Introduction
“640k ought to be enough for anybody.” Dating back to the year 1981, this is probably the most famous quote attributed to Bill Gates. In light of today’s software and accompanying memory requirements, most people can’t help but smirk at this statement. But all humor aside, Mr. Gates’ little mis-prediction has had a huge impact on the way we think about computer memory, leading to a “too much is barely enough” mentality. Just last year we witnessed the standard memory configuration for pre-built computers jump to 256MB RAM, thanks to new Windows versions and plummeting memory prices.
A similar development is evident in the graphics card market. The original monochrome graphics adapters only needed about 2KB of memory. The first video cards that really deserved the name were the VGA boards, which offered a resolution of 320×240 pixels at 256 colors or 640×480 pixels at 16 colors, requiring 32KB of video memory. Then S-VGA entered the picture and quickly became the accepted standard. EGA, on the other hand, was never able to gain a foothold and left the picture very quickly. The next logical step brought boards with 64, 128, 256 and 512KB of RAM. At the time, the larger memory sizes were needed mainly to render higher resolutions and not for the extra functions video cards have since added. For example, a card needs 6MB to display a resolution of 1600×1200-32.
With the advent of 3D accelerators (the first of which were really more “decelerators”), the video memory was burdened with the additional task of buffering the 3D data. A RIVA 128 with 4MB was only able to render a 3D scene up to 800×600 pixels – in 16Bit color, mind you! The follow-up card offered 8MB and was able to render 3D up to 1024×768 pixels. Starting with the 16MB generation of cards, the video memory began to be of only secondary importance for 2D resolutions. The quality of the card’s RAMDAC, and consequently the supported refresh rate, now determined the maximum attainable resolution.
Fast forward to the present, and conventional wisdom has it that a 32MB video card is sufficient to play current 3D games. The only feature that would truly require more memory would be anti-aliasing. Nonetheless, cards of NVIDIA’s GeForce2 Pro and Ultra generations and ATi’s RADEON line began shipping with 64MB. Considering the relatively modest (super sampling) FSAA performance, this was more than enough. Even today’s high-end cards like the ATi RADEON 8500 and NVIDIA’s Ti500 don’t need more than that.
Yet, true to the trend, the first companies have now begun selling Titanium cards with 128MB of RAM. Interestingly, it’s not the flagship Ti500 chip that’s receiving the royal treatment, but the more budget-oriented Ti200. Probably, the Ti500’s faster memory requirement made this option too expensive. So, does the additional memory translate into additional performance, or is it just another clever PR strategy some manufacturers are trying to see if it will help sell more cards to enthusiasts? That’s what we want to find out in this article.
Leadtek’s entry into this new niche is the WinFast Titanium 200TDH 128MB. Gainward sent us the GeForce3 Power Pack!!! Ti/500 TV Jumbo in two versions. The first is a standard version, the second a faster “Golden Sample.” The latter runs at higher speeds and is equipped with a more powerful fan. We’ll cover the differences between the cards in more detail later, though. Aside from the additional memory, the 128 MB Ti200’s are identical to the 64MB versions. They also use NVIDIA’s reference clockspeeds of 175MHz (GPU) and 400MHz (memory), except for the Gainward GS.
Gainward
The Taiwan-based company Gainward has recently added a GeForce3 Ti200 card with 128MB of RAM to its product line. Codename: Jumbo! We’ll also be taking a look at these cards, namely the GeForce3 Power Pack !!! Ti/500 TV Jumbo and its Golden Sample variation (Jumbo GS). In Gainward nomenclature, Golden Sample is synonymous with improved performance and overclockability. To this end, Gainward uses 4ns memory instead of the usual 5ns modules, and clocks the card at 210/480MHz, up from the standard 175/400MHz. There is a third member of the 128MB family (the GF3 PP !!! Ti/500 XP Jumbo “GS”), which is basically a TV Jumbo “GS,” with an additional TV-In.
Don’t let Gainward’s naming scheme confuse you. Despite the Ti/500 name, the board really uses an NVIDIA GeForce3 Ti200. Equally confusing is the fact that Gainward’s real Ti500 line carries the Ti/550 moniker.
Let’s check the specs. Technical data for The Gainward “TV Jumbos” is as follows:
Graphics Card | Hardware | Software |
GeForce3 Power Pack !!! Ti/500 TV Jumbo | GeForce3 Ti: 175 MHz 64 MB, 5 ns, 400 MHz (DDR) GPU Fan Memory Heatsinks TV-Out (CX25871-13) |
Driver-CD WinDVD v3.0 |
GeForce3 Power Pack !!! Ti/500 TV Jumbo Golden Sample | GeForce3 Ti: 210 MHz 64 MB, 4 ns, 480 MHz (DDR) GPU Fan Memory Heatsinks TV-Out (CX25871-13) |
Driver-CD WinDVD v3.0 |
Visually, the Golden Sample version of the TV Jumbo is identical to the standard model, aside from the more powerful fan, which also runs more quietly than that of the standard card. Gainward equips the GS with 4ns memory. The standard version has to make do with 5ns modules. Additionally, the Golden Sample is also clocked higher, running at 210MHz core and 480MHz memory frequency. Unlike Gainward’s previous forays into “out-of-the-box” overclocking, these higher frequencies are set in the card’s BIOS and are no longer a software overclock through drivers or utilities. Performance-wise, this puts the GS in a league with the real Ti500s (240/500MHz). And since overclocking is what the Golden Sample line is all about, rest assured that there’s still a lot of room for overclocking. The standard Jumbo TV sticks to NVIDIA’s reference speeds of 175/400MHz, however.
Gainward, Continued
The TV-Out is handled by a Conexant CX25871-13 encoder chip, allowing a maximum TV-Out resolution of 1024×768. Sadly, displaying the signal to both the monitor and the TV at the same time is not possible. Also, the PAL image is boxed in by the black frame on the TV that has become a trademark of all NVIDIA cards. To make use of features like overscan or dual view, you’ll need to get a utility like TV-Tool.
The cards share the same software bundle. In addition to the drivers, Gainward includes its own suite of utilities, called “EXPERTool.” These offer advanced display settings for the monitor, color adjustment and gamma correction, as well as an overclocking tool. Last but not least, a retail version of WinDVD 3.0 also ships with the card.
Leadtek
Leadtek’s product line also includes two 128MB cards, which differ only in memory speed. One card carries 5ns memory while the other ships with 4ns modules. We’ll be taking a look at the 4ns version. True to their TDH byname, both cards offer TV-Out, DVI and Hardware monitoring functions.
Specifications for the WinFast Titanium 200 TDH are as follows:
Graphics Card | Hardware | Software |
Leadtek WinFast Titanium 200 TDH 128MB | GeForce2 Ti: 175 MHz 64 MB, 4 ns, 400 MHz (DDR) GPU / Memory Heatsink TV-Out (CX25871-13) Hardware Monitoring DVI LEDґs |
Driver-CD DVD Software Gunlock Dronez |
Visually, the card’s most prominent feature is, without a doubt, its enormous, beautifully milled heatsink. It covers almost the entire circuit board, cooling both the GPU and the memory, and gives the card a very refined look, as well. The fan speed, core voltage and GPU temperature (measured with a dedicated thermal probe) can be monitored during operation using Leadtek’s utilities. The status LEDs on the top right-hand side on the front of the card are more of a gag than a useful feature. These three LEDs inform you whether the card is powered, if there is an error, and if the card is running in AGP 4x mode. While, in theory, the idea is quite sound, the execution doesn’t seem to have been thought through completely. Since the card is installed into the computer upside-down, you’d need a mirror to check the LEDs. Gainward should consider placing them on the back of the card instead of the front.
Leadtek, Continued
Leadtek uses the same Conexant CX25871-13 TV-Out encoder chip as Gainward. Therefore, they also share the same problems: no screen-filling PAL output (overscan), no dual view.
Leadtek complements its drivers with its own very colorful (not to say psychedelic) suite of tools. These include an overclocking tool, hardware monitoring and some monitor color and gamma calibration tools. Aside from the software DVD player, Leadtek also includes two games, both retail versions. Zetah Games’ DroneZ (http://www.dronze.com) already uses the GeForce3’s pixel and vertex shaders. Being a bit older, Rebellion’s Gunlok (http://www.gunlok.com) offers no such support.
Test Setup
Hardware Socket 478 |
|
CPU | Intel Pentium 4 2200MHz MHz 400 MHz QDR FSB |
Motherboard | ASUS P4T-E Intel i850 |
Memory | 256MB 400MHz RDRAM (2x128MB) |
Hard Disk | Seagate 12GB ST313021A UDMA66 5400 U/min |
Graphics Cards | |
Gainward GeForce3 Power Pack !!! Ti/500 TV Jumbo |
Chip Clock: 175 MHz Memory: 128 MB DDR-SDRAM Memory Clock: 200 MHz Driver w9x: Gainward v21.81 |
Gainward GeForce3 Power Pack !!! Ti/500 TV Jumbo Golden Sample |
Chip Clock: 210 MHz Memory: 128 MB DDR-SDRAM Memory Clock: 240 MHz Driver w9x: Gainward v21.81 |
Leadtek WinFast Titanium 200 TDH | Chip Clock: 210 MHz Memory: 128 MB DDR-SDRAM Memory Clock: 240 MHz Driver w9x: Leadtek v21.83 |
GeForce3 Ti500 | Chip Clock: 240 MHz Memory: 128 MB DDR-SDRAM Memory Clock: 250 MHz Driver w9x: NVIDIA v23.11 |
GeForce3 Ti200-64/128 | Chip Clock: 175 MHz Memory: 128 MB DDR-SDRAM Memory Clock: 200 MHz Driver w9x: NVIDIA v23.11 |
Driver and Software | |
DirectX Version | 8.1 |
OS | Windows 98 SE, Version 4.10.2222 A |
Benchmarks and Settings | |
Quake3 | v1.17 OpenGL with HW Transformation Support (Demo001) |
Aquanox | DirectX 8 Game |
3DMark 2001 | Synthetic DirectX 8 Benchmark |
Aquanox
Aquanox is a DirectX 8-based game and uses both pixel and vertex shaders, according to developer Massive Entertainment. Since it uses large high-detail textures and very complex scenery, it is a perfect benchmark to see if the Ti200 can benefit from the extra memory. After all, if 128MB are going to going to have any impact at all, it’s going to be in a complex game with a modern 3D engine.
Aquanox is completely unimpressed by the extra 64MB of memory. The 1fps difference we see here and there is trivial and well within the margin of error. The only card that can really distance itself from the rest is the Gainward Golden Sample, which is, of course, a direct result of the higher clockspeeds.
Quake 3
Quake 3 already runs very smoothly on 32MB cards. Therefore we shouldn’t expect to see any difference between the 64MB and the 128MB cards, all other things (namely clockspeed) being equal. Nonetheless, for the sake of completeness, here are the results:
The results show exactly what we expected. Thanks to its higher clockspeeds, the Gainward GS once again performs almost like a Ti500.
3D Mark 2001
Mad Onion’s 3D Mark 2001 is a synthetic DirectX 8 benchmark. As such, it is filled with complex scenes, large textures, and high-polygon models. In theory, the extra memory could be a real advantage here.
So much for that theory! There are no differences between the cards in 3D Mark 2001. Not even the individual scores show any kind of variation that couldn’t be explained by a margin of error.
FSAA
128MB could make a difference with FSAA enabled, theoretically (again). However, NVIDIA’s GeForce3 no longer uses the memory-hungry super sampling technique, but the more efficient multisampling, which requires a lot less memory than super sampling.
Nonetheless, the extra RAM should help out here. Here’s why: 64MB cards only support 4x FSAA up to 1280×1024 because of memory constraints. A 128MB board, however, can reach 1600×1200.
Once again, the 128MB cards show no real gain. While they do allow 4x FSAA at 1600×1200, this is only of academic interest, considering that the framerate remains somewhere in the 20’s.
Conclusion
These results speak very clearly. 128MB are practically useless in current games. It’s safe to say that getting a 128MB Ti200 is not worth the extra money. Any advantages these cards have are purely theoretical. While I concede that 4x FSAA is possible up to 1600×1200 on a 128MB card, the expected framerates of roughly 25fps make this setting practically worthless. I rest my case.
Nonetheless, keep one thing in mind. Back when Quake3 was released, many people, experts included, claimed that 16MB on a TNT card would be enough, and that the 32MB of a TNT2 card would be useless. Sounds a lot like that “640KB” idea, doesn’t it? Looking back now, they were obviously proved wrong. Since then, no one has ever doubted the usefulness of 32MB. Could we be at a similar point again now, with 128MB cards? Hard to say. Id Software’s next big title is surely going to set some new standards, as early videos have already shown. The question remains whether the Ti200 can exploit this potential in complex game engines or if the memory advantage is lost for lack of sufficient rendering power.
To summarize: none of the cards tested here is a bad choice or in any way disappointing. The Leadtek WinFast Titanium 200 TDH earns points for its extravagant and elegant cooling and its hardware monitoring functions. Gainward’s “GeForce3 Power Pack!!! Ti/500 TV Jumbo Golden Sample,” on the other hand, offers the higher performance. Then again, both companies offer comparable 64MB models as well…
In the end, it’s a matter of price. The 4ns 128MB Leadtek card costs about $35-40 (more than the corresponding 64MB card). A Ti500, on the other hand, will set you back a full $200 more. Gainward’s pricing scheme is similar. So, what’s the verdict? If you want to be on the safe side and wish to be prepared for future games, then a 128MB card is a solid choice. Otherwise, stick with a 64MB version and invest the money you save into other components – like system memory, for example. After all, you can never have to much RAM.