Asus V8440 128mb GeForce 4 Ti4400
Manufacturer: ASUSTeK Computer Inc. 
Purchasing: GB£215 or US$275 (average, subject to change).
Reviewed: 6th August 2002

Introduction

This review follows on from the initial look at the GeForce 4 Ti4200 reference board reviewed two months back. Whilst the chipset on that graphics card, and the one on this card are pretty much identical (as far as features go), the board reviewed here is the next-fastest model available. It is also important to note that this particular graphics board can be bought in the shop - the reference board previously reviewed could not be bought.

Due to the obvious similarities of the two cards, the main emphasis for this review is the complete package - what you get when you hand over your hard earned cash. Not just the pro's and con's of this chipset versus the next one.

To Summarize

The GeForce 4 is nVidia's current competitor for the consumer-level 3D graphics crown. It's long term rival, ATI, has the Radeon 8500 (reviewed here). Matrox have also recently re-entered the consumer level 3D graphics market with the parhelia 512 graphics board.

The GeForce 4 comes in two main flavours - the MX series and the Ti Series. The two GeForce 4's reviewed on this site have been from the better Ti series (the MX's are cut down budget level versions). The Ti series is then sub-divided down into 3 further categories - 4200, 4400 and 4600, which progressively get faster (4400=10% and 4600=20% faster than the 4200). ASUS offers a graphics card based on all 3 revisions of the Ti chipset: the v8420, v8440 and v8460.

The GeForce 4 is a fully compatible Direct3D 8 graphics card - it lacks some of the more advanced features present on the Radeon 8500, yet it generally appears to have a convincing lead in the performance stakes. Compared with the original GeForce 1 cards (2 years old now) performance has increased almost 4 fold - from 8.7 million triangles per second to 31.4 million triangles per second and 427 million texels per second to 2106.9 million texels per second.

The chipset makes use of some extremely clever technology - and with the new programmable shader architecture it allows for a huge number of effects to be written by the programmer(s). nVidia have also recently pioneered their 'Cg' language to further extend these features - special effects once the preserve of traditional ("offline") 3D renderer's only can now run smoothly in real-time.

The following table lists some of the vital statistics (as published by nVidia) on the GeForce 4 Ti series.

There are three areas that the nVidia engineers have been improving throughout the whole GeForce family (1,2,3 and 4):

Geometry processing - known as the Hardware Transform And Lighting (T&L) engine, capable of offloading a huge number of calculations from the CPU. To the end-user this basically results in much higher performance, to the developer this allows for far more complicated geometry to be rendered - a far richer environment.

Memory efficiency - this isn't one that's as well known as T&L, but deals with the now huge amount of data needed to describe these more realistic scenes. 136 million textured vertices would require around 4gb of storage space alone! Factor in texture reading/processing and suddenly the system->AGP and internal data structure becomes a very limiting factor (rate determining step if you like). The GeForce 4 chipset implements the "Lightspeed Memory Architecture II" (LMA2) technology to help solve this problem, using a crossbar memory controller and loss-less compression amongst other clever techniques they can greatly improve performance.

Image Quality - this is one of the current "buzz words" surrounding 3D graphics, we all know that high raw performance is possible, but high performance with excellent image quality is another thing. Enter the "nFiniteFX II" engine - vertex and pixel shaders to an experienced D3D programmer. We also get Accuview anti-aliasing to give the impression of a much higher-definition image than is actually rendered. The latter effect can be used without much hassle to the programmer, but shaders are a very complex topic - one for experienced/advanced programmers.

These areas have all taken a sizeable leap forward with the version 4 chipset; performance has improved over that of the GeForce 3 (it's predecessor) and programmers now have the right tools (Cg and shaders) to produce movie-quality special effects.

V8440 Performance

The ASUS V8440 is the main target of this review, and the first stop is a look at its performance. 3D Mark 2001 is the current standard for benchmarking the performance of video cards, the end result is a simple score - the higher the score the better. BUT, it will also output several other very useful statistics giving a good insight into how well the card will really perform.

Due to previous reviews, the results for the V8440 can be compared directly with it's main rival - the Radeon8500 and a GeForce 4 Ti4200 reference board. If nVidia's numbers are correct, the Ti4400 chip on display here should be roughly 10% (1.1x) faster than the Ti4200 previously reviewed. Here are the overall scores:

Test System: 700mhz Athlon Thunderbird (Socket A), 288mb PC100 RAM, Windows XP professional

The scores aren't as impressive as you might expect. A 1.1x speed increase should give scores of  6394 and 5921. The actual results gained were roughly 5% faster (1.05x) for low resolutions and 2% (1.02x) for higher resolutions. Both of the GeForce 4 cards were tested using the same computer, same operating system and same drivers (Detonator 29.42 WinXP reference).

Also to note, the out-of-the-box drivers for the V8440 would indicate that they are based on nVidia's 27.42 drivers, and performed noticeably worse than with the 29.42 drivers. Performance was actually lower than that of the Ti4200 in both cases - it has to be assumed that the 29.42 drivers being newer have optimized the cards performance somewhat.

 Individual test results pan out as follows:

1024x768x32

I've highlighted the "winner" of each individual test in blue - of 17 tests performed, the V8440 is the fastest in 13 (3/4 of the tests). In the 4 where it doesn't come out top, it is in a close 2^nd place. The test scores for all 3 graphics boards in the 4 "game" tests used by 3DMark2001 indicates that the processor is often to blame for poor performance; I have a 700mhz Athlon - not a fast processor by today's standards, but far from being a slow processor. 

Given the initial statement that the Ti4400 should be around 10% faster than the Ti4200 it would seem certain that the processor is not keeping up with the graphics card. Taking all the non-game tests (where the V8440 came out top) and doing some simple math, it is clear that it is often more than 10% faster. As two examples: Environment Bump Mapping is 23% faster (1.23x) and [1 light] high polygon counts are 12% faster (1.12x). With this in mind, it is even more clear that you really need a decent CPU/motherboard to take full advantage of these new faster graphics cards. The 3DMark2001 score is heavily influenced by the game-scores, which is why these single-feature test results don't translate to a higher overall score.

The next test of performance is one only really relevant from an end-user perspective: over clocking. Over clocking is something only recommended for the most advanced (and obsessive) users as it can lead to permanent damage to the graphics chip. ASUS does provide a simple-to-use "tweak" utility allowing us to set the memory and processor clock frequencies.

I don't profess to be the greatest over-clocker in the world (in my opinion it's not worth the risk), but I was able to add 10mhz to both the memory and processor clocks (560/285 respectively) and gain about 180 points in 3D Mark. Pushing it too far (295mhz processor and 610mhz memory) would only remain stable for 10 minutes before crashing - enough to register over 6000 points in 3D Mark for high resolution. Using a better cooling system (all I had available was a desk-fan and I removed the PC's case!) it may be possible to get a stable, over-clocked, system.

V8440 Features

Features are what should be important when it comes to choosing a 3D card these days. It is almost impossible to buy a truly slow graphics card - instead it is a toss-up over features, software and connectors.

The V8440 is a physically big card, considerably bigger than the Ti4200 reference card and the Radeon8500; at first it proved quite a challenge to fit into my computer (I have a smaller MicroATX case), but after moving a few cables around it fitted in. You'll probably only get this problem if you have a small case, or one already crowded with add-in cards and drive cables.

A view of the v8440 Back-Plate

On the back-plate of the card we're given three connections - DVI Output, TV In/Out, VGA monitor output. The latter is probably of most use to most; but the other two connections are very useful to have available. TV output would allow for large-screen projections AND the ability to record demo's to video. TV input allows for video capture and recording. DVI-Out allows for a higher quality signal to be sent to special flat-panel monitors. Of particular interest is that the package includes a small DVI-VGA adapter - which allows you to get two VGA outputs on the card. 

With two VGA outputs, you can run two monitors simultaneously (assuming you have 2 monitors) - this is an absolutely brilliant tool for any programmer or artist. I took the time to write an additional module for my forth-coming game to use the 2nd monitor (it's not hard). Using my older 2nd monitor as a desktop extension I positioned a simple window to output debugging information, and played the game normally on the first monitor. Doing this allows me (and any other testers) to watch real-time debugging information - namely frame rate graphs, process timing values and any error/warning messages. All this data is normally outputted to the log-file, but it is far harder to match up the timing values in the log file than to watch it in real time out of the corner of your eye. Dual-Monitor testing is really something that needs to be experienced to truly appreciate it; the only downside is that the system is notably slower - which should be expected given that the graphics card is having to run two displays at the same time.

As mentioned, the package comes with a DVI-VGA connector, it also comes with a split-cable for the TV in/out port... one side being a standard S-Video port, the other being a video-input port ('7pin S Compatible').

These are the only physical/hardware advantages provided by the V8440, but we also get a selection of software bundled in. The main show-piece is a full copy of the game "Aquanox", not a game I'd highly rate - but that would be the subject of a review on another site. We also get a CD with a set of demo's/movies to show off the capabilities of the graphics card - nothing hugely ground breaking, but it's worth a quick look.

Purchasing

I haven't included any particular links for purchasing this card, you can't buy them directly from ASUS - you'll have to go to an online store or a high-street retailer. At the time of writing this review you'll be looking to pay just short of US$300 for this particular model (GB£215). Expect these prices to fall significantly over the next few months, particularly towards the end of this year when the next round (D3D9) of cards is released onto the market.

Conclusion

This is definitely a great graphics processor, in a very capable package.

It would appear that the overall performance of this particular graphics card isn't as impressive as it maybe should have been, but short of the faster Ti4600 variants it is one of the fastest cards available today. The chipset has great features and this variant carries the maximum 128mb of video memory. The out-of-the-box capability for dual monitor support is great for professionals/serious developers. Given that it's also a very smooth purple and silver color (see image at top) - a refreshing change from the boring green PCB's of most add-in cards.