Pierre Legrand’s Pink Flamingo Bar

3dfx Voodoo 5 vs. LeadTek Geforce 3    
Well here we go again another comparison on my site between a Voodoo 5 and Nvidia product.  The last review generated some pretty intense flames with one person even going so far as to find out where I lived so that he could threaten me….weeeee. Lets hope it doesn’t degenerate into that again. I would think that all of us have gained a sense of perspective since 9/11. This is after all is said and done just some hardware to send some cartoons careening around the screen, fun without a doubt but of no real consequence.     

Geforce 3 x4FSAA	Voodoo 5 x4 FSAA
Crybabies should duck if they get out of line!!! One of the biggest disappointments was not being able to benchmark iL-2 in its native  Open GL environment, this landmark sim deserved its own benchmark tool built in!Â

First a little story about expectations. I’m something of a wine lover, early on I found that to get an honest opinion about wine it was absolutely essential to hide the maker of the wine by wrapping the bottle in a paper bag.  This was done so as to not prejudice the tasters because expectations being what they are, the tasters would respond to wine differently if it were labeled with the name of Domaine Romanee-Conti than if the same wine were labeled with a Gallo label. As proof of this a friend and I filled a empty bottle of Margaux with Chilean wine. We then served it to a group of friends that had shared the Margaux with us the night before. The unwitting seeing the label of Chateau Margaux assured us that this bottle was every bit as good as the bottle the night before. Expectations that were raised by seeing the label brought that on….the same thing happens in Video cards. Some see the Nvidia GF3 label and assume that it can only be faster than everything else especially an older card like the V5. This leads to them seeing a lot of stuff that doesn’t really exist, like oooh man this card is soooo much faster than my old V5 in Falcon. Me….I love that Missouri saying “Show Me”…!!!  
So what is the purpose of comparing old dog vs the big bad ass new dog aside from wanting to see for myself? Easy there are a lot of people who own a Voodoo 5 that are wondering whether spending the $300.00 for a new Geforce 3 will be worthwhile. A note, please understand that this card compares favorably to the newest GF3 Ti 500 since it was clocked at 220/525 for the duration of this test no anomalies were seen at this speed and I believe it could have hit much higher.  I hope to provide some comparisons with some apps to help you decide whether your $300.00 should be spent on the Winfast LeadTek Geforce 3 or a new motherboard and CPU combination. For the attention deficit disorder crowd I believe that unless you own an Intel i740 Graphic Chipset or the like that you will ALWAYS get better bang for buck going with a new CPU. 

There are several items that go into this decision and the weight that goes behind each of those items varies from person to person. A lot of gamers probably start by considering what games they will play. This is rational; the software the user plays should drive any decisions about hardware. This isn’t as important with new software as it used to be since the ascendancy of Dx and OpenGl means that open standards prevail with most new cards supporting both fairly well. But what if some of the games you love are Glide? 3dfx remains the best way to play those games and that will enter into your decision.  Red Baron only plays in 3D accelerated mode with the Voodoo chipset, Longbow 1 & 2 play best on the Voodoo, as does F-15 and there are others. It’s easy to forget that in the computer world a lot of people buy new hardware just because it’s new. To them it’s the hardware that’s fascinating not the software. Doesn’t make them bad, doesn’t make them right or wrong, but you do have to realize they may not have your habits in game play. For instance they may be perfectly happy not playing a game ever again, you may not. Making a hardware decision then discovering that you just ruled out playing your favorite game is pretty rough if you bought from somewhere that doesn’t take returns. Before buying anything make a list of the stuff you cannot live without and see what the requirements are, you may be faced with building a legacy box to play some of the older stuff, though considering the cost of a second set of TM controllers oh my. 

The next item will be image quality. One of the biggest advances in image quality in consumer cards was driven forward by the once king of 3-D Chipsets 3dfx. Known by its acronym FSAA (full scene anti aliasing) 3dfx managed to inject something really useful into the soup that comprises gaming. The experience was stupefying for those who witnessed the demonstrations of it in action. Here was a scene that didn’t swim around with textures dancing all around for no reason other than the player was changing his perspective on the scene. Instead the scene would hold still as it did in real life. Some have tried to say that Anti Aliasing is not required; here is what 3D Labs has to say about why it is. 3D Labs is a leader in state of the art 3D Professional Accelerators. (note: High resolution does not cure aliasing as can be easily seen when comparing 1600×1200 images with and without FSAA, remember that screenshots only show edges)
Why Is antialiasing Needed?
Anyone who has looked closely at computer generated graphics has seen the effects of aliasing — rendering artifacts that produce undesired visual effects such as the jagged, stair step appearance at the edge of lines and polygons. When scenes with aliasing artifacts are set in motion, the artifacts create distracting “crawling” or “twinkling” effects on the edges of the objects in the scene. This seriously impacts the sense of reality delivered to the user.

Antialiasing describes a range of techniques used to reduce these visual artifacts. Most of these techniques attempt to blend the colors of pixels next to the edge of points, lines and polygons to better represent a smooth edge. When viewed up close, the result is a fuzzy looking, soft edge. However, when viewed from a normal viewing distance the appearance is that of a smooth edge, free of visible “stair steps.”
http://www.3dlabs.com/product/technology/superscene_antialiasing.htm   

Here is what I see when I’m fighting in Flanker2.0. Notice how nice and clear the planes ahead are..?? FSAA makes the sim more playable and more real by rendering the wings the way we would see them…instead of twinkling in and out of view like they do in the Zero FSAA screenie next door.	Notice how invisible the wings are of the four planes ahead…? Thats what you see with no FSAA. This is 1024×768 resolution and those wings are really twinkling in and out of view when the sim is in play. When this is moving the effect is even more disturbing.

Nvidia was caught flat-footed by the reaction of the press to 3dfx’s FSAA implementation. Quickly they scrambled to introduce drivers that took advantage of their cards FSAA capabilities such as they were. But immediately it was obvious that 3dfx’s method of Rotated Grid Super Sampling was superior to Nvidia’s Ordered Grid Super Sampling, especially when it came to the biggest advantage of Anti Aliasing, which is to stop texture swimming and creepy crawlies. Nvidia’s solution was not able to come close to 3dfx’s solution but the problem for 3dfx was touting this to the public. Screenshots only show how good edge Anti Aliasing is being done which wasn’t the biggest advantage of RGSS method of FSAA. Unfortunately for 3dfx many supposedly competent websites put up screenshots to compare the two chipset solutions without adding that texture swimming and creepy crawlies, the worst offenders of aliasing, could NOT be shown by screenshots. Screenshots cannot show the best advantage of RGSS over OGSS that being the much improved over all scene stability when actually playing the game.  
To counter this advantage Nvidia worked hard on a feature called Anisotropic filtering which when it works dramatically improves the quality of the textures. But Nvidia seems to have some significant problems getting this feature to work in Dx apps. Witness that they do not enable it from within their card properties in Dx while there does exist a tab in the OpenGl portion of the video card properties tab. There is a significant performance hit associated with this feature. The bottom line remains that FSAA with the Nvidia line of cards does not compare with the Voodoo 5 in image quality, which revolves around not only edge anti aliasing but instead Full Scene Anti Aliasing which means taking care of creepy crawlies, texture swimming and the like.  
The final item should be driver stability with no graphical errors, a driver should be transparent to the game play experience. In the last review the Nvidia driver Ver. 5.11 was very much the reason I didn’t enjoy the Nvidia experience. Glad to say that this has largely been put away as I didn’t have any major problems with the Nvidia reference driver. It was transparent to the experience. The tables indeed have been reversed in that the Voodoo 5 driver is now showing some major problems with games. None that forced me to not play the game but for instance in Ghost Recon I was unable to play at x4 FSAA without major scene corruption. This will only get worse and so those of us who use V5’s are faced with the fact that as time goes by our choices will become narrower and narrower with regards to new games.  
To reiterate this article is for the people who are sitting on the fence trying to decide whether it’s a good idea to dump the V5 for a Nvidia GF3. I wanted to put up framerate scores to show what the actual differences are with apps that people are likely to be using right now. I concentrate on only one resolution 1024×768 with some scattered runs at 1600×1200 and all different FSAA settings that the cards supported. There isn’t a lot of 32bit stuff run just because I normally don’t have to play in 32bit to avoid banding with the V5 and I assume that you don’t either since running in 32bit limits your options with the V5’s FSAA. That is changing as well with some terrible banding and various other faults in one of Ghost Recons missions. 

As it turns out most of the apps I run are Direct X not OpenGL. I did include the requisite Quake 3 run just for fun. But everything else was run with Direct X using Fraps 1.5. Where I could I made recordings so that I could have absolutely identical runs each time. This was possible in Operation Flashpoint, Flanker 2.51, Ghost Recon Demo, Nascar 4, Falcon 4 Patched to RP5, eFALCON 1.10 with the rest of them I made missions that I would enable autopilot as soon as it entered the mission. All of these benchmarks come within a .5 frame of each other on succeeding runs. Repeatability is essential for benchmarks. All of the missions and recordings with instructions will be available for you to download to run your own tests. So on to the show….. 
The cards tested were the WinFast Lead Tek Geforce 3 clocked to 220 core 525 memory and the 3dfx Voodoo5 clocked to 185mhz.  I used the Detonator 23.11 drivers for the Geforce and Beta 1.04.01 for the Voodoo5. Each card for each test was set to the highest quality settings with the exception of the Geforce 3 which instead of 32/64 tap antistrophic was set to 16/32 tap for all tests.  I don’t go into complicated explanations of how to run the benchmarks because honestly if I make it too easy then the amount of email I will receive from people trying will overwhelm me. If I make the hurdle high enough to start with I tend to not have as much email asking stupid questions. (such as where does this file go…or what is fraps, questions like that deserve your research)

 Quake 3 Demo Well lets get this benchmark out of the way since it’s such a popular benchmark. I didn’t test this as thoroughly as I might have if it weren’t so popular.  The Voodoo 5 gets absolutely destroyed in this benchmark. All settings have been manually moved to the highest settings. I noticed that if you just select High Quality not all the settings are set to highest settings.

If you play Quake3 engine games you need the GF3 badly if you are using a Voodoo5. The GF3 is absolutely night and day better in framerates as can be seen by these astounding numbers. If they seem low to you, remember that I didn’t use the default high quality settings since that didn’t move all the sliders to WFO. Course if you are like me and stuck on a few wonderful Glide Games or love playing Operation Flashpoint in Glide don’t worry just don’t play Wolfenstein in 32bit…. play’s wonderfully well at 800×600 16bit x4 FSAA…. I know that’s how I play it now. Medal Of Honor though may force on me the fateful decision of moving towards some sort of upgrade. 
3D Mark 2001 I have to first admit that I massively messed up on this benchmark since I only tested x4 FSAA.  I decided on this test after I saw how unexpectedly close the Voodoo 5 was to the GF3 on games that didn’t support TnL or didn’t support it enough to be obvious or supported it but it didn’t make any difference….Nascar 4 anyone?  I ran this test to see how strong both cards were without any affect from TnL being considered. It was interesting to see how this highly respected and widely used benchmark showed the differences between the two cards….when TnL was factored out of the equation. I blew it since I don’t have the GF3 anymore and wasn’t able to run the tests at x0 FSAA and x2 FSAA where the Voodoo 5 was always much closer in the other benchmarks. Would have been a great counter argument to all those Hardware weenies who think we should all sell everything to buy the latest and greatest.

Yup no doubt that the Geforce 3 is able to sustain higher framerates when FSAA is pushed to the maximum on both cards. At x4 FSAA the GF3 pretty much dominated the standings of the other benchmarks as well. When Nvidia gets FSAA working properly then that will mean a lot more since x4 Nvidia and x4 3dfx are considerably different experiences. I just wished that I had been smart enough to run the GF3 at x0 and x2 FSAA without TnL support.
Nascar 4 is one of the top selling games of 2001 being in 8th position on the charts last time I checked. I made this recording at Daytona with a full field. To run the benchmark please select BenchmarkN4 from the recordings, make sure to have Fraps started, switch to Earnhardt then to cockpit, start the recording, start fraps averaging, press the space bar to lower the menu bars then let the recording run for a total of one minute fifty seconds, press the space bar to raise the menu bar and watch the counter, when it hits 2 minutes stop fraps.

This is another one of the sims I checked and rechecked and just was floored by the first result. Nascar 4 has support for Hardware TnL and yet here was the V5 significantly faster except when we get to x4 FSAA where the GF3 runs away and hides. If this is one of your sims you may want to hold off to see how good the next generation of video cards is before selling you V5.   

Flanker 2 was a bit of a problem for the Voodoo 5 since at x4 FSAA the mirrors would make it puke into framerates of less than 3 frames per second. But if the mirrors were turned off the sim screamed and the visuals were outstanding. Having the mirrors on did not affect the GF3 so it’s a problem perhaps with the way memory is handled in the drivers of the V5. Alas since 3dfx is out of business no more new drivers for fixing things.
For the Flanker 2 benchmark as well as Jane’s F/A-18 I made up two different tests. I decided to see if there were any significant differences in the percentage of difference in scores for the two cards based on the amount of load on the CPU. In English this means that BenchLow is a below 200 meter flight over Sevastopol with many flights and ships milling about.

 This is absolutely the worst-case scenario for a mission with nothing but the strongest Computers being able to achieve reasonable speeds. BenchHigh is a mission that starts up high and then goes lower then back higher with only a wingman for company. The benchmark high would be a general statement on how fast a card can be under ideal circumstances. What this will tell us is if a particular video card performs better than the other under load. From that we might be able to surmise which card would be better in a wider variety of missions. Pick up Benchmarks zip here.

Benchlow with Flanker & V5 x4 FSAA

The benchmarks for Flanker consist of two different trks as described previously where one is a low altitude heavy load and the other a high altitude light load. For BenchLow the mission starts coming in low over the harbor with the trk paused. Please press S to un-pause, then immediately press the scroll lock key to start up Fraps 1.5. Run the mission till it changes to waypoint 4 then when it is perpendicular to the runway of Sevastopol stop the Fraps counter. It’s crucial in comparisons to stop it at the same spot each time. For BenchHigh run the mission till it starts going uphill and watch the altitude. When it reaches 4,450 meters stop Fraps.   

 

Obviously Flanker 2.51 loves the Geforce 3 at x4 FSAA with scores that double what the V5 was able to achieve. The Voodoo 5 stays right with its newer rival though until that point is reached. It should be noted that at x4 FSAA the picture quality of the GF3 didn’t approach that of the Voodoo in regards to reducing texture anomalies. This was especially evident over Sevastopol where the buildings would shimmy and shake at the highest setting the GF3 could muster.   
Janes F/A-18 is a very demanding simulation and I decided to explore the low load and high load benchmarks here as well. This merely entailed putting together two benchmarks, where the heavy load had many objects, ships, planes and even paratroopers the light load was just you and your wingman. This gives you enough information to understand the strengths and weaknesses of both solutions (video cards). The heavy load naturally has much lower frames per second and the light load conversely has much higher frames per second. For the record I don’t believe that there are any missions in any campaign that stress your puter like the heavy load mission and so you can be assured that you will see higher framerates with comparable hardware. Pickup the Benchmarks here zipped.

Here you are dealing with two missions that must be “flown” using the autopilot. Put both missions in the correct folder then go into the mission and for the high altitude press the scroll lock key to start as soon as you enter but after you press the Autopilot key. Allow the mission to continue until 40 miles shows up on miles to waypoint indicator then stop fraps. For the low altitude mission when you roll out of way point 3 turn off Fraps.

 

Janes F/A-18 was puzzling, so much so that I ran and re-ran all the tests and came up with the same results each time. The Voodoo 5 flogged the GF3. This may have something to do with sparkling water setting inside the sim I came to find out later. I would test it again but the donated GF3 was returned. I would have liked to see what setting was making theGF3 puke and given it the same benefit of adjustment that the V5 got with Flanker.
Ghost Recon was problematic with the V5 since I could not get x4 FSAA to work at all.  Also for the first time 16bit picture quality was better on the Geforce 3 than the Voodoo 5 in a heavy fog mission, obviously some sort of bug since it wasn’t banding so much as just a plain error. That the V5 was excellent in x2 FSAA goes without saying but be mindful that in the heavy fog missions you will need to switch to 32bit which for me is a first, being ‘forced’ to go to 32bit on a V5. I made this recording in the Demo version since I started this review back before it was released.  Its easy enough to run, just remember to press the scroll lock key to enable Fraps when you first enter the mission, turn it off after you have completely circled the castle and have re-entered the backyard of the house. It takes about 6 minutes to run this benchmark. Pick Up Benchmark here.

Well here is a good reason to consider the GF3 since it really does play this simulator very well. You stay close in 16bit but when you go to 32bit poof there goes the framerate. And 32bit rocks on the GF3 with nearly double the framerates. Look at the 1600×1200 x2 FSAA scores and be amazed…actually is faster on the V5 and not only that but the V5 is faster there then at 1024 32bit. 1600×1200 x2 FSAA is where I play now on the V5 since it makes hitting the bad guys easier since you can see them better. But no doubt the GF3 is a good choice for this simulator because of its superior 16bit rendering in this sim and of course its awesome 32bit.
Operation Flashpoint was the first game to bring me out of the 9/11 funk. Smashing bad guys is a good thing and god is it fun to smash the bad guys in Operation Flashpoint. Having the right equipment to run Operation Flashpoint is crucial from my standpoint. This is one of those games that cause grown men to build computers so as to not suffer from bad FPS or ugly graphics. Luckily for me the V5 plays this well. The Benchmark mission has you are standing on a desert island with 4 tanks coming towards you, wait till the last one goes by and as soon as it does turn off Fraps. Viola benchmarked. Benchmark coming soon.

Notice that Im running this in Dx and yet OpFlashpoint supports Glide. Take note that with a Voodoo 5 FSAA is always better in Glide than Direct X with Direct x showing more aliasing anomalies. Don’t ask me why, the guys at 3dfx were sort of “concerned” that I pointed this out in the last review. But it’s the truth and its still true. Framerates are also usually higher. This sim looks much better at x2 FSAA in Glide than the GF in anything else but certainly those high framerates in Dx with the GF3 look very appealing. This benchmark will get the eye candy weenies and speed freaks at each others throats since its obvious that you have to decide between either or…  
Falcon 4 benchmark was the first attempt I made long ago at establishing something more useful for flight simmers than a Quake benchmark. Its still here and ignore those who say that it only tests CPU’s and that its not useful for showing differences in video cards. The evidence is in the scores below. With percentage differences that are absolutely significant, its obvious that different video cards can make a big difference with this particular sim. With this round of benchmark scores I endeavored to show the advantages of both of the most popular versions of Falcon 4.  With the advent of eFalcon various claims were thrown around about how much faster the Geforce series of cards now were than the V5. So back in the last review of Nvidia vs 3dfx on my site I compared them. I found then that eFalcon did indeed bring a significant speed increase to the Geforce group, but that the increase only brought them up to the speed of the V5. That’s still true. I included the Realism Patch without the eFalcon because that’s what I run since it still supports Glide and as I have pointed out FSAA always looks better on the V5 in Glide. I truly wish that other Sim makers would see the advantage of allowing us to benchmark their games. Pick Up Benchmark on this page.

For those who are warming up their fingers to complain about me using Glide to test this…step away from the puter. I don’t care about your protest. I am attempting to help people with a V5 decide whether they want to upgrade or not. Anyone with a V5 will be playing RP5 with Glide….simple fact.. So that’s what I measure.

Benchmarking eFALCON shows that the claims of the eRazor are true, he is actually improving D3D performance a great deal for the Geforce line of cards. But the V5 stays right there with the GF3 in this simulator. No matter which version you play the Voodoo 5 is faster….simple fact, that it looks better while being faster is sort of scary. Don’t get all warmed up to write me about how when you and your buddy are over the FLOT that your GF3 is faster, there are so many things wrong with that method of testing I wouldn’t know where to start to help you understand the problems. 
In Conclusion
So here we are at the end of another review. I cannot wait to hear what promises to be a sack full of complaints about why this or that wasn’t done or why it was done this way or that way. For those who plan on being rude about it go and re-read the opening paragraphs. For those who just want to share their thoughts let me lead off with what I thought could have been done better. In no particular order but numbered to make it easy to remember.

• Should have run Quake in 16bit as well…just for fun.
• Why didn’t I run 3D Mark 2001 in x0 FSAA and x2 FSAA that would have been so informative
• Why didn’t make more screenshots. (I don’t particularly see any benefit from it since FSAA is all about moving images not stationary ones)
• Why did it take so long….(9/11 a birth with a remodel in there with two people working full time…hey Im glad I get anything done, especially stuff that is not earning me anything except bellyaches)
• Should have run more games….
• Biggest wish would have been for there to be a way to benchmark iL-2 in OpenGL.  

That’s about all the stuff I think I could have done better….sure that you guys will think of more stuff. Pass it along and if I ever do something like this again I will try and do it better.    
The Nvidia Geforce 3 chipset is extremely impressive. The demos of what’s possible with the card are absolutely astounding. If you don’t have a GF3 then find someone who does and watch the various demos. Holy Smokes they had my entire family sitting there with our jaws firmly dropped to the ground. Impressive stuff and to think that one day we will be playing games that look like that makes me giddy. The Chameleon was absurdly beautiful, it’s a demo where this Chameleon is walking along a branch and it gradually becomes transparent then gradually becomes a walking mirror where the reflections are real time…WOW find someone with a GF3 and watch that demo. But and this is a big ass but….just like 3D Mark 2001 no one plays that demo. We buy these things to play games.  
This review was specifically about whether it is wise to replace the Voodoo-5 just yet. Of the 39 benchmark numbers I have 18 were wins for the Geforce 3, 13 were wins for the Voodoo5 and there were 8 ties (tie being defined as anything within 2 frames), that’s almost as astounding as the Nvidia demos. Who would have thought that a 3-year-old card would keep up that well with basically brand new card (at the clocked speeds the GF3 achieved its darn close to the Ti500). Just goes to show that you need to carefully analyze the games you play before buying hardware. In general you will be better off buying a faster CPU than replacing your V5, for instance going from a 750mhz P-III to a Athlon 1.4 almost doubled frame rates in Falcon, and was responsible for increasing on average my frame rates 40%! That’s the kind of gain you want to see when you dump $300.00 dollars on a new piece of equipment.  
It was great to see the Nvidia product really take on everything this time with not a hint of the driver instability when using FSAA that I encountered in the last test with the Geforce 2 GTS. Certainly anyone getting a GF3 should be very happy and they will marvel at their cards performance especially in 32bit. But if you are sitting there thinking that getting rid of your Voodoo 5 and buying a Geforce 3 is gonna fix your framerate problems….well you should consider other avenues first. Get the fastest processor you can buy with the best memory and motherboard as that will probably get you much more improvement for the money over the course of all your games. Plus you wont lose your compatibility with older games along with not losing that unbeatable FSAA.  
The Voodoo 5’s FSAA is simply in another class compared to the Geforce 3 and this is coming from a person who played the Geforce 3 for sometime before even starting to do this review to get a feel for it. I didn’t think there was that much difference until I put two computers side by side and ran the same missions on both cards at the same exact moments. Its pretty obvious at that point that Nvidia needs to put all those 3dfx engineers to work on Nv-25 to fix that FSAA. Rotated Grid Super Sampling blows Ordered Grid Super Sampling away…and they need to fix that since FSAA is important. To see how important look at January’s Computer Gaming World and take notice of all the jaggies, aren’t they ugly…? Well maybe you are used to them if you have a Geforce….but I’m not, my Voodoo 5 is still in my computer and it will stay there until something comes along that competes with it over the entire gamut of qualities that’s needed to qualify as better in my world. Hope this has been informative, have fun with whatever choice you make and don’t take life for granted.