Original Link: https://www.anandtech.com/show/406



Be sure to read Part 1 and Part 2 of our Athlon Buyer's Guide and keep a lookout for Part 4 in the coming weeks

In the last two parts of our Athlon Buyer’s Guide, we talked about what motherboard to buy and your possible options for overclocking the Athlon, but, for the gamer, the most important question is which video card is the best for your new Athlon system. In an ideal world we’d be able to simply recommend the fastest video card and be done with it. Unfortunately, oftentimes the fastest video card is also the most expensive, thus introducing a very realistic limitation to your decision.

In this guide we will be focusing on the major players in the video card market right now, NVIDIA’s TNT2, TNT2 Ultra, and GeForce, 3dfx’s Voodoo3 line, and Matrox’s G400 and G400MAX.

Choosing a Video Card

For an Athlon system, choosing a video card should be no different than with any other system, right? It would be very unfair if, by purchasing an Athlon, you were suddenly limited to only a few choices for video cards.

From a compatibility perspective, all of the video cards are "compatible" with the Athlon platform and the AMD 750 chipset. There are no inherent problems in either that could possibly prevent one of the aforementioned video cards from being incompatible with the Athlon platform. This is very refreshing news because it is one of the things we were quite afraid of when we originally recommended the Athlon back in August. We were left with a very sour taste in our mouths after the initial Super7 video card incompatibility problems, but, luckily, the same does not hold true for the Athlon.

Unfortunately, there have been reports of problems with certain combinations of components and video cards resulting in random crashes, lockups, etc… As far as the TNT2 goes, all of the currently available TNT2 cards work perfectly fine with the Athlon (including the now discontinued Hercules Dynamite TNT2 Ultra clocked at 175/200MHz) on all of the motherboards we have taken a look at thus far (those included in Part 1 of our Athlon Buyer’s Guide). So why is it that some users are experiencing random crashes and lockups with their TNT2 based Athlon systems?

Chances are that the culprit in this situation would be a power supply that is unable to deliver enough current to the system. But why would the power supply work perfectly fine in a Pentium III based system with the same video card? The fact of the matter is that the Athlon draws significantly more current than the fastest Pentium III, and thus its power supply requirements are much stricter than those of other systems. We will focus closely on power supplies in Part 4 of our Athlon Buyer’s Guide, but, for now, you’re probably best off looking at AMD’s Recommended Power Supplies page if you think this may be the problem.

How do you tell whether your lockups are related to your video card not receiving enough current? The easiest way to determine whether or not the AGP slot is receiving enough current for normal operation is to underclock the core frequency of the chip and see if that solves your stability problems. If it does, then the AGP slot is not receiving enough current to run the card at the higher core clock speed. For example, if you had a TNT2 Ultra clocked at 150/183MHz (core/mem frequency), you could underclock the core from 150MHz down to 125MHz and see if that solves your stability problems. If it does not, then the problem could lie elsewhere, possibly in a software/driver configuration problem or possible hardware failure/conflict.

This problem of under-delivering current seems to be most present with the TNT2 Ultra; however, we have seen cases where a G400MAX exhibited similar behavior in a system. The Voodoo3 seems to draw the least amount of current out of the cards we took a look at and had the least amount of problems with running reliably on systems with non-AMD recommended power supplies.

There have been recent discussions about problems with Athlon systems using the GeForce. We have been working on reproducing the problems in lab, but we were able to complete all of the GeForce tests without any problems on our test beds.



2D Image Quality

Of all of the cards tested, the Matrox G400 featured the best 2D image quality with NVIDIA’s GeForce 256 coming in an almost indiscernible second place. The higher speed RAMDACs of the G400 & the GeForce give it the advantage in driving higher resolution monitors (1600 x 1200 and above) at greater refresh rates.

At everything below 1600 x 1200, the 2D image quality of all of the cards was virtually the same. However, keep in mind that the cards all have individual features that may tilt the balance of favor in their direction. Read the individual reviews of the chipsets and respective cards listed in our Video Reviews Index.

The Test

Windows 98 SE Test System

Hardware

CPU(s)

AMD Athlon 700
AMD Athlon 600
AMD Athlon 500
Motherboard(s) ASUS K7M 1.04
Memory

128MB PC133 Corsair SDRAM

Hard Drive

IBM Deskstar 22GXP 22GB Ultra ATA 66 HDD

CDROM

Phillips 48X

Disk Controller

Promise Ultra 66 - Ultra ATA 66 Controller

Video Card

3dfx Voodoo3 3500TV
3dfx Voodoo3 3000
3dfx Voodoo3 2000
Diamond Viper V770 Ultra - NVIDIA TNT2 Ultra
Diamond Viper V770 - NVIDIA TNT2
NVIDIA GeForce 256 SDR Reference Board
Matrox Millennium G400 32MB
Matrox Millennium G400MAX 32MB

Video Drivers

3dfx Voodoo3 3500TV
3dfx Voodoo3 3000
3dfx Voodoo3 2000

1.03.00

Diamond Viper V770 Ultra - NVIDIA TNT2 Ultra
Diamond Viper V770 - NVIDIA TNT2
NVIDIA GeForce 256 SDR Reference Board

3.53

Matrox Millennium G400 32MB
Matrox Millennium G400MAX 32MB

5.30.007

Ethernet

Linksys LNE100TX 100Mbit PCI Ethernet Adapter

Software

Operating System

Windows 98 SE

Benchmarking Applications

Business

BAPCo SYSmark 98

Gaming

idSoftware Quake 3 Test 1.08 (OpenGL)



At 640 x 480 the strain on the video card is minimal and the performance is generally dependent on the power of the CPU and how well the drivers for the cards are written.  The GeForce 256 obviously comes out on top, a trend which you will see throughout this article simply because it is the faster chip, but the rest of the competitors perform very close to each other. 

One thing to note is that in 16-bit color the TNT2 Ultra is faster than the G400MAX, but once you make the move to 32-bit color, which consumes considerably more memory, the 200MHz memory frequency of the G400MAX gives it the slight edge over the 183MHz memory frequency of the TNT2 Ultra.  The performance difference isn't huge, but it's a point you'll see repeated as we go through the tests.

The performance all of these solutions offer is more than respectable, but then again, you don't buy an Athlon 700 just to run Quake 3 at 640 x 480.  So let's take a look at what happens when you bump up the resolution to 800 x 600.

Here we start to see some of the effects of the limits of the video card.  Matrox's TurboGL driver built into the 5.30.007 drivers help push the G400MAX slightly above the TNT2 Ultra in board 16 and 32-bit rendering scenarios.  All of the competitors offer above average performance, but it should be noted that the switch to 32-bit rendering leaves only two of the eight cards above the 60 fps mark. 

On the lower end of the graph, the performance difference between the G400, Voodoo3 3000 and TNT2 are all very similar.  Don't let the numbers fool you, chances are you wouldn't be able to tell the performance difference between those three contenders.  The reason that all of the cards perform within a reasonable margin of one another is partially because of the fact that we're testing on such a fast processor. 

At 1024 x 768 (especially in 32-bit color) we finally see some real limits of the cards.  The performance of the GeForce is incredible in comparison to the rest of the competitors; however, once we make the move to 32-bit rendering the performance drops to just about 5 fps faster than the Matrox G400MAX.  Increasing the CPU speed won't help performance too much here, as the limit is primarily the video card.  Increasing the core clock speed of the graphics card is the only way to improve performance significantly at this great of a resolution in Quake 3.  At the same time, overclocking the memory frequency of the card is the only real way to make a significant improvement in the 32-bit color tests. 



We're back to the starting point again, this time with a "slower" Athlon 600 at the helm.   The performance of all the cards is quite respectable but the important thing to look at in the tests as we change processors is how the standings change as the CPU performance decreases.  It's difficult to see any changes at 640 x 480 because of the already high scores, so let's take a look at how the 100MHz decrease in CPU speed affects real world gaming performance at higher resolutions.

The G400MAX begins to separate itself from the TNT2 Ultra by a noticeable margin.  For one thing the TurboGL driver is being used here and helps performance tremendously (the performance of the final drivers with the full and optimized ICD should be identical) and the second factor that affects the G400MAX's performance advantage over the TNT2 Ultra is the fact that the Athlon 600 is such a fast processor.  The G400 itself seems to do much better and truly shine on processors faster than a Pentium III 450, and, of course, an Athlon 600 is considerably faster than a Pentium III 450.  The performance difference in 32-bit color is fairly large, but the GeForce 256, even with its slower memory speed (166MHz vs 200MHz) manages to pull ahead of the G400MAX by a good 10 fps. 

Once again the 32-bit rendering performance gap between the G400MAX and the GeForce is reduced to around 5 fps when the resolution is increased to 1024 x 768.  This is due to the fact that so much memory is being consumed in the jump to 1024 x 768 that the dependence upon a fast memory bus grows as the color depth increases to 32-bits (higher color depths require more memory).  



The Athlon 500 is still a very fast processor, but it is the "slowest" out of all of the Athlons on the market so it becomes our "low-end" test bed in this case (wouldn't it be great to have an Athlon 500 as a "low-end" computer?).   Because of the already high performing nature of the Athlon CPU, you won't notice such a great change in the rest of the tests on this test system. 



One thing you need to remember about Expendable is that the difference in average frame rates will be very low, simply because of the complexity of the benchmark.  So a difference of 1 fps in the Expendable test translates into a fairly large real world performance difference. 

Here's something odd, the GeForce 256 comes in third place in this test, with the two Voodoo3's taking up the first and second places.  Earlier we mentioned that testing at 640 x 480 was a good way to test CPU performance as well as to see how well a driver was written.  As you can see here, even with the latest detonator 3.53 drivers, the GeForce 256 could still use some optimization with older DirectX 6 games.  The performance isn't "bad" but we'd expect more from the most expensive card in the roundup. 

Here the GeForce 256 regains its lead because of the dependency of the performance on the card itself and not on the quality of the drivers.  It is also interesting to note that the three Voodoo3's perform better than the TNT2 Ultra, and the performance difference is quite noticeable.  At the same time, all of the cards performed quite respectably and in 32-bit color the G400 came out on top over the TNT2 and the MAX finished higher than the the TNT2 Ultra.

Another very similar situation, first the GeForce, then two Voodoo3s and the Matrox comes out on top in the 32-bit color tests.



The standings remain basically the same with the two "slower" clock speed Athlon CPUs.  There is some switching around of places towards the lower parts of the graph but nothing major.





Summary

It turns out that the video cards we compared performed virtually identically across the three different CPUs.   While the scores were naturally higher on the faster Athlon 700, the breakdown of fastest and slowest of the bunch remained relatively similar at all three CPU speeds.   The GeForce 256 is the fastest of the group, but we already knew that from our review of the GeForce 256 itself.   

Because of the power of the CPUs themselves, the Matrox G400MAX fairs very well in the tests, often times beating out the TNT2 Ultra.  Not to mention that the G400MAX brings quite a few features to the table (i.e. DualHead) that the TNT2 Ultra cannot deliver.  The faster memory clock of the G400MAX gives it a good boost in 32-bit color rendering situations, sometimes even pulling ahead of the "faster" GeForce. 

The Voodoo3 performs quite nicely, but the Voodoo3 truly shines on CPUs slower than the Athlon where the TNT2 and G400 don't fair as well. 

In the end any of these cards would make for a fine matchup with the Athlon but if price is more of a factor then you may end up weighing your decision differently.  The most trouble-free card to install in an Athlon system would probably be the Voodoo3, but provided that you have adhered to AMD's recommendations you shouldn't have a problem with the other cards.   The highest performing card is obviously the GeForce, but at the same time it is also the most expensive and depending on your setup you may encounter some stability problems with the configuration. 

Also, with the GeForce's Hardware T&L, the card would probably last you longer than any of the other solutions as games that truly take advantage of Hardware T&L start to surface in the coming months. 

Be sure to read Part 1 and Part 2 of our Athlon Buyer's Guide and keep a lookout for Part 4 in the coming weeks

Log in

Don't have an account? Sign up now