Video Card Review

Voodoo 5 vs Nvidia Geforce 2

(Nvidia’s Geforce 2 chipset)

The Shoot-out at the Pink Flamingo Bar & Grill…!!!

Screenshots

I have been frustrated with Video Card reviews for quite sometime now and have been thinking that I could do better. The basis for that belief was that I knew that I would approach this like a gamer, indeed this review was somewhat delayed because I rediscovered an old friend in the form of a terrific Racing Sim, Grand Prix Legends. When it was time to either suffer thru benchmarks or attempt a new personal record at the Nurburgring, the Lap record always won, sorry. Gamers have a different set of priorities than the typical hardware reviewer and I hope that my humble effort at least begins to bridge that gap. Ultimately the decision to do this comparison was born from the frustration of reading other “reviews” that did basically two things fairly consistently. 

1. Spew forth a lot of techno-geek-jargon (which is useful when the right person is spewing it, witness Beyond3D.com* and Reverend) in an attempt to build credibility with the readership. The two sites mentioned Beyond3D and Reverend are excellent sources of information.*
2. Test Quake 3 and 3D Mark and call it a day. 

From a reviewers standpoint pretty cool since you can get that kind of review out in a couple of hours which means out before other sites and without stealing too many resources away from the other projects that keep the news flowing on whatever site is doing the review. From a player standpoint though it’s pretty useless unless you play a lot of Quake and 3D-Mark. Sadly 3D-Mark doesn’t do snap rolls and so I don’t play it… (inside joke amongst flight simmers)

To me a review is not a review unless it’s a test with a bigger variety of games, at least 8 current ones. Also while interesting and somewhat useful for tuning a computer, artificial benchmarks such as 3D-Mark are useless to me as a buyer. I want to know directly how a card or piece of hardware will affect my game playing experience and the only way to know that is to test it on games that people actually play. I am not a Techno-Uber-Geek…I’m not going inflict you with Super Neato Spanko Technical terms to dazzle and amaze your friends and relatives. What I am going do is to approach this like a gamer and have these cards show me what they can do with actual games. My goal for this is to enable you to make an informed decision about which card, based on what you see here, to be the better match for you. I well understand your stress about this decision cause I have felt it before, the hunger for some valuable bits of information to ease your purchase decision. This effort of mine is my humble attempt at changing the way reviews are conducted for video cards.  For me at least the days of accepting abbreviated reviews when making a decision on where to spend $300+ dollars are over, no more 2 application benchmarks and call it a day. We deserve better.

Out of Box Experiences 
Creative Annihilator 2 32mb Geforce 
vs 
3dfx’s Voodoo 5-5500 

Screenshots

Both cards were tested on a freshly formatted Hardrive using Win98se. I have a Asus P3V4X Via Apollo 133a motherboard with a P-IIIe running at 135×5.5 =742mhz with 128mb of Mushkin CA-2 SDRAM and a Vortex 1 Aureal soundcard. Win98se/Dx7a/Via 4.23 drivers were installed on freshly formatted disks for each cards install to eliminate any problems with confused drivers and installations. For the portion of the benchmarks using a 366 I simply turned my Front Side Bus down to 66mhz. I realize that this will not exactly simulate a 366 but it was as close as my limited (0$) budget could afford.

The initial hardware installation on the Geforce 2 was un-dramatic to the extreme given its very small size. The Voodoo 5 was on the other end of the scale with a length of 10.5” it required some forethought prior to installation. I had to move 2 hardrives for my installation and was thanking whatever stroke of good fortune had me in possession of an Enlight case with slide out drawers for my hardrives. Measure from the back of your case to the first obstruction along the line of your AGP slot, if you have 10.5 inches you should be good to go. Anything less means moving stuff around.

Before I started this review one of my goals was to test exactly what came in the box. I felt that a video card that costs around $300.00 should ship with drivers that work straight off the CD. This proved to be naïve and in the end I was forced to use the Nvidia Reference drivers ver. 5.22 for the Geforce 2. The drivers that shipped with the Creative Annihilator 2 32mb card had so many texture problems that testing was out of the question. Oddly enough, the game most used by reviewers to test Video Cards worked perfectly with the Creative drivers, Quake 3. Also the Creative drivers did not support the Full Screen Anti Aliasing (FSAA in the rest of this review) as claimed on the box, for that support later drivers were needed. For FSAA support I had to go to the 5.22 drivers but even those did not feature what I consider functional FSAA in OpenGL, as it was only supported at the most minimum setting. Considering that 3dfx’s V5 minimum setting was x2 I refused to test them against each other in OpenGL apps. By moving up to Beta Drivers Version 5.30 I could have gotten more settings in OpenGL by either getting a 3rd Party Tweaking program or by editing the registry myself. I consider editing the registry to be a asking alot of a customer spending $300.00 dollars and besides I didn’t want to get to far into testing Beta Drivers. For the 3dfx card I did not have to test any drivers except the ones in the box and considering this is a shootout between these two cards I thought I had already bent the rules in favor of Nvidia enough. Farther down in the benchmarks you will see that I did bend the rules just a bit more in favor of the Nvidia by testing the 5.30 Beta Reference drivers, but even those had problems. 

The 3dfx Voodoo 5-5500 ran straight out of the box with minimum difficulties with the drivers that came with the card. Probably some kind of record but then again when you consider the prices of these two cards I prefer thinking that I deserve functioning drivers straight out of the box. All features claimed on the box were working and overall stability was excellent. The 3dfx drivers did have some problems with the most obvious being a small white line that displays around the edge of the screen in some glide games. There have also been reports and I have experienced this of the screen resolution not resetting to the proper mode when exiting a game. But perhaps the worst bug in the driver for the 3dfx Voodoo 5 is the cards problems in D3D at 1600×1200 where it appears to be running on only one of its VSA-100 chips. I believe that I confirmed this bug using 3dMark Max where the Single Chip and Fastest Chip modes returned exactly the same scores in all area’s. Which would seem to indicate that above a certain resolution the driver shuts one chip down in Direct 3D. The Glide scores do not reflect this problem so perhaps its driver related. 3dfx has posted that they have repeated this bug in the lab and are currently working on a fix. Overall the driver stability of the Voodoo was superior with no real shenanigans to prevent you from playing. 

While I am on the subject of drivers it must be mentioned that the 3dfx Drivers came with an extensive and easy to use help application. It is a right click context sensitive application that gives you two choices in the degree of explanation offered, from “in depth” to “summary” very nice. Also worthy of note is that they did not shy away from explaining what each of the settings accomplished in FSAA. The Geforce 2 would have benefited tremendously from even a basic explanation of the different settings of FSAA. As it was there was no help file for the Geforce explaining the FSAA settings and any mention of editing the registry for modifying OpenGL settings certainly was not to be found.  The lack of help files shows a rushed product and it was the first sign that perhaps Nvidia was hurting themselves by being locked into a 6 month product release cycle.

TnL vs Full Scene Anti Aliasing…..that’s where the fight is…!!!

Screenshots

The toughest choice between these two cards for consumers is deciding what’s the more important feature.  Hardware transform and lighting with Nvidia’s Graphics Processing Unit which promises to make our games more compelling with any combination of these spanky new features:

1. Increased lights which already makes MDK 2 look real fine…you can see it with another card but without TnL the framerate hit is severe.
2. More polygons (rounder curves)…there are a couple maybe more Quake 3 levels that show whats possible with that feature.
3. Increased speed by taking  a load off the CPU which will be excellent for games with alot of Artificial Intelligence (AI) letting the CPU take care of the AI and the Video Card take care of the graphics.

All very neat things to do but sadly all of them require some type of developer support and that’s not due to hit the shelves in any appreciable manner till sometime during Christmas or right after.  Though Nvidia is using its TnL unit to do the FSAA that its new drivers are enabling. Or do you want to go with….

3dfx’s Full Scene Anti Aliasing that supports all games released that use D3D, OpenGl and Glide. Doesn’t require developer support so that you don’t have to worry about that support disappearing if the developer gets caught between the budget knife and additional features. Full Scene Anti Aliasing is a dramatic experience that mere screenshots really cannot capture. Sure I can (and will) show you screen shots where the edges of objects are beautiful and clean….but that’s not really the big deal about FSAA. Nope not even close to the big deal about FSAA, the BIG ASS deal about FSAA is no more creepy crawly textures, no more popping pixels not more scenery that’s alive…the hills may be alive with the Sound of Music but I want my hills to stay down and not move around…I will never play another game without my FSAA.  

Its a completely false and disingenuous statement to proclaim as some Nvidia company reps have done, that high resolutions obviate the need for Full Scene Anti Aliasing. High resolutions as high as 1600×1200 still exhibit jaggies (where straight lines exhibit jagged edges), but its the crawling textures and pixel popping that really catch the eye. Even in a sim as clean and uncluttered as Enemy Engaged Comanche Hokum these graphic artifacts can be seen.  The interview with the Nvidia reps can be read here. Ironically at the same time that the press guys are saying its not required and that high resolutions will do a better job of improving image quality the Nvidia engineers have been releasing driver release after driver release improving their version of FSAA. I guess perhaps its not a good feature until Nvidia can compete….

High resolutions are lovely especially in a flight sim but as you can see in the benchmarks its completely possible to have your cake and eat it too. High Resolutions and x4 FSAA is the only way to go…heck its so good it should motivate people to go out and buy faster processors to run every game that way.

Benchmarks

Screenshots

Saying that you are going to include more benchmarks from more games means having to first discover which games/sims that are amenable to benchmarking. This means having a method to display the frames per second then discovering a mission/save game/path that provides a consistent framerate result given the same machine settings and software settings. Measuring the framerate over the course of some time instead of taking an instant reading is important and in addition it’s crucial to the entire test to get repeatable results when none of the machine and software parameters have changed.

Its been said that the benchmarks are not valuable since many of the games I play are extremely CPU intensive. This to me is an interesting discussion but seemingly misses the point. When you test on the same machine only varying the Video Cards you will see differences in framerates. They may be smaller than if we only tested Q3 but the differences will still be there. Its also said that you must test with a Video Card specific Benchmark such as 3dMark so that you can see how the cards will run with the Future Gaming engines. My answer to that is if I want to guess which gaming engine is gonna be around next year the best places to go are the guys making the games….and the games they are making now. I like the Mad Onion guys but only using their benchmark gives a very distorted view of exactly how these cards will perform on your computer. How these cards run with the games we have right now is the best indication of how these games will run with the majority of games you are gonna play. Sure TnL enabled games can change the equation somewhat but since we don’t know how that will be implemented or when its pretty optimistic to be buying a card for it. Don’t let anyone blow smoke in your eye’s, artificial benchmarks are just that….artificial. For the best example of that, consider that 3dMark 2000 one of the most widely used benchmarks predicted that the Geforce 2 would be quite a bit faster than the Voodoo 5, only problem on the games we run it isn’t, this is a crucial bit of knowledge and we are being sold out to expediency by review sites that only use two benchmarks. But the truth is doing it the way I have done it is extremely labor intensive…in a fashion that I don’t think anyone wants to emulate. That’s cool….I will do this again when the need arises.

The actual method in gathering framerate information was to produce a save game that had a minimum of AI interaction since I had discovered that having Enemy AI can cause framerate to vary fairly widely. It is also important obviously to have a consistent path that the unit would take, whether it would be a human/plane or tank. The game settings would be pushed to maximum except where noted on the benchmarks. Then I would start the mission/save game enabling the framerate counter. Wherever the game did not have a native counter I would substitute a neat program called FRAPS which does an outstanding job of measuring instant framerate. At which point I would allow 5 seconds to elapse to stabilize the mission then I would begin to record framerates every 3 seconds thru a mission. It was important to the repeatability to end it at the same point so each mission was designed to have a recognizable point to end and that mission had to have been long enough to produce at least 20 samples.  I ran most of the tests three times so that I could test for repeatability then I would average the result. I believe that the minimum of 60 samples in 3 missions was a good minimum to assure repeatability of the final framerate number. Some of the tests ran this way were identical and so 3 runs were not needed in every resolution…Janes F/A-18 was one such.  Naturally this method wasn’t required on all the benchmarks such as Falcon and Quake 3.

All benchmarks are run with Max Details selected unless otherwise noted. The format for the report is resolution/FSAA setting/color depth. The Nvidia and Voodoo cards are very different in regards to FSAA and it should be noted that after comparing these two cards directly against each other using the same exact missions in two flight sims Falcon and Flanker I have the opinion that x2 Setting on the Voodoo 5 actually matches the x4 setting on the Nvidia (third notch over). This was a monitor next to monitor comparison running the same mission at the same time.  The x4 setting on the Voodoo 5 simply is without compare on the Nvidia,  this is probably as it should be considering the different implementations and the completely different focus. 3dfx offers FSAA as one of its main features and Nvidia allows a user to use it but the company doesn’t think its useful believing instead that high resolutions obviate the need.  Therefore its important to remember when viewing these benchmark numbers that for a given image quality you would have to compare the Voodoo x2 FSAA setting to the Nvidia x4 FSAA setting. 

Here is a comparison shot that shows some of the problems the Nvidia can have from time to time when the optimum angle is not met with its implementation of FSAA.  This is at x2 FSAA setting is one illustration of quality differences between the two cards in regards to image qualities, its no secret that the Voodoo 5 has better FSAA image quality.  Compare the rear of the wings notice the large jaggies on the Nvidia.*
Geforce 2 5.22 drivers at x2 FSAA	Voodoo 5 stock drivers in D3D (Glide in x2 FSAA exhibited no jaggies of any sort so improvements in the D3D driver may perhaps be on the table)
	D3D Version	Glide Version

Finally before we start I compare Glide numbers to D3D numbers, this is intentional. I’m not doing a comparison of the theoretical speed of these cards but of actual speed. Which means that if you owned the Voodoo 5 or Geforce 2 you would run them in whatever API was suitable. Naturally for the V5 that’s most often Glide.

Screenshots

The Voodoo series of 3d Cards have always been the best choice for Falcon and that’s not changing just yet. As can be seen with this benchmark there is no penalty with the Voodoo for running 1024 and x4 FSAA settings. This must be seen to be believed. Also there seems to be more of a penalty running the Geforce at high resolutions and x4 FSAA settings, notice that it loses 4 frames a second where the Voodoo loses maybe 1 frame. Image quality without FSAA is a toss up between the two when they both have been adjusted properly. Its often said that the Gamma must be adjusted correctly with the Voodoo to match the Geforce but I believe one of the secrets is to adjust the contrast and brightness settings as well as the Gamma. Both cards ran Falcon excellently with no obvious artifacts during the tests I ran. 

(remember FSAA means Full Scene Anti Aliasing)*

Screenshots

The advantage that the Geforce has in the Non FSAA scores evaporates pretty quickly in scores with FSAA enabled.  Indeed again we see the Geforce losing a lot to using FSAA and the Voodoo 5 not losing that much. Also the HUD on the Geforce at 640×480 is somewhat more blurry than the hud using the Voodoo. The Voodoo at the highest resolutions+FSAA actually cleans up the cockpit by making the text easier to read on the switches. Image quality with both using no FSAA was excellent. Using FSAA at high resolutions there were no obvious errors. At low resolutions the Geforce 2 was blurring some text and the HUD.

Screenshots

Flanker 2.0 still doesn’t like the Voodoo 5…indeed is almost like SSI has it in for 3dfx. 8-) When I installed the Voodoo 5 one of the first games I ran was Flanker cause I couldn’t wait to see how beautiful it ran. Luckily I had reinstalled Flanker for this review and so was able to verify that even with the completely capable Voodoo5 Card the textures are still blurry. Its a holdover from when the thought was that you had to provide textures smaller than 256×256 for the V3 and older 3dfx cards to run…this is incorrect. Easily fixed, indeed its detailed how on the front page or in the Old news section of the site. The Geforce 2 was ok (would have been excellent except hampered a bit by some problems accessing the menus causing crashes) in this test but again faded as we turned on FSAA. In the comparisons of speed notice the Geforce is faster without FSAA or 32bit but those are turned on the speed difference is gone. Indeed at high resolutions and high FSAA settings the Voodoo 5 is as fast to the Geforce as the Geforce was to the Voodoo 5 at low res.  We could say that the Voodoo 5 is a better card for people who want all the eye candy that Flanker has to offer.

Screenshots

There is definitely some weird things happening here and I’m not enough of a tech wizard to decipher them.  Obviously some driver problems hamper both cards to varying degree’s. The abysmal 1600×1200 score of the V5 points to that driver problem that was mentioned about only running on 1 chip at resolutions above 1248. The Geforce 2 was exhibiting texture flashing blue colors. There was no corruption where the Voodoo 5 was concerned indeed the image quality was astounding….

Both cards ran Panzer Elite perfectly and the speed difference was negligible. The quality of the FSAA of the Voodoo is superior.

Screenshots

This was one of two games I ran with the 530 Beta drivers for the Nvidia. They were rumored to provide alot more speed as well as stability and I certainly wanted to see if that was the case. Turns out they were a mixed bag. Sometimes they were more stable but they still had weird things like the Rogue Spear GUI (check the screenies)…the increased speed was there in some resolutions but not others. 

The bug that bothers the V5 when in D3D above 1248 is still evident here with the 1.01 drivers.  Both cards provide ample quality and speed to fully enjoy UT.

Quake 3 scores just for the hell of it…. My take on this is yes the Geforce 2 is fast in Quake 3 but then again it has the benefit of some driver releases, I believe it has had 3 releases prior to 522 Had I thought about it I would have benchmarked it with the original drivers to show the improvement possible. This would help to alleviate the defeat of the Voodoo card….after all its on its release driver. Course if your goal is the absolute speed crown in Quake 3 then the Geforce 2 is probably your card.

I figured out how to benchmark this program on the last day of my test and so wasn’t able to really test it the way I wanted to.  This is provided solely for informative purposes and no conclusions will be offered. Rogue Spear Urban Ops with either of these cards is an amazing experience and FSAA does contribute to the overall scene.

Turning down the Volume to see who runs best with the slow puters Benchmarks with a 366mhz P-IIIe

This was simulated by turning down the bus speed on my computer to 66mhz. Naturally this will not be an exact measurement but certainly close enough for our purposes.

This test shows a bit of strangeness that Im at a loss to explain. I re-ran these tests with the Voodoo 5 several times almost 2 weeks apart and the results were always the same. x4 FSAA was faster than x0 FSAA in Glide with Falcon 4. The Geforce 2 shows the expected decline in framerates as the resolutions and FSAA settings go up, but not the Voodoo. This probably can be explained by the relatively new drivers and the fact that 3dfx is concentrating on stability rather than sheer outright speed. 

Well at this speed neither card is gonna help you much. Time to invest some money and get both the CPU and card….

(remember FSAA means Full Scene Anti Aliasing)*

Screenshots

Conclusions and Recommendations Opinions even…hehe.

First off since both of these cards are extremely fast, speed is not an issue with either card. If speed is not an issue then what criteria can you use for making your decision on where to spend your money? Obviously features, stability and compatibility should figure very large in your decision. Also the two main features of these cards are very different, and that should also help you make your decision. Admittedly though this decision will be hard because actually there isn’t much separating these cards and either one would be a terrific addition to your computer.

Features present a difficult choice since 3dfx’s FSAA is available and brilliant right now with every title you own. Its hard to relate the difference it makes in games except to say that if you get one you should immediately enable the switches in both Glide/OpenGl and D3D that allow you to turn it off and on while running a game. That will show beyond a shadow of a doubt how important it is even in games being run at high resolutions. What’s more is that with 3dfx’s Full Screen Anti Aliasing you don’t have to wait for developer support, which given the state of PC gaming, expecting developer support of features not essential to releasing the game might be optimistic. Nvidia has a really neat list of developers on their website that are said to be supporting TnL on the Geforce, several games listed there have been released without the support due to…you guessed it time and budget pressures killed the feature. No doubt by Christmas we should be seeing a bunch more titles that support the TnL Hardware support of the Geforce 2. How that will work with what by then will be an old card is anyone’s guess. Sorry to be so pessimistic but having worked on a few beta teams and been shocked by what gets left out in the rush to release especially around Christmas I’ve learned to be pessimistic. Nvidia is certainly to be commended for pushing the envelope and bringing hardware TnL support to a consumer card first….but whether you need to be part of that gamble is up to you.

Nvidia does have FSAA enabled with their drivers after a fashion. But its obviously been rushed and should be considered work in progress instead of an available feature.  The settings are only explained by going to fan sites and reading FAQ’s and certainly that’s not a big deal but you would think that something the company is supporting would have a better explanation in the actual help file that comes with the driver.  Either the rush to meet that 6 month deadline or the rush to answer 3dfx’s shot across the bow with the V5 has caused some chinks to appear in what looked like solid steel armor.

Stability, here again Nvidia gets into problems probably chiefly for its insistence on releasing a product every 6 months. Obviously something gets dropped when you are as a company committed to a release every 6 months and in the Geforce 2’s case it was driver stability. GUI problems with games are the order of the day with little fun games like being forced to swipe your mouse around the screen to see the GUI of Motocross Action 2 demo, or the incredible melting GUI of Rogue Spear Urban Ops and the texture corruptions in other games like World War 2 fighters.  On the other hand 3dfx’s delay in releasing the Voodoo 5 played right into the driver teams hands allowing them to thoroughly test the driver for stability…it shows, very stable and extremely quick, but not without problems. The weird scores where x2 FSAA is faster than x0 FSAA in both Glide and D3D show that 3dfx has alot of room for improvement. But all the games I threw at it ran and I didn’t have any problems with GUI’s or texture corruption.

Compatibility…this is not an area for opinion. Flatly the Voodoo 5 can play all the games that the Geforce can plus all the Glide games still there and yet to come. This is a strong point since if speed is not a real factor then why limit yourself to only D3D and OpenGL games…?

In the end I was just floored by the improvements to the images when displayed using FSAA. I wont be playing any games with it disabled.  And if I asked myself which one played FSAA better the answer was very obvious and indeed most other websites agree that the 3dfx was hands down the choice. It has a better image quality when using FSAA and on top of that is usually faster on the whole when displaying that better image. Add to that the absolutely solid driver released with the card along with the stability in games and the choice was easy for me Voodoo 5.  When I saw for myself that there was NOT this big speed gap in games that the Lazy review sites led me to believe the deal was done, V5 for me thank you. But you could be satisfied with either card depending on your specific needs as both will do nothing but continue to improve markedly with better drivers.

Screenshots