provided by: 
Originally published at Internet.com 
STB Velocity V4400
Hercules Terminator Beast SC
Diamond Monster Fusion
Diamond Viper V550 The buzz these days in 3D graphics cards is the Nvidia RivaTNT2 Ultra, the 3dfx Voodoo3 3500TV, the Matrox Millenium G400 MAX, and to a lesser extent, the S3 Savage4. There is even a lot of talk about the future - the Nvidia NV10, the 3dfx Voodoo4, and the ever-anticipated, often-doubted Bitboys Oy Glaze3D. Yet, in all of this remains one constant: money, and lots of it. Almost no TNT2 Ultra, Voodoo3 3500TV and G400 MAX can be bought for less than $200, putting this out of reach for many people on a tighter budget. So where can all these people turn, scorned by the prices of the latest and the greatest? To those cards everyone has now forgotten - the RivaTNT, the Savage3D, and the Voodoo Banshee.
In this roundup, 4 cards were put through the paces on two different systems, with 5 different CPU/CPU speeds being used, although two extra Celeron speeds were added to the CPU Scaling graphs to round out the numbers. The 4 cards reviewed are the Diamond Viper V550 and the STB Velocity V4400, both powered by the RivaTNT, the Diamond Monster Fusion using the Voodoo Banshee, and the Hercules Terminator Beast SuperCharged (SC) with select Savage3D chips that were clocked at 120 MHz, 20 MHz above the original Savage3D's clock speed. All four of these cards are now sold for under $70, making them excellent contenders for consideration in a budget PC system.
Feature Comparison
Monster Fusion AGP Viper V550 AGP Velocity V4400 AGP Terminator Beast SuperCharged Chipset 3dfx Voodoo Banshee Nvidia RivaTNT Nvidia RivaTNT S3 Savage3D RAM 16MB 125MHz SDRAM 16MB 125MHz SDRAM 16MB "High-Speed" SDRAM 16MB 120MHz SDRAM RAMDAC 250MHz 250MHz 250MHz 250MHz AGP 1x 2x 2x 2x Composite/ SVideo Out No/No Yes (with cable)/Yes Yes/Yes Yes/Yes MPEG1/2 acceleration Yes Yes Yes Yes OpenGL ICD Not Completed Yes Yes Yes Glide Support Yes No No No S3TC Support No No No Yes Included Software Driver CD, MotoCross Madness (full), Unreal(11 levels), Zoran SoftDVD Driver CD, MotorHead (full), Zoran SoftDVD, Fremont3D, MS Game Sampler, Photosuite SE, VRCreator Driver CD, Forsaken (full), Digital Video Producer, Web 3D, 3D Sensations, 3Deep Driver CD
Supported Refresh Rates
bits-per- pixel (bpp) Monster Fusion AGP Viper V550 AGP Velocity V4400 AGP Terminator Beast SuperCharged 640x480 8 60-200 60-200 60-160 60-160 16 60-200 60-200 60-160 60-160 32* 60-200 60-200 60-160 60-160 800x600 8 60-200 60-200 56-140 56-160 16 60-200 60-200 56-140 56-160 32* 60-200 60-200 56-140 56-160 1024x768 8 60-160 60-170 60-120 60-130 16 60-160 60-170 60-120 60-130 32* 60-160 60-170 60-120 60-85 1152x864 8 60-160 60-120 60-85
Installation
With Win98, each card's setup was a breeze. Plug in the card, let Win98 detect it, insert the driver CD and install the drivers, reboot, and away we go. Installation went without a hitch for all Win98 installations, regardless of motherboard, CPU, or what else was in the system, even for Super7 motherboards. The Terminator Beast SC didn't quite live up to it's promise for refresh rates - however. It would never do better than 85 Hz at 1024x768 and was a dismal 60 Hz at 1280x1024. Using the latest drivers from Hercules didn't help at all. This was the case both with Win98 and NT.
However with the Monster Fusion, Windows NT4 wasn't quite so forgiving. At first, it worked fine in Windows NT. But when it came time to install it in the testbeds for benchmarking, NT wouldn't cooperate. The drivers would be installed, which required reboot, but upon bootup, the system would still be using the VGA driver. Nothing I tried to do would work: I used the 3dfx Reference drivers, and I even tried Xenomorph's NT4 .inf file. Everything responded the same. My only guess as to why this happened would be the fact that it worked fine on a SP3 installation, but the machines I tried installing it on were using SP4 and SP5.
With the Velocity V4400, I had the same problem one time, but after installing the Nvidia Reference Drivers with the Viper V550, I was able to swap that out for the V4400 and everything worked fine. It seems odd this would happen only once and could point to a problem with the installation on the testbeds themselves, not with the cards or their drivers.
OpenGL was where things really got interesting. Since many people are already playing Q3Test as an actual lifestyle, one can only imagine how popular the final game will be. So, I made it a point to do my absolute best to get each card working with Q3Test 1.08. The Viper V550, Velocity V4400, and Terminator Beast SC all worked on the first try. Their ICDs were mature enough to handle Quake3 without problems, although the Terminator Beast was noticibly slower (more on that later). Once again, it was the Monster Fusion having the problems. The Monster Fusion drivers only include a MiniGL, not a full OpenGL ICD. Even the latest drivers from Diamond's website don't include an OpenGL ICD. The 3dfx Reference ICD drivers were weak at best, and could be termed "Quake3-only" drivers. This wasn't good enough so once again, I turned to Xenomorph. He created a set of drivers which were basically the latest 3dfx Reference drivers, but also included the latest Voodoo3 ICD from the Voodoo3 drivers.
With these drivers, I was able to get Quake3 to run and 800x600 was a playable resolution. However, 640x480 was still quite a bit smoother, and for those that prefer smoothness over looks, that will likely be your resolution of choice. There were some occassional graphical hiccups with Q3Test, but I don't know whether to blame the drivers or the game, which is still in beta. My favorite and most common glitch occured when the game would start up in what looked like CGA color, though the resolution would still be 800x600. It was like playing Q3 on a great system, while having a CGA color monitor hooked up to your PC. Quite entertaining at first, but quite annoying after the euphoric effects wore off. I discovered that the way to fix it was simply to close Q3Test, restart it, and continue opening/restarting until it opened properly. It usually did it on the first restart, and never took more than 3 times.
Stability
For the most part, the cards were completely stable in both Windows NT4 and Win98. However, in the Super7 testbed (with the Epox MVP3C-M motherboard), both RivaTNT cards were less stable. It seemed to be a driver-related issue. When either the Viper V550 or the V4400 used the Detonator 1.10 drivers, stability was fine, but when they used the newest Detonator 2.08 drivers, they were faster, but less stable. For the Quake II benchmarks later on, I used the Detonator 1.10 drivers for the Viper V550, but the 2.08 drivers for the V4400. The Viper just couldn't get through any benchmarks without crashing using the 2.08 drivers. The Velocity was stable enough to get through a 3D benchmark or two, but would crash after the 3rd or 4th one. I checked PowerTweak to make sure it wasn't a clock issue, and both were cranking away at 90 MHz. The board was also configured at 2/3 speed for AGP. I can't explain it, but it could account for the reason why the Velocity was significantly faster than the Viper in Forsaken.
The Terminator Beast SC is extremely sensitive to AGP speed. It wasn't even completely stable at 112MHz FSB, or even 75 MHz AGP bus speed. Nothing in 3D would run at 124 MHz FSB for longer than 1 second before freezing. Obviously, for overclockers this is definitely not the card to get. It will severely limit the FSB choices.
Image Quality
The RivaTNT cards had the best image quality of the bunch in both 2D and 3D. Clean, clear, and just plain good. At 16-bits-per-pixel (bpp), I couldn't notice any difference between the cards. However, 32-bpp was noticibly cleaner on the RivaTNTs vs. the Terminator Beast SC. The Monster Fusion was limited to 24-bpp in 2D and 16-bpp in 3D. The 3D limitation was quite a bit more noticeable than the 2D limitation. In 2D, it was just about as good as the RivaTNT in quality and is a card I would be happy with in my own system. However, the Savage3D is severely limited by its lower refresh rates, making 1024x768 the lowest endurable resolution, and 1280x1024 absolute torture.
The Savage3D has the added advantage of S3TC, which can have the effect of accelerating memory transfers or allowing for extremely large textures to be stored in the onboard RAM. By far, the most pleasing of these is the extremely large textures. Unfortunately, all my attempts at getting the Unreal S3TC level to load were unsuccessful, resulting in immediate crashes in the drivers. Since I had no other S3TC games, I could not provide screenshots of S3TC. However, there are screenshots on the S3 website demo-ing S3TC. 
Viper V550 Forsaken 1024x768 
Viper V550 Descent3 1024x768 D3D 
Terminator Beast SC Forsaken 1024x768 
Terminator Beast SC Descent3 1024x768 D3D 
Monster Fusion Forsaken 1024x768 
Monster Fusion Descent3 1024x768 Glide
Testbed Configurations
Slot1 Testbed * Pentium III 450 @ 558 MHz * Celeron 300A @ 450 and 504 MHz * Shuttle HOT-641 Dual CPU motherboard * Corsair 128MB CAS PC100 SDRAM * Quantum Fireball EX 13GB ATA/33 * 12x IDE CD-ROM * Generic 10/100 NIC * Windows 98 retail * Windows NT4 SP5
Super7 Testbed * K6-2 350 @ 300 and 400 MHz * Epox MVP3C-M motherboard * 64MB PC100 SDRAM * Quantum Fireball 3.2GB ATA/33 * SoundBlaster AWE32 * 8x IDE CD-ROM * Generic 10/100 NIC * Windows 98 retail * Windows NT4 SP5
The Viper V550 and Velocity V4400 were benchmarked using the Detonator 1.10 drivers. However, in Forsaken, the Velocity was tested with the Detonator 2.08 drivers to show the difference between the 1.10 and the new 2.08 drivers.
The Monster Fusion was tested using the latest drivers off the Diamond website. The NT drivers came from the Driver CD.
The Terminator Beast SC was tested using the latest Win98 and NT4 drivers from the Hercules website.
How the Tests were Run
There was nothing running in any of the tests except Norton Crashguard and Norton Antivirus Auto-Protect (these things are tough to keep from loading at startup), but they were running in the background on every test.
All 2D tests were run in 32-bit color, except on the Monster Fusion which is limited to 24-bit color. The cards were run at 85Hz refresh rate in every resolution, except on the Savage3D which was limited to 75 Hz at 1024x768 and 60 Hz at 1280x1024.
All 3D tests were run at 16-bpp color with VSync OFF.
Winstone99 Windows98 Business Winmarks
As you can see, the Savage3D is significantly slower at 1280x1024 than at 1024x768, but still lags behind the pack. Both the Banshee and RivaTNT handled well in the K6-2 tests, albeit gaining little when moving up 100 MHz from 300 to 400 MHz.
WinBench99 Windows98 Business Graphics Winmarks
This test shows just how bad the Terminator Beast is in graphics applications, especially on the K6-2. Even with the 24 vs. 32bpp advantage that the Monster Fusion has, it still can't keep up with the two TNT cards on the K6-2. The Slot1 side is a better race between the Voodoo Banshee and the RivaTNT.
WinBench99 Windows98 High-End Graphics Winmarks
The Terminator Beast makes a valiant effort on the Slot1 CPUs, but dies in the water in the Super7 testbed. The Monster Fusion performs admirably, but is outshown overall by the two TNTs.
Winstone99 NT4 Business Winmarks
The Terminator Beast has a better showing in NT4 on the K6-2. I suppose by now, we have become used to watching it descend like a rock in 1280x1024.
WinBench99 NT4 Business Graphics Winmarks
WinBench99 NT4 High-End Graphics Winmarks
Overall, the Terminator Beast performs quite a bit better in NT4 than Windows 98, but it's still a ways back in the pack, especially on fast CPUs and K6-2s. I apologize for not being able to provide Monster Fusion results, but the board just wouldn't cooperate.
QuakeII 640x480
I'm not quite sure why the Velocity trounced in the K6-2 tests. I benchmarked them over and over, and the results were the same. Same drivers as the Viper, same everything. However, it was much more unstable than the Viper - which makes me wonder about clock rate. PowerTweak reported 90 MHz on both cards. Not that I have reason to doubt it except for the vast difference between two very similar TNT cards.
Other than that, the Banshee performed admirably on slower CPUs until it hit the Crusher benchmark. It seems a bit odd since the crusher.dm2 benchmark is considered more of a CPU benchmark. The TNT cards began showing their power at 450 MHz and above, leaving the rest in the dust.
QuakeII 800x600
Again, the STB card kicks tail on the K6-2, but it was quite unstable, crashing ever 3rd timedemo or so. So take those numbers with a grain of salt. The Viper V550 is more likely to be a stable setting for the RivaTNT for whatever reason.
QuakeII 1024x768
The Slot1 benchmarks are pretty predictable, but the K6-2 benchmarks get funky again (noticing a pattern yet? :) ). This is the section where none of these cards have the fill rate to keep up with the CPU as long as it is a 400 MHz or better CPU. If you want to run faster than 30fps at 1024x768, don't get these cards. You could have a 1GHz CPU and the benchmarks wouldn't go up much. And the stench of the Terminator Beast is just awful - not pulling above 15fps at all, dorpping under 10fps on crusher.dm2 on both K6-2s.
QuakeII 1152x864
The 3dfx MiniGL simply takes higher resolution and renders only the portion of the sceen that would fit in 1024x768, rendering benchmarks any higher inaccurate. The Terminator Beast did a great .2 fps (yes, that's point 2) on the first demo1.dm2 on the P3 558 so I didn't bother with the rest for it. So, that makes this into a RivaTNT head-to-head, and it's pretty boring to see two lines right next to each other for 5 graphs, so I cut the fat and put in the fastest CPU and the Celeron 300A. Not too suprisingly, the numbers are nearly identical, only taking a slight hit on the C300A on the crusher.dm2 benchmark.
Direct3D: Forsaken
The Forsaken marks caught me off-gaurd. At 1024x768 on all three Slot1 testbeds, the Terminator Beast pulled a worst to first move. Granted, the difference isn't that much, but on the Celeron 300A, it's quite a bit - better than 10fps. The two TNT cards simply can't keep up with either the Banshee or the Savage3D throughout much of it.
Direct3D: 3DMark99
This benchmark was an interesting one to watch. The one where the AGP implementation came into play were the 16 and 32 MB texture sizes. The RivaTNTs and the Terminator Beast SC were quite smooth rotating a 16MB texture, but the movement on the Monster Fusion was very jerky with a slow frame rate. On the 32MB texture, the Terminator Beast SC started doing much the same except it was more a delay than anything and it wasn't as painful as watching the Monster Fusion run through the test.
The performance of the Terminator Beast was surprising on the K6-2 was quite surprising, as was the numbers on the faster CPUs, nearly matching the TNTs at 450-504 MHz across the Celeron and P3.
CPU Scaling: Direct3D
For the CPU Scaling tests, I chose 800x600 because it was the resolution most likely to be used by a gamer. Although Forsaken frame rates have always been amazingly high, the graph of them points out the scaling across CPUs very well. I chose the QuakeII massive1.dm2 benchmark because it is the most likely to be real-world gaming. It doesn't depend on fill-rate like demo1.dm2 or CPU like crusher.dm2 does. 
Of all the cards, the one that scales the nicest in Direct3D is the Monster Fusion. It starts out near the top and finishes near the top. The Terminator Beast is terrible on the K6-2, but kicks tail on the P3 450.
CPU Scaling: OpenGL
These are pretty similar except the very lackluster performance of the Terminator Beast SC. It's pretty clear the OpenGL ICD in the Hercules drivers needs some work before it can be competitive with other low-cost video cards. However, with support like it is, don't expect such drivers anytime soon, or ever.
Conclusion
Supply anb demand dictate prices in a free-market economy. The more demand, the higher the prices, unless there is too much supply. Conversely, the fewer demand, the lower the prices, unless there is too limited supply. Simple supply and demand economics.
There has been such little demand for any of these cards recently that quite frequently they can be had at great prices. There have been a great many produced and out on the market. Even though stocks are now low in these cards, they can often be found at some great prices with vendors trying to clear shelves to stock items that are in greater demand. If you are looking for a video card, don't ignore any of these products, and now you have their features and limitations in mind.
At the top of the list are definitely the two RivaTNT cards. It's tough to make a choice, but if I were making it, I'd probably go for the Viper. I always liked MotorHead better than Forsaken. They are so close in most performance tests that it doesn't make much of a difference, especially for Slot1 system owners. Neither STB, which no longer exists, or Diamond can be expected to put out any new drivers for these two items. Because of this, that leaves the Nvidia reference drivers and eliminates any advantages one might have over the other.
Coming in a close third is the Monster Fusion. It's lacking on the feature set, but it's always close to the top in performance. WIth the OpenGL implementation lacking a bit, it has it's drawbacks, but it still performs competitively and is worth considering.
The Hercules Terminator Beast Supercharged: What can I say, but that I was disappointed for the most part. The OpenGL ICD should have much better performance, but it was quite lacking. The Direct3D performance was quite good, but the card lacked higher resolutions in both 2D and 3D. I recommend only considering this card if you have a 17" monitor or less and prefer the 1024x768 resolution or less.
For the value shopper, the RivaTNT is the best chipset, but the Voodoo Banshee shouldn't be ignored, especially for those that have several Glide games. ---------------------------------------------------------------------
Author: Eric Grevstad
Read article at Internet.com site