<%BANNER[top_768x90]%>

<%BANNER[banner_468x60_h]%>

3dfx Voodoo Scaleable Architecture Preview

Well, it finally happened. 15 November 1999 3dfx officially announced its new VSA technology (VoodooScaleable Architecture) aka Napalm and the first graphics processors and cards supporting it.

by FastSite
11/26/1999 | 12:00 AM

Well, it finally happened. 15 November 1999 3dfx officially announced its new VSA technology (Voodoo ScaleableArchitecture) aka Napalm and the first graphics processors and cards supporting it.<%BANNER[article]%>

Only a very lazy and indifferent user or observer didn't blame 3dfx for something during the last six months. Totell the truth there were enough reasons for that, however, the company didn't give in and kept working on its newoffspring. First of all, 3dfx had to choose the correct strategy in the current situation. The people at the head ofthe company as well as the engineering stuff had to compose a clear image of future 3dfx products, namely to define theirfunctional peculiarities. The new products were supposed to be a combination of features, which could ensure their popularityamong the customers and let 3dfx at least remain one of the leaders in the 3D graphics mass market and at the most become theonly indisputable leader in this market. Since most 3dfx's competitors could already boast new products with a large range ofvarious functional innovations, 3dfx engineers would have to create something at least yielding to none of these. However, thiscould be enough only to stay in the market for another while. If 3dfx really wanted to resume its leadership, they had to dotheir best and to provide their coming product with something barely new, which would please the users and at the same timebe absent by the competitors. Or they had to accumulate all the best and most progressive ideas of the competitors and tointroduce them in one single product, which should be still better than all the analogues. 3dfx company started gettingready for the presentation of the new technology and the products working with it in summer already. Anyway, the realinterest towards this mysterious product from 3dfx had been existing for quite a long time. Some people were very anxiousabout the newcomer from their favorite company, while some others also looked forward to this event secretly hoping that3dfx would fail and disappear from the market clearing the way for more successful and strong competitors. About a few weeksbefore 15 November you could see the following on 3dfx web-site:

We were invited to visit 3dfx site at 7 p.m. (Moscow time) and to find out what surprise 3dfx had prepared for us. Theyseemed to really like the idea of building up tension and drawing as much public attention as possible according to NVIDIA'slive example. It's quite logical actually, because all successful ideas are always borrowed by competitors. But the mostinteresting thing about it was the fact that already by 2 p.m. (Moscow time) curious users managed to search 3dfx Europeanweb-site and to find a secret press release with all the info on the magnificent event about to happen five hours later.By the way, when time "X" came 3dfx site got overloaded with requests and was hard to access for the first 30 minutes thougha bit later things returned to their natural course. It was probably due to a lot of information available at differentcomputer sites, which appeared sufficient for some inquisitive users and hence reduced the amount of those who hunted forthe latest news on the matter exactly at 7 p.m. We think that this leak of information in no way spoilt the triumph of thenew VSA architecture and new graphics accelerators. With the time we will undoubtedly understand if this event has reallybecome a landmark in the development of 3D-graphics mass market. Now we can just state that all our forecasts came trueagain and take a closer look at what 3dfx offers us this time. Nevertheless, you should feel free to celebrate this day asa holiday or just to ignore this event if you like, it's up to you.

Almost all the products competing in the field of 3D-graphics support 32-bit color in 3D-graphics, AGPtexturing and larger textures, as well as a few other features, which have already become a de facto standard. Actually,these and some other standard parameters of the today's graphics accelerators seem to be absolutely clear to all of us. Ofcourse, you can argue about the necessity of some functions, which appear not supported by certain applications, but thesedisputes are absolutely useless. If there is a set opinion that something is a necessary requirement, especially speakingabout functional features, then all the products simply must have it, otherwise - asta la vista, baby. A new 3dfx productalso had to meet all the standards existing in the today's market. But what else could they offer their customers besidesthe standard combination of features? Note that we still consider an integrated geometric coprocessor to be an extraordinary,new and progressive feature of mass graphics processors. Surely, they could also introduce hardware T&L support, following theexample of NVIDIA and S3, which were the first to integrate geometric coprocessors into their chips intended for the mass marketof 3D-graphics. No doubt, this is a way to the guaranteed success and we have every reason to believe that in half a yearor a year we will be able to see a lot of games requiring hardware T&L. However, though it is a good way it is not the onlypossible way to develop further. After T-Buffer technology had been announced it turned out absolutely evident that verysoon 3dfx products would boast T-Buffer hardware support. Since we mentioned this technology, we would like to remind youof T-Buffer and its main principles.

The main idea of T-Buffer technology is to make it possible to operate several frame buffers, which allows using somespecial effects. The most important digital effect is full scene spatial anti-aliasing (or simply FSAA). Besides, thetechniques of such effects as motion blur, depth of field, soft shadows and soft reflections have already been developedas well. In the first place T-Buffer technology should be regarded as a tool. Able to operate the rendering process inseveral frame buffers simultaneously, application developer can easily create his own cinematic effects. You don't needany special interfaces to use full scene spatial anti-aliasing, while all the other effects applied with the help ofT-Buffer have to be supported by some interfaces. However, this support can be provided by OpenGL and Direct3D extensions.So, the main thing T-Buffer is initially aimed at is the visualization of higher-quality 3D-graphics.

Well, after the announcement of 3dfx technology with a very intriguing name - VSA-100 it was clear that the main stresswas made on fastness and quality, and "Scalability" appeared the key word here. T&L hardware support was postponed for sometime.

Actually, it's not so difficult to understand the idea of 3dfx engineers. What's the use of making a product, whichfunctional features will be utilized to the full extent only in half a year or maybe even in a year? It's up to NVIDIAand to S3 to choose their way and to decide if they really want to play the role of a pushing force promoting the ideaof T&L hardware support and speeding up the game developers' efforts in this direction. Besides, they will also be busycompeting with each other. Moreover, in spring both: NVIDIA and S3 will launch a new generation of their GPUs and then3dfx will also join the competition. Of course, it will happen in case the slightest doubts about T&L support vanish andall cons and pros become evident. However, if at that time it still remains unclear whether the mass market really needsgraphics cards with the geometric coprocessors on board, 3dfx will be able to postpone the launching of its T&L hardwaresupport till autumn of the year 2000. But if in spring of the year 2000 T&L hardware support suddenly turns out a necessaryrequirement for all the graphics accelerators of the mass market, 3dfx will simply equip its graphics cards with the externalgeometric coprocessor. Here we would like to mention that we think it seems more logical to expect an external geometriccoprocessor from 3dfx for several reasons. First, according to the info provided by some non-official sources, 3dfx isalready working on the use of an external geometric coprocessor from Mitsubishi - IMPAC-GE. The major problem about it isthe drivers support. And the primary advantage - minimal developing expenses, because IMPAC-GE chip is already designedand has been existing in the professional graphics market for over a year now, which means only one single thing: it is avery well checked solution. Second, they can introduce a couple of changes into their products of VSA family and add ageometric coprocessor there. However, this solution will have a tangible drawback: too much money needed for development,design and check, because everything will have to be done practically from the very beginning (of course, if they haven'tstarted working on it yet). But the important advantage of this solution will definitely be the potential computing capacity,i.e. the performance. The matter is that the key feature of VSA architecture is scalability, in other words it means that therecan be several graphics processors installed onto a single card. If each of them possesses an integrated geometric coprocessorwe will get a super computer in a format of an ordinary AGP card. On the other hand, no one is against installing severalgeometric coprocessors onto one PCB, the only thing that really matters is the sufficient power supply and free space onthe card for additional components. At the same time, the main principles of SLI technology can be applied to the parallelcalculations carried out by several geometric coprocessors. We don't know yet which way 3dfx will finally decide on: timewill show. Nevertheless, we would like to quote 3dfx CTO Scott Sellers: "3dfx will have HW T&L when it matters in the market".

VSA Architecture

The abbreviation VSA is derived from Voodoo Scaleable Architecture. Here we have to point out right away that 3dfxdecided not to give up a well known and promoted Voodoo brand name as we had actually expected. At least it helped tosave time and money on advertising and promotion, which is especially important for 3dfx because its Voodoo3 cards arenot selling well enough to cover all the expenses and the new cards are expected to appear in the market only in thecoming spring - after the New Year hot sales time. The first chip with VSA architecture was called VSA-100. Simple, eh?

So, what does this scalability of the new architecture actually imply? VSA Scalability allows from 2 to 32 graphicsprocessors to work parallel. There are two ways to make use of this parallel work. Now we will specify what exactly ismeant.

The first way implies the use of SLI technology (scan line interleave) very well known since the times of theclassical Voodoo2 chipset. At that time SLI looked as follows: two graphics processors rendering different lines ofone and the same frame, which in the end formed an image. And all the chips shared the same frame buffer for thispurpose. After that the formed frame was taken from the frame buffer and displayed on the monitor. VSA architecturesupports new SLI realization, which is deprived of some restrictions typical of the older SLI technology, such as theparallel use of only two graphics chipsets and no support for the resolutions starting from 1024x768 (now it supportsall resolutions up to 1600x1200). Moreover, the version of SLI technology embodies an absolutely different idea. Takinginto account that today's graphics accelerators were initially designed to use AGP interface and, as to 3dfx, VSA-100chip was specially optimized for working with AGP, from now on SLI regime supports the parallel work of several graphicsprocessors located on one and the same PCB. In case of VSA-100 we are considering, this new version of SLI-technologyensures parallel operation of up to 32 graphics processors. And so every chip in SLI regime can work on a certainsuccession or band of lines of a single frame, where every succession like that can include from 1 up to 128 lines.Moreover, the number of lines in this succession can get dynamically changed. Here we mean that every chip forms thelines following one another and making a band and these bands in their turn form the final frame. For example, if thereare four chips working in SLI regime, each of them can create a band of 40 lines the final frame will be composed of. Ofcourse, it is also possible to use alternating lines instead of line successions. For instance, let's take the samesituation with four processors in SLI regime and assign a CN index to each of them, where N=1, 2, 3, 4. Then C1 chipwill be responsible for the lines S=1, 5, 9, etc. or S=4*M+N where M=[0…1200] and N is the number of the graphics processor.

Why did they have to use exactly these bands, which allowed changing the number of lines they contained? The thing isthat this possibility helped to arrange the most optimal SLI utilization depending on the color depth, the amount ofpolygons used for the scene and their location. And it is also possible that a small polygon will be solely formed byone graphics chip within one line band. In this case the use of SLI won't be as efficient. On the other hand, if the graphicschip forms more than one line at a time, i.e. it forms a band, the system gets more evenly utilized and the texturecache will be used with higher outcome. In the end they decided to use the bands with a dynamically changing number oflines from 1 to 128, which provided even system utilization in SLI regime.

We would like to mention that each of VSA-100 chips used in SLI has its own frame buffer for rendering lines and linesuccessions (bands). After that the data from all these frame buffers is submitted to the general frame buffer where thecomplete frame is created and transferred to the monitor. Moreover, each VSA-100 chip is equipped with its own memory busand local frame buffer and supports up to 64MB local graphics memory. A part of the available memory is used for a framebuffer (with double of even triple buffering) and the rest - for storing textures and Z-buffering. In case the availablememory turns not enough, 3dfx suggests resorting to one of the two supported texture compression methods: DXTC or FXT1.The way the work of graphics processors is arranged helps to avoid memory bus overloading. By the way, the total memorybandwidth significantly increases due to this solution, namely due to the possibility to evenly distribute the data streamsbetween the graphics processors working in SLI regime. That is why we shouldn't disregard the possibility to use standardSDRAM/SGRAM memory, which is very likely to happen, actually.

The second way implies the use of T-Buffer technology. In this case every graphics processor is responsible for thewhole image frame, which it forms in its frame buffer from the very beginning to the very end. And after that T-Buffercombines the contents of several frame buffers and, as a result of these manipulations, displays on the monitor as thefinal frame. We won't go into details about the main advantages provided by the T-Buffer. We would like just to mentionthat the main thing is the improvement of image quality and hence better perception of the end-picture. The image qualitygets better due to the hardware full scene anti-aliasing, and the perception - due to a number of cinematic effects such asdepth of field, soft shadows, etc.

Well, in our opinion there is only one more important question left. Will it be possible to simultaneously use T-Bufferand SLI? Actually, it looks as if it were possible, but it will work differently for two-, four- (and more) VSA-100 chipsetconfigurations. And here is the reason. Hardware full scene anti-aliasing uses four pixel samples for the end-pixel of thefinal image. If the card is assembled only with two VSA-100 chipsets, they can form exactly four pixel samples per time step,but not more, in other words it will be just a waste of time if we use alternating lines or line bands. On the other handhowever, T-Buffer effects can be applied not to the complete frame but to its certain parts. Then it is probably possibleto alternate lines in order to optimize memory bus utilization, i.e. it is possible to use T-Buffer and SLI regimesimultaneously. If our suppositions are correct then there should be at least four VSA-100 processors working simultaneouslyin order to make optimal use of T-Buffer and SLI, i.e. to apply hardware FSAA (full scene anti-aliasing) and some otherspecial effects without any tangible performance drop of the whole system. In case of only two VSA-100 processors, theuse of SLI technology will be limited, or it won't be involved at all when T-Buffer is active. However, the opposite isalso possible in this case: the activated SLI regime aimed at achieving maximum performance while the T-Buffer isn'tutilized to its full extent.

So, if we regard these ideas as true, the user is supposed to get the opportunity to choose between high performanceand high quality and image realism. There is something to prove this point, actually. According to 3dfx, the user will havethe possibility to disable T-Buffer (in other words, to disable hardware FSAA effect) for the sake of maximum fillrate andhence max fps. Let us explain. If the user selects the regime providing higher graphics quality and image realism (itautomatically implies that he decides on T-Buffer), it will never mean sacrificing the gaming performance. At least3dfx promises that high graphics quality and realistic images achieved due to such effects as motion blur, depth offield, etc. won't tell on the fillrate. If we make it simpler, it will look like that: the user gets not only highquality graphics and the whole bunch of realistic effects but also high frame rate, of about 60fps at 1024x768, andT-Buffer technology. If the user doesn't give a damn about the cinematic effects and he is quite satisfied with theimage quality he's getting from NVIDIA, Matrox or ATI products, but he lacks fastness and longs for crazy fps rates,he should just disable T-buffer and put the SLI regime on, as simple as that.

We would like to draw your attention to one more important thing. Compared to MAXX technology from ATI, SLI from3dfx doesn't suffer from lags. They owe the absence of this unpleasant problem to the fact that all graphics processorsworking in SLI regime never store up the ready frames and work only on the current ones.

And now it's high time we talked a bit about the main features of the first chip supporting VSA architecture.

VSA-100 General Characteristics

3D Acceleration Features

Supports all the functions typical of Voodoo3 family and some more:

Software Support

The first thing that arouses numerous questions is the 0.25 micron technology chosen for VSA-100 chip manufacturing.And it is inQ1 of the year 2000, when all the competitors are about to shift to 0.18 micron! S3 is already using 0.18 micronfor its Savage2000 chips and GeForce256 GPU by NVIDIA is manufactured with 0.22 micron. Scott Sellers answered this questionin the interview to "Thresh's FiringSquad". Here is a quote: "NVIDIA uses a 5-layer metal 0.22 process, which results primarilyin smaller die size and lower power consumption. By using a six layer 0.25e (enhanced) micron process, 3dfx is able to shrinkthe die accordingly by cramming more onto the same relative area, and retain improved product yields". Well, we can't doanything but take it for granted and console ourselves with the fact that 3dfx realize what they are doing.

But why did 3dfx use 14 million transistors in their VSA-100, we wonder? Savage2000 from S3 has 12 million transistors butlike VSA-100 it also has 2 pipelines with 2 texturing blocks on each. However, Savage2000 also possesses an integrated geometriccoprocessor, while its counterpart doesn't have any. Or maybe it does, but we simply weren't told about it? Just try to imaginethat somewhere in March 2000 3dfx suddenly announces that VSA-100 has a geometric coprocessor, but they preferred to keep silentuntil the supporting software was completed. Like it? Well, sounds not bad, actually, but these are just our absolutelyungrounded dreams.

We would like to stress that 3dfx borrowed some developments from its previous graphics processors families and implementedthem in VSA-100 chip design, which was expected to provide absolute compatibility with all the today's applications optimizedfor these processors. Although we are not quite sure that with this card we will be able to do without the patches for gamesintended for the previous Glide versions. That's why if one day we decide to enjoy a couple of old games with a new coolgraphics accelerator, we will still have to wait for these patches to come out.

Another very important issue, unfortunately, still lacks clear-cut explanation. It deals with the local memory busbandwidth and memory interface. The thing is that VSA-100 chip provides pretty high fillrate and if the card is equippedwith two or more VSA-100 chips, the fillrate gets even higher. And the higher rises the fillrate, the higher appears thememory bus utilization. If the local graphics memory bus bandwidth becomes insufficient, then the potential fillrate willremain potential forever and will never turn into reality. The memory may simply fail to transfer the required amount ofdata within the available time, which will result into a significant performance drop and the user will never get the awaitedfps at higher resolutions. The question about the local graphics memory type turns out especially crucial for 3D graphicsin 32-bit color regime supported by VSA-100. Nevertheless, there is a way-out: a new version of SLI technology, which allowsus to beautifully solve the problem of memory bus utilization. The memory bus utilization gets lower when every VSA-100 chipcreates just a part of the lines forming the final frame. So, a 128-bit memory bus per VSA-100 chip will be just enough.

Besides, nothing is clear yet about DDR SDRAM/SGRAM memory support. We don't see any obstacles in the way towards thissupport in the near future, for instance. Moreover, the price of the graphics cards on VSA-100 is far not so low, especiallytaking into account that they are expected to appear in the market in Q1 2000. However, if the product offered at this pricewill be provided with the on board DDR SGRAM local graphics memory, products from 3dfx will definitely gain in popularity.Besides, on the other hand, 3dfx can simply drop the price without any radical technical innovations.

We would like to point out that the memory bus working frequency will be again synchronized with that of the graphics coreas we saw by Voodoo3 products.

As we have expected, VSA-100 chip supports 32-bit 3D-graphics with 24-bit Z-buffering and 8-bit stencil buffer. All theyear everybody has been blaming 3dfx for the absence of this functional feature, and finally its support is introduced.However, the question about the quality of 3D graphics in 32-bit color still remains open, because 3dfx hasn't yet gotenough experience here. In fact, the graphics models should be of relatively high quality especially since there is thewhole lot of supported functions contributing to quality increase: single-pass tri-linear filtering, color blending regimeswith alpha channel, bump mapping and multitexturing. Anyway, let's hope that the quality won't disappoint us.

By the way, since we came to speak about 32-bit color depth, there arises a really important question: will the performanceof the graphics subsystem suffer a lot in this case? 3dfx promises that even in 32-bit color regime the user will be able toplay at the average of 60fps, which is more than enough to feel at home, as to 3dfx. And in this situation it doesn't matterany more how many fps you can get when running the same application at the same resolution in the regime with 16-bit colordepth. Everything seems quite logical. If you are satisfied with the performance and quality then no problem. Hopefully verysoon we will be able to check on practice if our suppositions are right.

Note that VSA-100 supports 32-bit textures with the resolution of 2048x2048 pixels and AGP texturing. These are alsoanother two functional peculiarities, which we longed to see in 3dfx's new products. And again have patience: we'll seehow it all works a bit later...

VSA-100 chip supports two texture compression technologies: DXTC and FXT1. The first one is a version of S3TC licensedby Microsoft and included into DirectX. And the second one was developed by 3dfx Company on their own and is absolutelyopen. There is one pretty interesting observation: the fact that FXT1 is supported by VSA-100 chip guarantees that DXTCwill be also supported in hardware. The thing is that DXTC uses one compression algorithm, while FXT1 - four. Besides,since FXT1 is supported through hardware it seems quite logical to add hardware support for DXTC as well especially sincethe product will only gain in popularity and appear more attractive to users. We can only hope that it is really as coolas we imagine.

However, it is still not very clear which bump mapping method is supported by our VSA-100 chip. Now we can just supposethat it supports Embossing and Dot Product. And again we can do nothing but wait and wait...

MPEG2 video decoding is also worth mentioning. Unfortunately, we can't draw a happy picture here. The data transformationfrom the planar into the packed format and that's all. Nothing more. Not a word about overlays or about motion compensation.We only dare hope that at least overlays are supported, but who knows.

Graphics Cards on VSA-100 Chip

3dfx decided to clearly distinguish between the solutions for the mass and professional market. For the mass market 3dfxsuggests making the whole family of graphics cards based on VSA-100 chip:

Voodoo4 4500 PCI/AGP:

In fact, Voodoo4 is the card from the Voodoo3 family but in addition it supports AGP texturing, larger textures,32-bit color depth in 3D-graphics with 24-bit Z-buffering and 8-bit stencil buffer. If this card had appeared in themarket in spring 1999 or at least in autumn 1999, its prospects could have been pretty clear and understandable. Itcould have competed with the graphics cards based on NVIDIA Riva TNT2 Ultra and Matrox G400 MAX chips. It could havesold very well and 3dfx could have at least retained its leading positions in the market. But Voodoo4 graphics cardsare expected to enter the market only in the first quarter of the year 2000 and the only indisputable advantage theywill be able to boast in front of the competitors, which have been present in this market for about a year already, isAPI Glide native support. We believe that Voodoo4 graphics cards have simply come too late. And we can't quite make outwho will buy these graphics cards for 179 bucks. If its price drops below $100 then the card will be able to conquer itsterritory, otherwise...

Voodoo5 5000 PCI:

And why did they announce only a PCI version, we wonder? What prevents them from developing an AGP version as well?Maybe, AGP optimization is very special? To tell the truth, it's a rather strange card. 3dfx says that they takecare of the users with older systems, which don't have an AGP port. And at the same time 3dfx positions its cardsfor the systems with 700MHz CPUs. Could you show us a system with PCI slots only working on Athlon or Pentium III EB?

If there is no AGP version, the market share this card manages to win for itself will be very limited. Besides,the price strikes as too high. This card will mainly compete with the products from the price group <$200, and wereally doubt if there will appear a lot of users willing to pay extra for hardware FSAA and cinematic effects if allthe other functions and parameters are the same as by cheaper competitors. And for sure nobody will be willing tooverpay for Voodoo5 5000 if it turns out that the use of T-Buffer (in particular the use of hardware FSAA) willtangibly influence the card's general performance. On the other hand, if the user gets the average frame rate in hisfavorite applications of about 60fps at the resolutions up to 1024x768 and with 32-bit color depth, then the card willdefinitely look much more attractive. Anyway, most users prefer playing at the resolutions set up to 1024x768 that iswhy we have just to check all the hypotheses and to make a decision.

Voodoo5 5500 AGP:

This is an enhanced Voodoo5 5000: with 32MB more memory and AGP interface, and of course at a different price -$70 higher. So that the AGP version of the graphics card could work normally each VSA-100 chip requires 32MB memory.But in this case it appears unclear why they managed to do with only 16MB local graphics memory in case of PCI version.Take a look at the picture of Voodoo5 5500 card. Do you see a white connector in the upper right corner? This is a standardpower supply connector for the PC internal peripheral devices. We have already seen something very similar to it on CanopusSpectra 7400 based on GeForce 256 GPU. Frankly speaking, there is nothing to be surprised with in this solution. Two VSA-100graphics chips consume together about 25W. Using the power directly from the power block guarantees reliable supply and as aresult stable work of Voodoo5 5500 graphics card. This card is most likely to enjoy the highest demand and popularity amongthose, who want to have FSAA and other cinematic effects provided by the T-Buffer in games. If 3dfx reduces the card's price,success is granted.

Voodoo5 6000 AGP:

This one is the most powerful of all the announced cards. It can boast 4 VSA-100 chips, 128MB local graphics memory andan impressively high price. In fact, it looks as if only Voodoo5 6000 graphics card would be able to demonstrate all theadvantages of hardware FSAA without threatening to worsen the performance of the whole graphics subsystem. First of all itbecomes possible due to 4 VSA-100 chipsets working in SLI regime. You will probably be able to feel the difference betweenVoodoo5 5500 and Voodoo5 6000 only at higher resolutions. For example, 3dfx promises that with this card you can play at1600x1200 in 32-bit color regime at an average of 60fps using all the T-Buffer effects including hardware FSAA. And arelatively big local graphics memory shouldn't confuse you because it is then distributed among the four VSA-100 chipsand is not used as one entire piece. And as for the available memory, 32MB seems to be quite enough to satisfy the needsof each graphics processor, namely for buffering and textures. Nevertheless, despite such enormous price the card you willget for this money will boast just incredible fillrate. In fact, if nothing extraordinary happens if none of the dark horses,such as BitBoys for instance, suddenly works a wonder, then Voodoo5 6000 AGP becomes the first graphics accelerator intendedfor the mass market capable of displaying on the monitor over 1Gpixel per second. We'll see what it really looks like onpractice.

By the way, you can easily notice on the picture of Voodoo5 6000 AGP an Intel chip. It is most likely to beNon-transparent PCI-to-PCI bridge. What's the use of this non-transparent bridge on the graphics card? Thedescription of the Non-transparent PCI-to-PCI bridge says that it allows presenting a certain subsystem to thePC CPU as a single device. Besides, the non-transparent bridge also hides all the resources of this subsystem fromthe CPU. So, it is possible to place a really powerful graphics subsystem composed of several parallel working chips.Does the use of this bridge imply that Voodoo5 6000 won't support AGP texturing? We are not sure. We think that thisnon-transparent bridge serves to help recognize the graphics card representing a complex graphics subsystem as a singledevice. It would be very stupid to give up all the AGP advantages only for the sake of four VSA-100 chips working at thesame time.

We would also like to note that according to some unverified info Intel microchip is none other but an i960 processor,which is so universal that it can easily perform as an AGP-to-AGP bridge or fulfil some other specific functions.

Voodoo5 5000 graphics cards can't boast the same bridge (at least we can't see it on the pictures) probably because incase of only two VSA-100 chips there are no problems with the graphics subsystem and the system CPU. And the access to AGPmemory can be divided on the software level, i.e. through the drivers. ATI for instance resorts to software in its MAXXtechnology, when they have to share the access to AGP memory between two graphics chips located on the same graphics card:and there is no need to use any bridges in this case.

And now it's exactly the time we dwelled on the power supply. As we know, Voodoo5 5000 graphics card with 2 VSA-100 chipsrequires additional power, but how do things stand with the four-chip card then? Each VSA-100 chip consumes 12-13W, which isabout 30% more than used to require Voodoo3. Voodoo5 6000 has twice as many chips: four VSA-100 chips and besides, 128MB localgraphics memory and an additional Intel bridge. In this respect, 3dfx suggests a very outstanding solution. Voodoo5 6000 graphicscard will be provided with its own power supply unit called "Voodoo Volts" with the power of 100W. And it is really a separatepower unit, which will be plugged into a separate standard 220 or 110V outlet and provide Voodoo5 6000 card with the requiredpower. The exterior of this power supply unit looks very much like that of a notebook. It will be connected to the graphicscard through a connector located near the VGA-out. This solution will guarantee stable operation of the device in any PC.

Despite huge power potential of the graphics card and the presence of the external power supply unit, the price ($599)still seems to be too high. If 3dfx reduces it at least till $499, they can count on selling a certain amount of thesemonsters to real hardcore gamers.

And for professional use 3dfx recommends solutions from Quantum3D known under the general name of AAlchemy. AAlchemyis a new family of powerful systems for 3D graphics real-time modeling. An AAlchemy graphics subsystem consists of 1, 2or 4 cards with 4 or 8 VSA-100 chips on each and with 32 or 64MB local graphics memory for each chip. AAlchemy graphicssubsystem can potentially provide the fillrate of 3Gpixels per second! They are intended mostly for Heavy Metal systems.However, the cost of these ones is also quite impressive and can be written in four- and even five-digit numbers.

Conclusions

So, let's try to make a shot summary. 3dfx Company announced not just a new VSA-100 chip. It announced a new ideologyof high performance graphics solutions. The mere idea of parallel data processing is very interesting and promising. Infact, it's now too hard to foresee the reaction of the market to these new solutions from 3dfx. Especially since thereare still a few unclear and moot issues. There is only one indisputable thing: it is too untimely to make any definiteconclusions before we see with our own eyes how it all works.

In fact, the users will have to choose between the T-Buffer (including FSAA effect) and hardware T&L in the Q1 2000.Besides, the situation with the geometric coprocessors support in the existing applications will matter a lot and is verylikely to influence your decision. However, you should clearly understand one thing. The bare announcement that this or thatgame supports hardware T&L is not enough. Only if this game run on a graphics card equipped with the geometric coprocessorlooks indisputably better on the user's monitor than in case he uses a Voodoo5, the user will prefer competitors' productsto those from 3dfx. But if most games with the announced hardware T&L support turn out not to make full use of these graphicscards features (in other words, they use the same amount of polygons in the scenes as if they were run on a system without anygeometric coprocessor), then there will be hardly any sense in buying graphics cards with the geometric coprocessor. Really,what's the use of buying a graphics card with the geometric coprocessor if in autumn 2000 we are offered another generation ofthe graphics cards of this type? In this case, we don't doubt that Voodoo5 cards will take the lead because they offer a widerange of features the user can apply immediately in new games as well as in older ones without waiting for anything. Here wemean T-Buffer, namely the major effect it provides - full scene anti-aliasing (FSAA). And that is not all! T-Buffer alsoprovides hardware support for the whole lot of various cinematic effects. They have to be supported in the interfaces,while hardware FSAA is supported by all applications and API. All in all, we are facing a very interesting period of timewhen we will have to make our own choice and to pay for it.

By the way, if 3dfx announced hardware T&L support in its new products the situation with geometric coprocessorssupport in applications could seriously change. And these changes could be directed towards wider spreading of geometriccoprocessors support. And in the meanwhile we can only hope that NVIDIA and S3 will be eloquent enough to convince the gamedevelopers of the necessity to use certain new features of the today's graphics accelerators, such as hardware T&L.

As for the pay-off, we would like to stress once again our idea about the prices set for Voodoo5, which seem to us a bittoo high and even scary. On the other hand, try to remember the prices on Obsidian2 X-24 from Quantum3D, which used 2 Voodoo2chipsets in SLI. This card cost $650, though it had only 24MB local graphics memory and only 2 Voodoo2 chipsets. By the way,even in Moscow these cards sold pretty well, while two Voodoo2 cards went for about $480 ($240 each). So, if Voodoo5 graphicscards are really worth this money, they will sell. Especially since no one can prevent 3dfx from dropping the prices.

And now we can just try to satisfy our curiosity, lay up some money and wait for the first real graphics cards on VSA-100.

<%BANNER[banner_468x60_f]%>