GLQuake, Quake II or Turok, it scales alright, although I consider the performance increase of 24 to 21 (Pentium MMX 200/Pentium II 300) as too little. I guess that the limitations of the Rage Pro are lying somewhere around the PII 266 or 300.
CPU Scaling of 3D Graphic Chips – NEC PowerVR
As well known, the PowerVR scales extremely well, too well for owners of slow systems. This chip would be interesting for Pentium II owners if its absolute 3D performance was a little bit better and would get closer to Voodoo’s performance.
CPU Scaling of 3D Graphic Chips – NVidia RIVA 128
There’s no doubt about it, the RIVA scales very well, easily visible at the jump between P5 MMX and PII. However, GLQuake and Quake II are currently not one of its strengths. New OpenGL beta drivers are supposed to be close to their release though.
CPU Scaling of 3D Graphic Chips – Rendition Verite 2×00
You can see it, Slot 1 doesn’t really benefit from this chip anymore. The increase from P5 MMX 200 to PII 300 is too small to justify the use of the V2x00 in a Pentium II system. Voodoo or Voodoo 2 is the better choice here.
CPU Scaling of 3D Graphic Chips – SiS 6326
Only to be complete, here’s the amazing results of the SiS 6326. It seems to scale somehow, down there in the frame rate nowhere land.
Image Quality 3Dfx Voodoo and Voodoo Rush
The above image is the ideal look, produced by cards with the 3Dfx Voodoo or Voodoo Rush chip, running under the 3Dfx own Glide engine. You can try whichever card you want, the image quality won’t get better than that.
Running the Voodoo under the Direct3D engine shows that the quality gets definitely worse. The sky looks less smooth, the transparency of the fog is less than above. Other cards can do it better under Direct3D.
Image Quality ATI Rage Pro
Running Turok with ATI Rage Pro cards under the Direct3D engine is only for forgivable people, but a little bit faster than under the Rage Pro engine. However please look at the sky and the artifacts at the rock on the right. You can’t really play with this engine. I just wonder what’s wrong with the Rage Pro’s Direct3D interface. It can’t be Turok, because otherwise an unknown card as the SiS6326 wouldn’t look as good as it does under Direct3D. The statement that the Rage Pro runs every Direct3D game is simply wrong.
Even the image quality produced under Turok’s special Rage Pro engine is simply bad. As addition to this, the frame rate is bad too.
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Image Quality NVidia RIVA 128 Cards
NVidia’s RIVA 128 is not famous for highest image quality and I am sure that you can see the reason why. The waterfall is not filtered as fine as in the reference picture below and the sky and clouds don’t look as good as on the Voodoo either. However, the RIVA doesn’t seem to lack of any particular feature when running Turok, it just doesn’t look quite as good as it could. At least you get a fair trade off here, since the RIVA may not be the most beautiful, but the fastest at Direct3D.
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Image Quality NEC PowerVR Cards
Although Turok uses the the dedicated engine, the PowerVR doesn’t produce a very nice picture. Compared to the reference picture the sky and clouds aren’t transparent at all, looks as if the PowerVR produces some really heavy weather. The waterfall isn’t fultered very well and there seems to be a Z-information problem with the rock on the left. All in all is Turok playable, doesn’t look too bad, but most other cards are offering a better image quality under Turok.
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Image Quality Rendition Verite 2×00 Cards
The cards with Rendition’s Verite 2100 or 2200 produce an excellent image quality. If you compare it to the reference picture of the Voodoo below, you’ll only notice slightly less transparency of the fog as well as a less realistic water surface. Considering that the Rendition chip doesn’t even use its own RRedline, but the Direct3D engine, shows how high the image quality of this chip is.
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Image Quality Diamond Fire GL 1000 Pro
The Fire GL 1000 Pro is producing an impressively good image quality. If you compare it to the reference picture below, you’ll only see that the fog produced by the Fire GL 1000 Pro is less transparent and that the reflexions on the water are looking less realistic. The 3DLabs Permedia 2 chip that’s used here runs also only under Direct3D, not under a special engine. Turok does definitely look well on the Fire GL 1000 Pro.
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Image Quality Elsa Winner/Office 2000
This picture looks just the same as what the image sample of the Fire GL 1000 Pro looks like. However the smoke that’s produced when shooting is somehow rastered, as you can see below.
These strange rasters are not produced by the Diamond Fire GL 1000 Pro, although it uses the same graphic chip.
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Image Quality Hercules Dynamite 3D/GL
That’s what Turok looks like if you are using the driver build 254. No bilinear filtering makes the image look ridiculously, it’s no miracle that the framerate is higher than of other Permedia 2 cards.
That’s what Turok looks like when using driver build 286 with the Dynamite. Please notice the nice waterfall and the little birdies ….
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Image Quality SiS 6326
This slow and pretty unknown graphic chip produces a quite nice image quality. It’s the sky and the fog that are looking worse than the reference picture. This is how Direct3D can look on a chip that is certainly not specially supported by Turok and has probably never been tested by Acclaim either. ATI should have a close look at this picture, before they accuse Turok to be the reason for the bad image quality of the Rage Pro chip in Turok.
Reference Picture
The above is the ideal picture, produced by a 3Dfx Voodoo card under Glide engine.
Discussion of the Rendition Verite 2×00 Family of Graphic Chips
When the new generation of graphic chips started shipping last year, it took quite a while until the first cards with Rendition’s Verite 2×00 turned up on the market. 3DLabs and NVidia were a lot more successful in marketing their chips and opposite to these two companies, who got a lot of vendors using their chips, Rendition had first only one, Diamond, and only now the Thriller 3D of Hercules starts shipping. If you listen around large graphic card vendors, you wonder what the future of Rendition looks like, since I hardly came across any vendor that is planning future graphic cards with Rendition chips. This seems pretty sad to me, because the current Verite 2k chips are far from uninteresting and I think that Rendition certainly deserves a lot of respect in the 3D chip market still. Let’s hope that they’ll get a better reputation soon, because I think that the current products with Rendition chips are far from unattractive and I can take pride in saying that I was one of the first who published that.
As mentioned above, there are currently only two mainstream graphic cards with Rendition’s Verite 2×00 chips available, Diamond’s Stealth II S220 and Hercules Thriller 3D. They don’t use the same chip, the Stealth uses the cheaper V2100 with only 170 MHz RAMDAC and Hercules uses the more expensive V2200 with 220 MHz RAMDAC. This is not the only difference between the two cards, because whilst Diamond puts the Stealth II S220 into the low market segment and sells it for practically only $99, the Thriller 3D comes with a lot of fancy additions and is placed into the mid to higher market segment. I don’t want to talk about the nice video in/out and the stereo glass feature of Thriller 3D, because I’ll do that on the dedicated review page for this card. Rather than that is this article meant to discuss the 3D performance aspects of these two cards.
First of all, what’s the difference between V2100 and V2200? It’s actually less than you might think. Due to the different RAMDAC speeds, the V2200 enables higher refresh rates. This can be a performance advantage in case you have a monitor that supports 160 Hz refresh rate at 640×480, because the 3D performance of the V2x00 is depending on the refresh rate due to the well known ‘flip at vsync’ issue. This ‘flip at vsync’ means that the graphic chip may only draw a new page into the screen buffer memory when a whole picture has been displayed on the screen. The lower the refresh rate, the more often the graphic chip has to wait until it can transfer a new picture into the frame buffer and hence the fps go down. This issue is well known from the 3Dfx Voodoo cards, where almost everyone has switched off this feature with the special Glide setting ‘SET SST_SWAP_EN_WAIT_ON_VSYNC=0’ in the autoexec.bat or in the batch that starts the game. Hercules also allows you to switch off this feature, but the game becomes unplayable by that, which does not happen in case of a Voodoo. The ‘flip at vsync’ is a specification in Direct3D which was modified by Matrox’ Mystique for the first time last year, which upset a lot of people that said Matrox is cheating. Nowadays it’s pretty common to switch off this rule as long as you can still play the game decently. Anyway, the performance of a game running under the Rendition V2x00 increases with higher refresh rates and hence the V2200 can have an advantage here, the V2100 is limited to a maximum of 120 Hz at 640×480. The 200 Hz setting on my Thriller 3D doesn’t work, although my 21′ monitor would be able to display that, 160 Hz works though. The other advantage of a faster RAMDAC is the ability to have a faster 2D engine, just because the memory is not used as long by the RAMDAC and hence free for the graphic chip.
There are a little bit more advantages of the Thriller 3D however, its SGRAM can be clocked higher than that of the Stealth II. This is quite surprising, because both are using the same kind of SGRAM, only 10ns one. Since the memory usually runs at double the Verite’s clock, it’s very surprising that it survives almost 140 MHz. This should normally take 7 or even 6 ns SGRAM. Anyway, my Stealth II could only be clocked up to 62.6 MHz, the Thriller runs up to 69.8 MHz. By default the Stealth II is only running at 40 MHz, which is very little and that’s why you can get a significant improvement in 3D performance by overclocking it to at least 60 MHz. The Thriller already starts off with 62.6 MHz and so the increase isn’t quite as high as what you can reach with the Stealth II. The interesting thing is in any case, that the Stealth II at 62.6 MHz is just as fast as the Thriller 3D at standard settings. This is so impressive, because the Stealth is about half the price of the Thriller 3D.
The most important things you have two know about both cards are that they produce a very good image quality. Also are cards with Rendition’s Verite 2×00 chip the only ones that can do anti-aliasing currently. This is mainly a very fancy feature and hardly ever used in any game due to its performance impact, but it shows how seriously Rendition takes the implementation of as many 3D features as possible. A special thing of all Verite chips is that it can be programmed with a special microcode. This enables the Verite to easily understand OpenGL when needed, you just have to load the correct OpenGL microcode. The Verite 2×00 is just as the V1000 using Rendition’s RRedline graphic engine, which is supposed to be very powerful as well, however, there are hardly any new games that support this particular engine right now. Another advantage of the V2x00 is the ability to run with up to 8 MB of onboard memory, which enables 3D at a resolution of up to 1024×768 and more. The typical 3D performance of the V2x00 is pretty good in systems with lower CPU 3D performance, but it doesn’t scale as well as e.g. the NVidia RIVA 128 chip. This means that it’s not the greatest choice for high end systems, although it’s definitely scaling better than the Voodoo Rush chip from 3Dfx. Socket 7 systems benefit greatly from it though and that’s currently still the majority. Altogether can’t you go too wrong getting a Stealth II or a Thriller 3D if you have got a Socket 7 system. You will get good 3D performance at very good image quality. However, the 2D performance of the V2x00 is a different story and I couldn’t see any impact of the overclocking on 2D performance at all.
Specification of the Rendition Verite 2×00
This is the specification of the V2200:
- 55 million pixels per second fill rate (high quality pixels)
- 2 million triangles per second (edge walk)
- 550,000 triangles per second (fully setup, high quality triangles)
- Over 50M WMs 2D acceleration (WinBench’96)
- 100 MHz SGRAM / SDRAM memory (up to 16MB)
- Exceeds Microsoft “Entertainment PC 97” requirements
- Exceeds Intel “Graphics Controller 97” requirements
- Complete on-chip triangle setup
- Asynchronous rendering with seperate setup, triangle and pixel engines
- Richest set of advanced 3D features
- True per-pixel perspective correction
- Sub-pixel / sub-texel positioning accuracy
- Flat and Gourad shading
- Z-buffering(optional)
- Anti-aliasing
- Comprehensive texture mapping support inc. video textures
- MIP-mapping
- Bi-linear and tri-linear filtering
- Support for texture animation, morphing and other special effects
- Complete set of blending functions including alpha blending, fog, chroma-keying
- 60 fps full-screen MPEG-2 video playback
- Video input port
- Digital video output port
- “SofTV” for flicker-free TV output
If you want more in depth information about the V2x00 chip from Rendition, why don’t you just read what Rendition has to say about their product.
For all those of you who went crazy already, because you don’t know how to overclock your V2x00 card, here the downloadable V2x00 overclocking program and I can also explain how you can do without it:
You need to create a text file called ‘verite.ini’ and put it into the Windows directory. If you’ve got the Thriller 3D, it’s already in there. It should have the following lines:
[display] m=x
n=2
SClkP=4
MClkP=2
x is the only variable you want to change, the rest is responsible for the settings of the ratio between chip and memory clock, which is 1:2 by default.
x = Chipclock / 14.318 * 8,
where ‘Chipclock’ is the clock frequency in MHz at which you want the chip to run at. Due to the odd 14.318 (typical clock quartz) you’ll have to round x up and down a bit. This calculation stuff is all done by the above mentioned Win95 program, so you may as well get it.
There is something I almost forgot, because I tend taking this for granted, which is certainly wrong. Overclocking the Verite is something you do at your own risk and there is definitely a variation amongst the cards with V2x00 chips regarding how far you can overclock them. This depends on the used SGRAM chips as well as on the V2x00 itself. My experience also showed that it differs from motherboard to motherboard, which is most likely due to variations in the PCI voltages. I don’t think you take a big risk in damaging the card, but it can happen that your system simply freezes and you’ll lose your data. Hence I’d recommend overclocking the V2x00 only for playing games, especially since the 2D engine does hardly benefit from the overclocking at all. It also makes sense putting a flat heat sink on the chip and especially using a 3Dfx Cool Fan, which works just as well as for Voodoo cards. Please be careful with the overclocking and increase new clock frequencies slowly step by step when testing it out.
For the ones that haven’t got the latest miniGL port for the V2k yet, download it here
Summary
After running all these game benchmarks on all these cards and this on two different systems I feel that I know well, which 3D accelerator I would recommend. Sure, in a few months the Voodoo 2 will rule the world, but what’s the beef right now?
In the first place you should be clear about your budget. If money isn’t of importance, the Quantum3D Obsidian 100SB 4440 is the card you have got to get. This card offers Voodoo2 like performance now and blows away anything else on the 3D game accelerator market. The cheaper version 100SB 4400 does it for people that have some concerns about how much money they spend. It runs Direct3D games exactly as fast as the 4440 and Glide games that don’t support the 2 TMUs of the 4440 wil also not be faster on the 4440. Only GLQuake and Quake II can take advantage of the 4440 right now. The architecture behind the Obsidian 100SB is like 2 Monster 3D with 4 MB texture and 4 MB frame buffer combined on one card, doing the now famous SLI. Add another two Texelfx chips with 4 MB each and you’ve got a 4440. If you’ve calculated correct you came to 16 MB RAM, 2 PixelFX and 2 TexelFX chips on the 4400 and 24 MB RAM with 2 PixelFX and 4 TexelFX chips on the 4440. Quite impressive, ain’t it? A Monster 3D has got one PixelFX, one TexelFX and 4 MB RAM.
If you should think that money is certainly an issue, but you still want to get the best possible performance and the best quality as well, then get yourself a add-on card with 3Dfx Voodoo chip. The best bet is currently the Canopus Pure3D or in Europe the Miro Highscore 3D, which are identical. These cards offer 6 MB RAM, 4 MB texture memory and a beautiful video out for NTSC as well as for PAL. Since the above cards are hard to get, you can still go for a Diamond Monster 3D or for one of the other 3Dfx Voodoo clones. Just be particular about the pass-through cable, it should be of high quality, especially if you’ve got a big monitor and should do some normal work on that monitor as well.
Now people that still think that a Voodoo is too expensive or who prefer to have only one card have got the option to decide between NVidia’s RIVA 128 or Rendition’s Verite 2×00 chip. If you want to play GLQuake or Quake 2 the decision is currently pretty easy, go for a Rendition Verite 2×00 card. If you have faith in NVidia’s ability to come up with a really decent OpenGL ICD soon, and if you want the best Direct3D performance currently available, then go for a RIVA card. The RIVA card that I recommend is Canopus’ Total3D 128V because of it’s nice video in/out feature and the good video capture software that comes with it. However, there’s still no AGP version available, which may be an issue to you. So if you really want an AGP card with the RIVA, I’d recommend the Diamond Viper V330, just because I’ve got more faith into Diamond’s drivers than in drivers of STB, Elsa or Asus. If your choice should be the Rendition chip, it really depends on what exactly you expect and how much you want to pay. The difference between the Verite 2100 and the V2200 is only the internal RAM DAC. The V2100 has only got a 170 MHz RAM DAC, the V2200 has got a 220 MHz RAMDAC, which enables higher refresh rates. The Diamond Stealth II S220 comes with the V2100 and it’s originally clocked with only 40 MHz. My Stealth II can be overclocked to 62.6 MHz and shows a frame rate increase of over 25% in the tested games. Clocked at this speed it’s exactly as fast as the Thriller3D from Hercules, which clocks its V2200 at 62.6 MHz by default. The advantage of the Thriller3D is the better overclockability, mine can be clocked up to 69.8 MHz, better refresh rates and video in/out as well as 3D shutter glasses output. That makes the Thriller3D much fancier. I personally like the 3D glasses output a lot, if you’ve ever tried these cool H3D glasses with a game that supports it, you won’t forget it. However the Thriller3D will be double the price of the Stealth II as it seems.
If you are still not sure what you want to buy, you’re probably someone who works on his computer professionally as well. For this case you may want to consider cards with the Permedia 2 from 3DLabs. This 3D chip wants some good CPU performance if you plan on playing games with it, but its professional OpenGL support and the excellent 2D performance are something that might be more important to you. As long as you don’t care for video in/out as offered by Elsa’s Winner/Office 2000, you should really go for the Diamond fire GL 1000 Pro. This card was giving the least trouble of all three Permedia 2 cards that I’ve tested. It comes with OpenGL support for 95 as well as NT, whilst the Elsa only supports OpenGL under NT and the Fire GL’s drivers are offering better performance and image quality. The Hercules Dynamite 3D/GL is mainly good for trouble right now. Repeated driver problems as well as image quality issues wouldn’t let me recommend this card at all.
So what’s with the ATI Rage Pro chip? I can’t help it, but this chip lacks in too many cases to be worth a recommendation. It does not support GLQuake’s or Quake II’s OpenGL engine, it has obvious problems with Direct3D, it has got only very weak support of professional OpenGL under NT, so that it doesn’t leave much else than it’s excellent 2D performance in combination with its video in/out features. However, if I want the best 2d performer with the best picture quality and the best RAM DAC I rather go for a Matrox Millennium II. The Rage Pro is neither fish nor meat, it’s no gamer’s card, but it’s also not really a professional card either. So what is it?