Thursday, August 19, 1999 - Hilbert Hagedoorn

V3 Vs. TNT2U

Ever since Orchid released the Righteous3d many moons ago there has been a gunfight to knock the Voodoo PC gaming platform off it's high horse. Through the years 3dfx has taken plenty of shots but have always remain standing for the next fight. Probably the most serious cowboy to raise it's weapon at Voodoo has been nVidia. Their TNT2:
Ultra card with it's "Voodoo-like" performance and added features have
tempted many 'a Voodoo-fan to switch sides, or at least consider it.

Mark Crowley wasn't sure which card to go for.... He knew the
Voodoo3's legendary speed and compatibility, but read about TNT2 Ultra's extra features and improved image quality. He HAD to find out for himself which card was best for HIM.

I found this report on the 3dfx Newsgroup and asked Mark permission to reprint in the Voodoo Review. I think you might find it interesting:


I can't tell you how many times I have gotten into "religious"
arguments about the TNT2 Ultra Vs. the Voodoo3 3000. Personally, I HATE to be misinformed about products because of someone's personal
bias enough, that I had to find out for myself, as objectively as
possible, which is REALLY better (This review was done for ME, after
all). I did try to make this little analysis as unbiased as possible,
and I hope that you take what I have here at face value, except where I noted "IMO" or "in my opinion".

I purposefully did not include any Frame rate numbers, because, in my opinion, a single number, or benchmark number cannot represent the overall performance of all games. The purpose of this test was primarily to satisfy my curiosity over the real difference playing the games that I play (kinda selfish, isn't it ;-), and in reality, benchmark numbers really do not do much justice to the feel of a game.

I have gone through and tested 3 different brands of TNT2 Ultra boards against a Voodoo3 3000. In order to be fair, I balanced the TNT2 cards by using the Nvidia Reference drivers (Detonator 1.88). (Note: I did test with the Creative "Unified" Beta drivers on their TNT2 Ultra card, and performance, particularly on Tribes was BETTER with that driver (albeit much slower than on the Voodoo3), but since every D3D game tested with that driver set was SOOO much slower, in relation to the Detonator 1.88 drivers - as much as 30%-, I did not go further).

First: To run these tests, I stripped my machine down to the bare minimums - I removed everything but the video card, and the sound card, and completely wiped and reinstalled Windows 98 for each card.

The system: -ABIT BX6-R2, P-III 504 (112X4.5), 256 Megs CAS2 SDRAM -19" Iiyama Vision Master Pro 450 monitor(Gorgeous) - 640x480@ 160Hz refresh, 800x600@140Hz refresh, 1024x768@120Hz refresh, 1280x1024@85Hz refresh, 1600x1200@75Hz refresh. -3 button Logitech PS/2 mouse(PS2 Rate@100). -Diamond Monster sound MX-300 (with latest posted Diamond Drivers). I used the NVIDIA Detonator 1.88 drivers on the TNT2, and
the Quake3 Compatible Voodoo3 drivers.

Clock speeds:
Voodoo3 3000: 166Mhz core, 166Mhz memory
TNT2 Ultra: 166Mhz core, 202Mhz memory
(Note: The Core/Memory speed ratio was kept the same as factory recommendation to be consistent)

Games Tested:
Starsiege Tribes V1.50
Unreal V2.25
Need for Speed III With DX6 drivers and Voodoo2 updated drivers.
Mechwarrior III V1.0
Quake2 V .93
Quake3Test
Half Life
Sports Car GT

Comparative analysis: 2D: Quality is marginally better on the Voodoo3 than on ANY TNT2 that I have tested. At my running resolution of 1152x864@100Hz Refresh, text stands out better, and is clearer with the V3. DirectDraw programs are a bit faster on the V3. BOTH are WORLDS better than the Banshee, or any other Video card that I have experienced, thus far.

3D: Clock for Clock in the same rendering mode on both cards (16 bit) the Voodoo3 is about 10% faster with everything but Q3Test, Unreal, and Tribes. In Q3test, the performance nod goes to the TNT2, by maybe 1% (Almost impossible to tell a difference in speed). In the TNT2's 32 bit mode, there was a noticeable drop off in smoothness, which was alleviated by a drop in resolution. Tribes was nearly unplayable at
any resolution on the TNT2, which indicates a technical problem.
Unreal's performance also showed a marked drop off in overall
smoothness, and is very jerky, which also indicates a technical
problem.

On Quake2 Engine games, the performance is indistinguishable from card to card (The TNT2@ 125 MHz would run these games well enough for most)@800x600. At the higher resolutions in Half life, there was a bigger drop off in performance with the TNT2.

Games that run Glide Natively (Unreal, and Need for speed 3) tend to have a significant performance improvement. With most of the games that run with the Glide interface, I was able to bump up one resolution (from 640x480 to 800x600, or 800x600 to 1024x768) in Glide mode over D3D mode, and retain full smoothness. In Need for speed3, the smoothness of operation at 640x480 and 800x600 is not possible.
Some say that this result is caused by a problem with the D3D driver in EA Games (most of their 3D enabled games use the same driver structure, and are often interchangeable from game to game) or is the Glide driver just that good? My experience with other, NON EA games (Tribes and Unreal) points towards the latter.

Installation: This is a no-brainer for either card. The installation
with either driver was completely flawless. Control panel operation was painless with either card. One interesting point - As there are no custom drivers available for the 3DFX cards (as there are not several manufacturers vying for your dollars) a 3DFX owner ALWAYS uses reference drivers. With the TNT2 cards, ALL of the drivers that came with the cards, and all of the drivers that are available from the different manufacturer s web sites, were all back rev. As the earlier rev drivers did not perform as well as the latest Nvidia reference drivers (Detonator 1.88), I ended up using the Nvidia reference
drivers for my tests. This has the side effect of disabling the extra functions of the cards, such as TV out, and performance sliders. This was NEVER the case with the 3DFX cards.

Compatibility: I found more games to run and run WELL with the 3DFX product. D3D compatibility was VERY good on either card, and I never experienced any problems, on either product. Since the TNT2 does not support Glide (Except with the Creative Labs Unified drivers, but Creative Labs Glide is not native, nor is as compatible or as fast as the real thing), there is a real problem with the Unreal, and with Starsiege Tribes. I expect to see the problems resolved (to some extent) in future releases of those respective games, but, for now, prepare for disappointment. In most cases, Game titles that support
Glide offer better performance and visuals under Glide than they do under OpenGL and Direct3D. Case in point: The Unreal D3D and OpenGL both exhibit a visible mip mapping error, where the Glide driver does not. I do not expect this situation to change).

Overclockability: In the tests, I over clocked all of the TNT2 Ultras to match the Voodoo3's default clock rate, to keep the tests fair. In fact, the TNT2 ultra, on many vendor's boards, is clocked anywhere between 150MHz and 175Mhz. Of course, there are variances from chip to chip and card to card. Personally, I have over clocked a Voodoo3
2000, and 3000, a TNT2 Ultra with 5ns SDRAM, a TNT2 Ultra with 5.5ns SGRAM, and a TNT2 Ultra with 5.5ns SDRAM. Out of the box, I can get the both the Voodoo3s to 170Mhz, and with cooling, the 2000 gets to 175Mhz, and the 3000 gets to 185 MHz. Both Vooodoo3s have 6ns SDRAM, so I can realistically attribute the difference to the Tennmax Stealth Cooler (Available from: <http://www.tennmax.com/>) that I put on the Voodoo3 3000 (The 2000 has a glued-on heatsink that I screwed a fan on, is the same fashion as with processor heat sinks). All of the TNT2s were able to have their memory clocked to 210Mhz, with the Creative card (The 5ns memory card) being able to run memory at 220Mhz. I couldn't get any of the TNT2 ultras that I tested up past 170MHz reliably, no matter what heatsink/fan combination used on them
(Stock Cooling, along with add on fans and coolers, and I tried the Tennmax on all three).

Note: For my comparative performance tests, remember that I did clock the cards to equal the base clock speed(and memory speed) of the Voodoo3 3000, 166 MHz (Memory on the TNT2s was set to the factory recommended core speed/memory speed ratio, so came out to 202 MHz)

Quality Analysis (My opinions):

(Note: In this comparison, I will try to describe, to the best of my ability, what it FEELS like and LOOKS like, playing the games on both cards. Since the performance of Tribes and Unreal are more indicative of technical problems between the games and the TNT2, I did not provide any performance opinions on them)

The Voodoo3 is noticeably smoother in all of the games (including Q3 Arena). I noticed a number of "Hitches" in the game play in Q3 Arena with the TNT2, while other times the game was smooth. In the games that run well with both cards (Not Tribes, and Unreal), the Voodoo3 seems to run smoother at the same resolutions than the TNT2.

Visual Quality: I have been in so many arguments about this one. The    problem is, that an estimation of visual quality is completely up to the user's point of view. For some, the small things are vitally important to their personal immersion into the game, and for others, those same things are unnoticeable. I will admit that from my perspective, frame rate, and smooth game play is everything. Others will tell you that image quality is everything! Both of us are "right. However, for this review, I will try to describe what I saw out "coloring" it with my point of view.

16 Bit mode: Winner - Voodoo3. the 22 bit post filtering (which has been ON for the whole test) really helps here. Blending error (Banding on Light flares, "cheese-cloth" effect on smoke trails) is actually WORSE on the TNT2, as the Voodoo3 makes an attempt to dither around the lost pixels while the TNT2 does not. There is evidence of the "Green"(Note: in 16 bit rendering, the Green channel has one more bit per pixel than both the red and the blue, which is why when a blend reduces the intensity of the pixel, and therefore the value in each channel, the lowest value ends up being 0 for red, 0 for blue, and 1 for green, causing the pixels to look green before they
disappear.) effect with both cards, but it is more pronounced on the
TNT2. Overall, the colors on the Voodoo3 and the TNT2 are similarly saturated. A subtle difference that I noticed was that the TNT2's mip map divisions were more noticeable than on the Voodoo3. This is best noticed on Level2 of Quake3 arena when bouncing on the center pad, and looking down. What you will see is a light "shimmer" effect, when the graphics chip is changing bitmaps as you are coming and going. The Voodoo3, In my opinion, used a better blending technique than the
TNT2. However, I believe this to be more a decision made in driver development than in hardware development.

32 bit mode: Winner - TNT2 .Since the Voodoo3 does not have a 32 bit mode, I will attempt to discuss the differences between the TNT2's 32 bit mode and it's 16 bit mode. Since the only game that I have played that shows an easily identifiable difference with 32 bit operation is Q3 Arena, I will attempt to describe the differences - With 32 bit rendering and 16 bit textures, the performance hit was noticeable, but not huge, especially at lower resolutions. The visual benefit was
mostly limited to smoke trails, in that there was no cheese-cloth effect present. With 32 bit rendering and 32 bit textures, the performance hit was much more severe. So much so, that the game had to be reduced in resolution to keep the frame rate up in the "smooth" range. The difference with 32 bit textures was present throughout the levels, with more saturated colors, and a quite noticeable smoothing
effect (The textures looked a bit less like a game and more like real bitmaps). However, the change in screen resolution caused a bigger visual quality loss, than the improvement in detail provided with the 32 bit textures. On the other games, there is so little difference that I have trouble believing that there *IS* any real difference (except for the noticeable drop with frame rate).

The bottom line: A Voodoo3 2000, will outperform, in any mode, a regularly clocked TNT2 card. In order to get the same performance level, you will really need to get a standard clock rate TNT2 Ultra card, which will set you back at least $200.00. The Voodoo3, in my experience, is more overclockable, and is, in my opinion, generally more satisfying to play on than any of the TNT2s that I have tested (I guess that the TRIBES problem bummed me out more than I thought).
Since there are going to be paradigm-changing products coming out next year (products that are going to change how we play games on the PC) I would recommend that you spend as little as possible on your replacement card a possible to save for these cards. Therefore then, I would have to recommend the Voodoo 3 2000 card. It provides the best bang for the buck, and should easily hold you to the end of the year.

Voodoo Magazine , July 4, 1999.