3dfx Voodoo4 4500 AGP Review

by FastSite
11/03/2000 | 12:00 AM

A year ago 3dfx promised to launch three new graphics cards right after the company announced its VSA architecture and VSA-100 chips, which were supposed to give birth to a new graphics cards family. Dual-chip Voodoo5 5500 was assumed to be the main product of this family. 3dfx started promoting this graphics card in the beginning of summer. We have already tested this graphics card (see our 3dfx Voodoo5 5500 Review) and we arrived at the conclusion that this product wasn't worth the hopes all of us had pinned upon it. Nevertheless, 3dfx still had the chance to restore its reputation by launching a faster four-chip Voodoo5 6000 graphics card. However, even though 3dfx promised to start selling this monster in July, we do not see it yet on the shop shelves. Moreover, we really doubt that this graphics card will come out at all one day. So, the only chance for 3dfx to regain its popularity appears a single-chip Voodoo4 4500 AGP aimed at sub-$150 graphics market.<%BANNER[article]%>

Well, today we managed to get hold of this graphics card and we will do our best to find out whether it will be able to compete with other cards from the same price group. Among its major rivals we should probably mention NVIDIA GeForce2 MX, Matrox G450 and STM Kyro. As for the first one, GeForce2 MX is a really high-performance low-cost solution, based on GeForce2 GTS architecture, which fastness is restricted only by the reduced memory bus bandwidth. The other competitors can hardly boast many advantages. The 3D performance of G450 is even worse than that of G400 that's why it will hardly manage to compete with Voodoo4 4500. As far as STM Kyro is concerned, the problems with the drivers seem to prevent this solution from finding its application in the nearest future. So, what do we have left? Only NVIDIA GeForce2 MX. Therefore, we will compare the performance of our today's hero with that shown by NVIDIA's low-cost offspring.

Closer Look

Now let's take a look at 3dfx Voodoo4 4500 graphics card.

As you can see, this card is almost twice as small as Voodoo5 5500. However, it is not only the PCB size that matters here. Unlike Voodoo5 5500, Voodoo4 4500 has only one VSA-100 chip and is equipped with 32MB SDRAM. There are 4 64-bit memory chips by Hyundai featuring 6ns access time and supporting 166MHz working frequency (HY57V653220B). As far as we know from our previous experience, these memory chips can work at higher frequencies as well offering quite a lot of opportunities to overclocking fans. Today Voodoo4 4500 sells for about $150, which is almost twice as cheap as Voodoo5 5500. No wonder since Voodoo4 4500 is designed with only one VSA-chip and hence needs twice as few component parts as its elder brother.

Moreover, if you look at the card, you will definitely notice the absence of the power supply connector and all the elements dealing with the voltage regulation, which makes the graphics card design really nice-looking. In fact, Voodoo4 4500 consumes much less power than Voodoo5 5500 (almost twice as little) and hence dissipates less heat, which is also a positive factor.

Well, before passing over to our today's hero, Voodoo4 4500, we would like to say a few words about the VSA-100 chip our graphics card is based on. VSA-100 is equipped with 2 pipelines capable of processing two pixels per clock with multitexturing disabled and one pixel per clock in multitexturing mode. So, Voodoo4 4500 architecture appears a bit similar to that of Voodoo3 family. Although unlike the Voodoo3 cards, VSA-100 acquired 32bit color and large textures support. For more details on VSA-100 you may see our 3dfx Voodoo Scaleable Architecture Preview and A Bit More on Voodoo Scalable Architecture article).

So let's compare the features of 3dfx Voodoo4 4500 with those of NVIDIA GeForce2 MX, which also falls into sub-$150 price group since the cards based on this chip cost around $130.

As you can see, GeForce2 MX and Voodoo4 4500 have a lot in common. However, in multitexturing mode the pipelines of VSA-100 get combined, which causes a 50% fillrate drop. As for GeForce2 MX, its pipelines feature two texturing units each, which doesn't result into any fillrate reduction when two textures are laid over one pixel. That's why speaking about equal performance level of both graphics chips will make sense only in case of laying one single texture over one pixel. Unfortunately (or fortunately), almost all the today's games work in multitexturing mode and hence Voodoo4 4500 looks a bit less attractive in them.

Here we can add only one thing: 3dfx' brand technologies, such as T-Buffer and hardware FSAA, do not work on Voodoo4 4500 because they need at least two VSA-100 chips. Also Voodoo4 4500 doesn't support Dual Monitor Output and has no TV-Out, unlike GeForce2 MX. So, the only feature absent by GeForce2 MX, which Voodoo4 4500 can boast, is the support of API Glide that is getting somewhat outdated already.

We would also like to stress that Voodoo4 4500 doesn't have HW T&L engine, which isn't so badly needed for the today's applications, but which may appear a must for the games of the future. As for the 3dfx Geometry Assist driver, which allegedly allows implementing hardware T&L on Voodoo4/Voodoo5 graphics cards is none other but a software emulator, working only in Direct3D on AMD Athlon based systems.

Performance

For our tests we used the following system:

• Intel Pentium III 866 CPU;
• ASUS CUSL2 mainboard;
• IBM DTLA-305020 HDD;
• 256MB (2x128MB) PC133 SDRAM by Hyundai;
• Windows 98.

For our investigation we selected the following gaming applications:

• Quake3 Arena 1.16n. We tested the graphics cards in demo001 and quaver to check their performance in OpenGL mode.
• Unreal 2.26. We used timedemo to test the graphics cards in API DirectX and Glide.

We decided not to test our cards in Unreal Tournament, because in this game the performance mostly depends on the CPU rather than on the graphics card.

As we have already mentioned, we will compare the performance of Voodoo4 4500 with that shown by NVIDIA GeForce2 MX based graphics card, namely, MSI MS-8816 (StarForce 816).

The tests were run with the latest drivers, which we managed to successfully download from the manufacturers' web-sites: v.1.03 dating back to 9.25.2000 for 3dfx Voodoo4 4500 and Detonator3 dating back to 9.20.2000 for NVIDIA GeForce2 MX.

We didn't check the graphics cards performance with anti-aliasing enabled, since it is implemented on the software level by both cards and hence will cause a nearly 50% performance drop. In this case, the gaming performance will appear below the acceptable level.

So, here come the results.

So, what do we see here? Voodoo4 4500 lost in all resolutions. If we consider the percentage of the performance shown by the graphics cards in these benchmarks, we will get that in 16bit color Voodoo4 proves about 37.9% slower than GeForce2 MX and in 32 bit color - 34.8% slower.

However, let's not make hasty conclusions. To check whether these results are stable, we suggest looking at a bit more complicated quaver demo, which loads the CPU as well as the graphics processor a bit harder using larger textures, such as 512x512, for instance.

The percentage in this demo shows us the following: GeForce2 MX is all in all 37.5% faster than Voodoo4 in 16bit color and 35.6% faster in 32bit color.

Well, GeForce2 MX wins nearly all the competitions. The gap between this card and our Voodoo4 4500 appeared so tangible due to GeForce2 MX' architectural peculiarities, which allow this graphics card to process two pixels per clock against only one pixel per clock by VSA-100. So, 3dfx seems to have failed one more project, doesn't it? Wait! What about Unreal? This is exactly the game, which was developed particularly for 3dfx chips.

Now we are passing over to Unreal tests. We used Unreal v.2.26 flyby demo. All the tests for GeForce2 MX were carried out in D3D and for Voodoo4 4500 - in Glide and D3D.

At 800x600 Voodoo4 4500 managed to surpass GeForce2 MX in both: D3D and Glide modes, while at all other resolutions the card appeared not so fast. Almost everywhere Voodoo4 4500 proved a bit faster in Glide mode rather than in Direct3D.

Since Glide doesn't support 32bit color, we will refer to Direct3D only here.

GeForce2 MX appeared incredibly cooler than Voodoo4 here and the higher was the resolution, the greater appeared the gap between these cards. If at 800x600 the performance difference mad only 28%, then at 1280x1024 this value grew up to 64%! And at 1600x1200 Voodoo4 4500 refused to work in Direct3D at all. And we do not know what is to blame for that: Unreal game or Voodoo4 drivers…

Scalability

We have tested the cards with a pretty powerful CPU, however in order to get a better idea of the dependence of the performance on the CPU we decided to check the card working in a system with a slower one: Celeron 566. The results were taken at a low resolution set to 800x600. As for Voodoo4 4500, its performance dropped by 54.3% in 16bit color and by 25.5% in 32bit color. GeForce2 MX lost 59.4% and 28.1% correspondingly. Comparing these numbers, you will see that GeForce2 MX is much more CPU-dependent than Voodoo4 4500.

Overclocking

Having tried almost everything about the graphics card we had at our disposal, we finally turned to overclocking. The card got easily overclocked with the help of a standard "Overclock" utility from 3dfx. Having installed it, we got a new properties page in the display menu:

This engine was exactly the one to help increase the working frequency. Note that since the memory and the graphics processor of Voodoo4 4500 work at the same frequency equal to 166MHz there is only one single engine for both. Though we managed to make the card work at 190MHz memory and chip frequency, the performance was very unstable and the system crashed several times (because of VSA-100 overheating, probably). That's why we set the working frequency 1MHz lower, i.e. to 189MHz. Here is what we obtained when testing an overclocked Voodoo4 4500:

Overclocking made Voodoo4 4500 about 16.3% faster in Quake3 in 16bit color, and about 17.8% faster in 32bit color mode. As for Unreal, the performance grew by nearly 13.2%. However, even this pretty tangible performance gain didn't let Voodoo4 4500 catch up with GeForce2 MX.

Conclusions

Well, we have every right to state that 3dfx Voodoo4 4500 failed to prove up to our expectations. Compared to NVIDIA GeForce2 MX, available for the same sum of money, this graphics card has the only advantage over its competitor - Glide support. However, we very much doubt that this advantage will have any determinative effect, especially since there are not so many games supporting this API now. Only the PCI version of this graphics card may appear more or less demanded, because it will allow upgrading the graphics in systems without the AGP slot. Therefore, we would recommend 3dfx to reduce the cost on its Voodoo4 4500 down to $100, otherwise, it may suffer a total fiasco. Of course, the market will force 3dfx to drop prices one day, however, we would consider it much cleverer not to set the initial cost that high. Voodoo4 4500 needs a good start…

We also hope that the next 3dfx product, aka Rampage, due in the end of 2000 or in the beginning of 2001, will return at least a bit of 3dfx' popularity.