RSS Feed

HCW Tech Blog

For the latest info on computer hardware, tech, news, video games, software tips, and Linux, check out our new improved front page: HCW Tech Blog

Reviewed by: Bryan Pizzuti [09.05.02]
Manufactured by: MSI
MSRP: $199 

Discuss this review in our forums! NEW! You DO NOT have to register to comment on a review!


The Test

We're testing this particular TI4200 against its immediate predecessor, the GeForce3TI200 (This particular one is the Chaintech SE version with 128 MB of RAM). Here is a table of vital statistics for each card:

Card GF3 Ti200 128MB NVIDIA GF4Ti4200 Ref. MSI VTP128
GPU Speed 175 MHz 250 MHz 250 MHz
Rendering Pipes 4 4 4
Texture Units/Pipe 2 2 2
RAM 128MB 64MB 128MB 128MB
RAM Speed 400 MHz 500 MHz 444 MHz 500 MHz
Memory Bandwidth 6.4GB/sec 8GB/sec 7.1GB/sec 8GB/sec

Unfortunately, the card we received would not detect any more than 64 MB of RAM on the board, even when playing around with the memory clocking. We aren't sure if this is an incompatibility with our testing rig, or if it's a defect in the card, but it's a defect we can live with, since we can test this card as an ordinary 64 MB TI4200. The performance numbers you see from these tests will be the MINIMUM numbers that can be obtained from the purchase of one of MSI's TI4200 products. Also, when buying this particular model of card (The VTP) performance may be greater, due to double the amount of RAM running at the same speed.  (Does this explain why the RAM is actually running at the speed intended for the 64MB cards? I hope people don't end up buying a 64MB card when they paid for a 128MB! -Ed)

The test system
AMD Athlon-C 1.2 GHz (266 MHz FSB)
512MB DDR RAM (266 MHz)
Turtle Beach SantaCruz
Pioneer 12x DVD
Microtek 17 inch flat panel
Windows XP Professional

Detonator driver version: 30.82

Synthetic Benchmarks

DirectX7 - 3DMark2000

At this point, we only include this for reference purposes, but it's still important to know how well a card will do in a non-DirectX8 situation. The addition of pixel and vertex shaders to a card doesn't always guarantee that it's base polygon processing is up to snuff.

As you can see, there isn't a huge gain in performance between these two cards. When playing older games only, money might be saved by going with a cheaper GF4MX product. But like a lot of people, we don't stop with DirectX7.

DirectX8 - 3DMark2001SE

MadOnion's 3DMark2001SE is the current standard for DirectX8 testing, and we tend to break this down by a selection of categories, including raw multitecture fill rate, and a test of the pixel and vertex shaders.

As you can see, the performance gains in DirectX8 functions are a bit more noticeable, being that much greater.

Z-Culling - PowerVR's VillageMark

This benchmark is made by PowerVR. Considering PowerVR hasn't brought out a new GPU in a while (and might never, since STMicro, their parent company, is selling it's video business) and we aren't involving a PowerVR card here, the use of this mark shouldn't tilt in any card's favor.

As you can see, NVIDIA has GREATLY improved their Z-Culling engine in the GF4 GPU.

OpenGL - Vulpine GLMark

This OpenGL benchmark supports all of the GL extensions that support DirectX8 features like shaders and such. All of these features were activated for this run.

This is a long benchmark, as well as a grueling one (it brings a lot of older cards to their knees). As you can see, even with this in mind, there's quite a performance difference.

Pixel Shader Performance - NVIDIA's ChameleonMark

This is the first time we're using this benchmark, which is made by NVIDIA. Since both of these cards are NVIDIA based, using it should be quite fair for judging pixel shader performance. The GF4 shows quite a jump, indicating a lot of improvement in the pixel shader engine.

Next Page: (5)