GeForce4 Titanium
So, what does GeForce4 Titanium chipset really offer over the GeForce3 Titanium series? Firstly, the clockspeed of both memory and core are higher, which is no great surprise:

Of course, there's not a linear correlation between clockspeed and video performance. New chips are usually more efficient and contain new features which, when software is written to take advantage of them, can provide more substantial performance gains than the clockspeed changes would indicate. On nVIDIA's product page and technical page for the GF4 Ti, they list a few main features, my translation and comments below:

• nfiniteFX II Engine - this is the core technology of the card, featuring such buzzwords as dual programmable Vertex Shaders, faster Pixel Shaders and 3D textures. There's more info here but the main concern with these new features is whether, and when, software will know to take advantage of them. This document explains how this technology can be used to draw more realistic furry animals, for example. From my perspective at the moment, the main purpose of this technology is to "unlock" some more sections of the 3DMark benchmark - it knows about the special features and tests them, but an older game won't. However, as software is released or updated to use these features, they will - as shown by the nVIDIA demos - be able to create more realistic, more immersive characters and environments.
• Accuview Antialiasing - Anti-aliasing is in itself not new with the GF4 Ti, it's a technique used to smooth edges onscreen for a more realistic, less artificially-sharp appearance and has been around for a few generations of video cards now. It helps prevent the "jaggy" effect of diagonal lines or curves displayed onscreen. The GF4 Ti allows some extra anti-aliasing modes which are intended to give comparable visual quality without the huge performance hit normally associated with full-screen anti-aliasing. FSAA warrants an entire article to itself, with visual quality (screenshot) comparisons between the various methods as implemented by various cards as well as benchmarking - for this reason I will cover antialiasing in a separate article.
• nView display technology - I cover this in some detail on page 3 of this review.
• Lightspeed Memory Architecture II - There's a brain-popping technical PDF about this here and there's quite a few separate technologies involved, but the basic purpose is to allow the GPU more efficient use of the 10.4GB/sec bandwidth available across the 128-bit (DDR) memory bus. This is achieved by using 4 separate memory controllers, 4 separate data caches and techniques such as data compression, pre-charging sections of the memory array that are likely to be used soon and rapid clearing of the z-buffer.

Performance:
Buzzwords are all well and good, but this is the bit I imagine most of you want to see - the magic numbers. So, let's get into it. The testbed configuration is as follows:

Hardware:

• Soltek SL-75DRV5 KT333 motherboard (also from Altech, review soon)
• AMD AthlonXP 1600+ (1.4GHz, 133x10.5)
• KingMax DDR400 SDRAM (running at 333MHz @ CAS 2.5)
• IBM 75GXP 30GB ATA-100 7200rpm hard drive
• Enermax EG651P-V(E) 550W ATX power supply

Software:

• Win2K SP2
• DirectX 8.1
• VIA 4-in-1 drivers version 438(2)v(a)
• DetonatorXP drivers version 28.32

3DMark2001SE from MadOnion is of course the standard DirectX benchmark these days. We ran it at 3 resolutions and the results speak for themselves:

As you'd hope, the Ti4600 card streaks well ahead of the older GF3 and the "Value" chipset, GF4 MX440.

Quake3Arena from id Software is a popular game in itself but also serves well as an OpenGL benchmark. The engine is also used for many other, newer games, so despite Q3's age we still consider it a valid tool for comparison. We used our standard OCAU Slayer demo, check that page for full info about the benchmark. Default settings were used for High Quality 800x600 and Fastest, but for 1600x1200 we set Geometric Detail to High and the Texture Detail slider was dragged all the way to the right.

Here the results are closer, but as the resolution increases, so does the Ti4600's lead. The "Fastest" test shows the cards equal, but that of course is an unusably-low resolution and detail setting, only really serving to show that the testbed was identical for all 3 cards. The real limit there is the CPU, but as the resolution increases the video card has to do more work and the slower cards get left behind.

So, a healthy performance increase from the new chipset over the old - admittedly, the GF3 Ti500 (no sample available for comparison) has been shown elsewhere to be a smidge faster than the "classic" GF3 compared above, but there's not a lot in it. The GF4 Ti4600 definitely takes the Speed King title for now and Asus's implementation leaves little to be desired in that regard. Of course, we couldn't just leave it at stock speeds, though..