With the debut of the GeForce4 Ti 4400 and GeForce4 Ti 4600 in March of 2002, the GeForce4 Ti 4200 is scheduled for a retail appearance at the end of April. Based on recent reports, we've also learned that NVIDIA will be phasing out the GeForce3 Ti 200 and GeForce3 Ti 500 which will eventually, if not already, be in short supply.

Based on our preview, the GeForce4 Ti 4600 achieved a 30-40% increase in performance over the GeForce3 Ti 500 in higher resolutions when antialiasing was disabled. With antialiasing enabled, there were cases where frame rates doubled. With its unmatched performance, the GeForce4 Ti 4600 also commands a premium price, which is customary for new high-end products in the computer hardware industry. At the time this preview was published, the GeForce4 Ti 4600 still retails for around $350 or more, while the GeForce4 Ti 4400 is selling for about $250.

The exciting news is that the GeForce4 Ti 4200 is destined to overtake the GeForce3 Ti 200 as NVIDIA's mid-range graphics offering. The 64MB version of the GeForce4 Ti 4200 will debut with a suggested retail price of $179, while the 128MB version will cost $199. All four GeForce4 Ti's are based on the same architecture, with the exception being the graphics processor and memory speeds.

Note that the 128MB version of the GeForce4 Ti 4200 has a lower memory speed than the 64MB version. Also of interest will be the positioning of the GeForce4 MX 460 in the GeForce4 lineup. I briefly tested the GeForce4 MX 460 here.

Since antialiasing is used throughout this preview, it's a good idea to list the modes which exceed frame buffer storage for the 64MB and 128MB models of the GeForce4. If an antialiasing mode is selected whereby the memory requirements exceed the memory available, the hardware may fall back to a lower antialiasing mode or possibly no antialiasing.

A GeForce4 with 128MB of graphics memory can support 2X, Quincunx, 4X, and 4XS antialiasing at all resolutions, with the exception of 2048x1536, in 16 and 32-bit color. The 64MB variant can do the same in 16-bit color, but is limited in 32-bit color. For example, Quincunx antialiasing can be used at 1600x1200 in either 16 or 32-bit color with 64MB of memory.

The following table lists some of the specifications for the GeForce4 Ti series of graphics cards.

Of particular importance is the Accuview Antialiasing subsystem, which was designed to increase the performance of Quincunx antialiasing and matches 2X antialiasing performance in most cases. However, a significant performance penalty is realized when using anisotropic texture filtering.

While there are a few web sites that consistently publish good graphics card reviews, I've found that most of them rarely comment on gameplay. This observation isn't intended to criticize since such an analysis is time consuming as well as subjective. One such review I did was with Prolink's GeForce3 Ti 200 and I decided to do the same with the GeForce4 Ti 4200. Keep in mind that gaming and benchmarks were done using a relatively high-end processor. A general rule of thumb is the more powerful the processor, the better the 3D graphics performance will be on a high-end graphics card like the GeForce4 Ti.

System Configuration

• AMD Athlon XP 1800+ @ 1.53GHz
• NVIDIA Reference Motherboard (nForce Chipset)
• 256MB Corsair PC2400 DDR RAM
• 21-Inch Sony Multiscan E500 Monitor
• NVIDIA Reference GeForce4 Ti 4600 (300MHz/650MHz) - 128MB
• NVIDIA Reference GeForce4 Ti 4200 (250MHz/500MHz) - 64MB
• NVIDIA Detonator XP Driver Version 28.32
• 32-Bit Color / Sound Disabled * / Vsync Disabled / 75Hz Refresh Rate
• Windows XP Professional / DirectX 8.1

Notes

• A SounderBlaster Live! Value was used during game play

Software

• Quake 3 & Team Arena - Version 1.30
• IL-2 Sturmovik - Version 1.03a
• Nascar Racing 2002 Demo - Version 1.0.0.1
• Jedi Knight 2 - Version 1.02a
• Serious Sam 2 - Version 1.05
• Max Payne Demo - Version 1.02
• Comanche 4 Demo - Version 1.0.1.18
• Dungeon Siege - Version 1.0
• Tiger Woods PGA Tour 2002 Demo - Version 1.0

Screenshots

Note that all full size screenshots in this preview were saved in portable network graphics (PNG) format. While PNG format results in a large file size, it does a better job at preserving image quality than JPEG format. This was an important consideration since antialiasing is shown extensively in this preview. Dungeon Siege screenshots were taken with antialiasing enabled on the GeForce4 Ti 4200 in lieu of using the aa_screenshots=true command line parameter.

Quincunx Screenshots

Due to the implementation of Quincunx antialiasing on the GeForce4, screenshots do not show the associated "blurring" effect. While this effect is noticable, it is less prevalent than Quincunx on the GeForce3.

OpenGL game performance begins with version 1.30 of Quake 3 using the Demo4 map. All benchmark results were based on a maximum quality setting, which consisted of the standard high quality setting with high geometry and maximum texture detail enabled. The GeForce4 Ti 4200 was tested at resolutions of 1024x768, 1280x1024, and 1600x1200. Each resolution contains results without antialiasing enabled, and with 2X, Quincunx, and 4X antialiasing. Each test also contains texture filtering results using trilinear filtering, 2X anisotropic filtering, and 4X anisotropic filtering. High quality sound was enabled.

The GeForce4 Ti series of graphics cards are about choices. As a gamer, the GeForce4 Ti 4200 provides a greater number of options, such as antialiasing versus no antialiasing or trilinear filtering versus anisotropic filtering, to improve performance or enhance image quality. Of the 36 Quake 3 benchmark results, 23 have frame rates that exceed 60 frames per second, 10 are less than 60 frames per second, and 3 are because 64MB is not sufficient to enable 4X antialiasing at 1600x1200.

Game play performance with the GeForce4 Ti 4200 in Quake 3 and Team Arena was measured using two different graphics settings - 1280x1024 with Quincunx antialiasing and 2X anisotropic filtering and 1600x1200 with no antialiasing and trilinear filtering. High quality sound was enabled as well as background music. With Quincunx antialiasing enabled at 1280x1024, the GeForce4 Ti 4200 offers solid performance in heavy multiplayer Quake 3 action.

Quake 3 tests consisted of setting up a local multiplayer match on level Q3DM1 against eight bots with a frag limit of 20. With Team Arena, I created a six-on-six capture the flag match with a 5 minute duration using the outdoor map Overdose. The maximum frames per second setting was changed to 999 (seta com_maxfps) while FRAPS was running in the background and logging the minimum, average, and maximum frame rate during each match. Each match was played twice and I made sure not to hold anything back. In other words, it was an all out frag fest.

Here are my results:

• 1280x1024 - Quincunx Antialiasing - 2X Anisotropic Filtering
  
  Q3DM1 - Quake 3
  
  Match 1 - Frames: 9055 - Time: 120313ms - Avg: 75 - Min: 30 - Max: 118
  Match 2 - Frames: 9425 - Time: 123298ms - Avg: 76 - Min: 25 - Max: 120
  
  Overdose - Team Arena
  
  Match 1 - Frames: 14697 - Time: 293693ms - Avg: 50 - Min: 27 - Max: 98
  Match 2 - Frames: 14931 - Time: 299350ms - Avg: 50 - Min: 24 - Max: 98

• 1600x1200 - No Antialiasing - Trilinear Filtering
  
  Q3DM1 - Quake 3
  
  Match 1 - Frames: 10600 - Time: 101125ms - Avg: 105 - Min: 45 - Max: 179
  Match 2 - Frames: 10128 - Time: 103822ms - Avg: 103 - Min: 49 - Max: 179
  
  Overdose - Team Arena
  
  Match 1 - Frames: 20644 - Time: 294964ms - Avg: 70 - Min: 36 - Max: 115
  Match 2 - Frames: 21837 - Time: 297698ms - Avg: 74 - Min: 35 - Max: 128

In regards to antialiasing image quality, I've put together a comparison much like the one in my GeForce4 preview. A key point of this comparison is that there was a difference in Quincunx antialiasing on the GeForce4 between that preview and this one. Using the latest official Detonator drivers, the new Quincunx sampling pattern on the GeForce4 results in less "blurring" of textures under OpenGL. However, unlike the GeForce3, you will notice that screenshots of Quincunx on the GeForce4 don't pick up the post filter effect. Clicking on the image below allows a comparison of antialiasing image quality by using a java applet that superimposes images over one another. All of the images were taken from the same in-game location.

In the event that the image quality page doesn't work with your browser, here are the links to each image used in the comparison.

• GeForce3 - No Antialiasing - 775 KB
• GeForce4 - No Antialiasing - 775 KB



• GeForce3 - 2X Antialiasing - 777 KB
• GeForce4 - 2X Antialiasing - 777 KB



• GeForce3 - Quincunx Antialiasing - 729 KB
• GeForce4 - Quincunx Antialiasing - 775 KB

It's not often that we cover simulations in our articles, but I felt IL-2 Sturmovik deserved recognition as Maddox Games has developed a top notch WW2 flight simulation. A quick glance at the SimHQ message board shows the IL-2 forum, and it's two massive archives, bursting with activity. I'd like to thank UbiSoft's Sean Kauppinen for providing me with a copy of the game which I've enjoyed playing.

After having spent the better part of the past two weeks scouring the IL-2 Sturmovik forum at SimHQ, I ran across a track that was recorded by forum member Ming_EAF92 for benchmark use. The 16Plus32BigOnes.trk places a heavy load on the graphics processor, as well as the central processor (artificial intelligence and physics), by simulating a dog fight with 48 aircraft. The GeForce3 setting with trilinear filtering was selected from the video setup menu while the excellent picture quality setting was chosen from the in-game video option.

Sound was disabled. The average frame rate was determined by using the playback feature in IL-2 Sturmovik while FRAPS was running in the background. Benchmarks were run under the Direct3D (D3D) and OpenGL (OGL) API's.

One glitch I noticed during playback was a moderate stutter around the 2:30 mark where the frame rate dropped in the upper teens to low twenties which was due to a nearby aircraft explosion.

The graphics in IL-2 Sturmovik are admirable assuming you have a powerful enough system to enable high quality settings. With the GeForce4 Ti 4200, I've been using excellent picture quality under OpenGL at a resolution of 1280x960 with Quincunx antialiasing along with the SoundBlaster Live setting. IL-2 Sturmovik does have a few problems such as "micro stutters" that occasionally occur during offline play and my first and only attempt at mutiplayer wasn't a pleasant experience. However, there is some good news for IL-2 players as a major patch has been released.

As for antialiasing image quality, there's no question that 4XS, which currently works under Direct3D, is desirable. Since the release of the GeForce4, we've learned that 4XS uses multisampling and supersampling antialiasing schemes. In this comparison, the "blurriness" associated with Quincunx antialiasing can be seen under Direct3D, but not under OpenGL. Note that these screenshots weren't taken from the exact same position.

Again, in the event that the image quality page doesn't work with your browser, here are the links to each image used in the comparison.

• Direct3D - No Antialiasing - 587 KB
• OpenGL - No Antialiasing - 589 KB



• Direct3D - Quincunx Antialiasing - 592 KB
• OpenGL - Quincunx Antialiasing - 597 KB



• Direct3D - 4XS Antialiasing - 611 KB
• OpenGL - 4X Antialiasing - 599 KB

In order to better appreciate the results of antialiasing on the GeForce4 Ti 4200, I enlarged a section of the top right cockpit by a factor of two. What you will discover is that while the results of antialiasing are beneficial, they can vary based on the API (Direct3D vs OpenGL) and by the view (horizontal vs vertical).

When comparing Quincunx under Direct3D and OpenGL, notice that under OpenGL the blending of colors along the edges provide a slightly smoother looking appearance.

When comparing 4XS and 4X antialiasing, notice the difference between the top left edges, which are aligned horizontally, and the bottom right edges, which are aligned vertically. You may wish to magnify the images by another factor of two, using a freeware program like Irfanview, but 4XS does a better job than 4X on horizontal edges while the opposite is true on vertical edges.

I'll end this section with a screenshot showing 4XS antialiasing at a resolution of 1920x1440. What's exciting is that within the next two years, we'll probably be gaming with similar graphics settings.

Nascar Racing 2002 is the latest racing simulation from Papyrus and runs under Direct3D and I believe the retail version also supports OpenGL. My first use of the demo was in my GeForce4 Ti 4600 preview to show the benefit of 4XS antialiasing. Since then, I've a recorded a demo race which is close to 5 minutes in length. This series of tests was based on playing back the demo with all graphics options maximized with the exception of the highest level of anisotropic filtering. I manually changed the value from 8 to 2 in the rend_d3d.ini file.

Note that benchmarks with antialiasing enabled at 1280x960 or 1600x1200 were not provided as some textures were severely blurred at those resolutions. I don't know if this is also an issue with the full version of Nascar Racing 2002. Sound was not used during play back, but its affect on performance isn't too drastic during actual game play.

Papyrus recommends an average frame rate of 30 frames per second for good game play in Nascar Racing 2002. Again, the GeForce4 Ti 4200 offers a number of settings that can be used that achieve this average.

Jedi Knight 2 was released recently by Lucas Arts and contains a benchmark feature, which isn’t surprising, since the game is based on the Quake 3 Team Arena graphics engine. With Raven Software being the lead developer and experienced with the Quake 3 graphics engine, I was certain that Jedi Knight 2 would turn out to be an effective benchmark.

I noticed a couple of new graphics options in Jedi Knight 2, one of which raises the bar as far as graphics quality and performance is concerned. A very high texture detail option was added to support textures that are of higher quality than those used in the high texture detail setting. Even with texture compression enabled, the very high texture detail option caused lengthy loading times in single player mode and game play suffered as a result of continued disk access when retrieving texture data. In mutliplayer games the frame rate dipped below 10 frames per second during periods of heavy graphics activity. While Raven recommends at least 512MB of system memory if very high texture detail is going to be used, the game ran and looked great using high texture detail.

The following graphics settings were used when benchmarking Jedi Knight 2 using the included FFA demo. Sound was disabled.

I didn't conduct a complete set of benchmarks for Jedi Knight 2 since I was mainly interested in the settings that could achieve 60 frames per second or higher. Notice how performance declines on the 64MB GeForce4 Ti 4200 when very high quality textures were used at 1280x1024 and 1600x1200. This wasn't an issue with a 128MB GeForce4 Ti 4600.

After examining the benchmark results, I selected a resolution of 1280x1024 with Quincunx antialiasing and 2X anisotropic filtering, which was benchmarked at 70 frames per second, and proceeded to test on-line performance. High quality sound was enabled with FRAPS running in the background. I played on servers hosting a variety of games including capture the flag, saber dueling, and free-for-all. My last official test was participating in an eighteen player FFA match. Here are results from my various on-line sessions with Jedi Knight 2:

• 1280x1024 - Quincunx Antialiasing - 2X Anisotropic Filtering
  
  Session 1 - Frames: 34569 - Time: 409909ms - Avg: 84 - Min: 33 - Max: 142
  
  Session 2 - Frames: 12931 - Time: 305249ms - Avg: 42 - Min: 23 - Max: 98
  
  Session 3 - Frames: 20836 - Time: 314422ms - Avg: 66 - Min: 32 - Max: 110

As with most new games, there are usually features that make each of them unique. In Jedi Knight 2, saber dueling adds a new twist along with the various force powers (push, pull, sight, heal, protect, etc.) that can be configured based on a style of play.

To take a high quality screenshot in Jedi Knight 2, bind a key like this - bind F11 "screenshot_TGA".

Developed by Gas Powered Games in partnership with Microsoft, Dungeon Siege is a hack-and-slash type of role playing game with limited character development. The role playing genre of games have been a favorite of mine and Dungeon Siege is the talk of the community as it was the third most requested search phrase at Google for the week ended April 8.

Dungeon Siege's graphics engine is based on DirectX 7 and consists of a third person view that can be rotated in all directions. The point of view can be shifted vertically, which takes getting used to, and a zoom in/out feature is present. Objects such as trees and branches will become transparent in order to keep the line of sight visible. The sprawling landscape and dungeon artwork is visually stimulating and weather effects such as fog, rain, lightning, and snow are a nice touch. Character animations are convincing and the spell effects are effective.

At the 2002 Game Developer Conference, Chris Taylor, who is the president of Gas Powered Games, commented on the GeForce4 saying that "it's architecture was superior in every way" and even asked NVIDIA for samples of the card for his development team. And it gets better as Dungeon Siege 2 is expected to further push the GeForce4 by increasing the polygon count and texture resolution.

At this point, I've logged about 20 hours playing the single player version of Dungeon Siege using a variety of graphics settings. Using 2X and Quincunx antialiasing causes the game to display a black screen and I understand that an upcoming patch will correct that problem. Resolutions of 800x600 and 1024x768 are officially supported while higher resolutions can be enabled with command line parameters. FRAPS will work with Dungeon Siege, if the bltonly=false command line parameter is added, but I was able to measure performance with the games internal frame rate counter. I've read that the frame rate is not being reported properly, but it seemed to operate correctly for me.

It was a challenge to come up with a favorite graphics setting when playing Dungeon Siege. Some graphics elements, such as the character inventory and text, become difficult to read because of their size at resolutions over 1024x768 - even on a 21-inch monitor. Although objects are based on a smaller scale unless zoomed in, antialiasing is quite beneficial, as I've become accustomed to using it. Another dilemma I faced was whether or not to enable environmental audio extensions (EAX), which offers enhanced sound effects but would eat up CPU cycles. During periods of extremely heavy activity, frame rates would drop by as much as 40% with EAX enabled.

Nevertheless, I began playing Dungeon Siege using 800x600 with 4SX antialiasing, maximum object detail, simple shadows, and EAX disabled. I later moved up to 1024x768 with 4XS antialiasing after runningtests with the in-game frame rate counter enabled. In the majority of cases, I would estimate that frame rates averaged around 30 frames per second until a major battle took place where the frame rate would dip into the upper teens and low twenties. Even when I used the GeForce4 Ti 4600, performance wasn't noticeably better. If I had to guess, I would say it performed 10% faster than the GeForce4 Ti 4200. With or without antialiasing, frame rates didn’t average over 40 frames per second very often. I was able to dig up some comments in regards to performace in Dungeon Siege by Gas Powered Games at the Planet DungeonSiege forum.

This is a DX7 game (though networking uses DX8). We install DX8.1 only because it contains post-DX7 bug fixes, but do not actually use any features like pixel/vertex shading from 8 for graphics. This is why we can run on Voodoo 2 boards.

Anyway, there are two frame rate issues.

First is the problem where people are expecting 80 fps because they have a maxxed out system, and that's what they get in Q3. In DS, the frame rate is way more dependent on CPU, hard drive subsystem, and memory than the video card. All that CPU goes into AI, line of sight calculations, physics simulations, pathfinding/following, and a bunch of other stuff. Most importantly it goes into loading and unloading objects in background for our continuous world (though a dual-proc will help you out there). This is a very CPU-intensive game.

Second issue is the problem where people have bad frame rates relative to each other with similarly configured systems. I recently discovered that problem myself at home (and I feel your pain, folks, ouch) and it appears to be configuration-specific. :( We are working on figuring out what's going on there, and hope to put a fix into the next build. If you have a maxxed out system you might not get 80 fps, but you certainly shouldn't get 15! We're working on it.

Other games I checked out with the GeForce4 Ti 4200 included Serious Sam: The Second Encounter and the demos of Max Payne and Comanche 4. Serious Sam 2 was played using quality graphics settings and trilinear filtering. The test consisted of playing the Sierra de Chiapas level on normal difficulty until reaching the palace gate. Details on how Max Payne was tested can be found here and Comanche 4 consisted of trying to blow as many things up, using the first person view, before my helicopter was blown up. Sound was enabled for all tests while FRAPS was running in the background.

The Sierra de Chiapas level contains a generous number of trees and lush vegetation making it a good stress test. Average frame rates increase by 2X or more in indoor levels. Much of the Comanche 4 level takes place over water, which is rendered with the assistance of pixel shaders (see screenshot below). I also played an earlier demo of Comanche 4 back in February on a land-based map.

Paul Olshan, who is a Product Manager at NovaLogic, dropped by our forum last month and provided some feedback in regards to Commance 4 performance.

Thank you for your comments. I've been following your progress on this forum and I'm concerned that you're missing some important information about our game and this benchmark.

First of all, this benchmark is not a synthetic test of GPU performance. Rather it is indicative of the game performance in Comanche 4. We render over 200K triangles per FRAME and nearly 10 million polys per second on GF4 and fast P4 systems. Fast CPUs, fast memory, fast AGP and fast GPUs are required to render this much data. At lower resolutions, the CPU, AGP and main memory bandwidth are the performance limiting factors of the game. If you run the test with FSAA and resolutions above 1024x768, you begin to load up fast GPUs such as the GeForce 3/4 and the Radeon 8500.

Comanche 4 does not offer the same performance profile as Quake or Unreal which is why we think it is interesting. It offers something different and offers a realistic look at games that render LOTS of polys in a real game scenario. Comanche 4 is probably heavier on AI and physics than many upcoming first person shooters but we think it does closely resemble the amount of data that your PCs will have to manipulate for high-quality, rich environments.

Anyway, I hope this clears a few things up and explains the results you are seeing.

At this point, you've probably noticed that I'm normally using some form of antialiasing with the GeForce4 Ti 4200. While antialiasing on the GeForce4 is a second generation multisampling solution, NVIDIA has come a long way since the days of the supersampling on the GeForce2 Ultra. And the best is yet to come as better forms of antialiasing should be delivered in the future.

Why a gamer would invest in a GeForce4 Ti is quite simple actually. Wouldn't you want to play games with graphics that look like this? While I didn't cover the pixel shader capability of the GeForce4 Ti 4200 in this preview, you can take a look at what game developers can do with such a tool. Tiger looks great in his antialiased attire as does the water, which is a product of pixel shaders.

I will say that the GeForce4 Ti 4200 is an outstanding product and is the most impressive graphics card for the price I've had the pleasure to test in over four years in this business. So do yourself a favor and consider the GeForce4 Ti 4200 if you're in the market.

Back to nV News