by FastSite
11/19/1999 | 12:00 AM
Everybody wants to be a happy owner of the fastest graphics card. There are two traditional ways to make thisdream a reality. The easiest and probably the most traditional way is to go and buy the latest graphics card withthe coolest 3D accelerator on board. To tell the truth it is not the most optimal way, especially for our far notFortunatus's purses. And if we try to keep pace with the rushing technology, we will have to repeat this operationevery 6-8 months. Besides, this will definitely be not so cheap. We believe that not everyone can afford to spend200-300 real American bucks on a graphics card, which can get outdated in 3-6 months already. As for the secondmethod, we called it "traditional" because at least some users resorted to it in real life and all the othersundoubtedly have a very clear idea of what it actually is. Here we mean 3dfx SLI. Are you feeling the magic ofthese letters? Aha, you recalled the past and lovingly patted on the dear minitower case on your table, didn'tyou? Yeah, even the bitterest enemies of 3dfx know what SLI is and what is the connection between this regime andclassical Voodoo2 based graphics cards.<%BANNER[article]%>
To cut the long story short, SLI (Scan Line Interleave) technology allows parallelly installing two graphics cardsbased on Voodoo2 chipset (or just two Voodoo2 chipsets on a single PCB). Each graphics card based on Voodoo2 isresponsible either for even or for odd lines of the displayed frame, which makes the graphics subsystem performanceover 1.5 times higher and as a result it appears possible to play at 1024x768 using z-buffer.
However, SLI technology has a few drawbacks:
In fact, it is not quite correct to call the second modernization method SLI. It would be more exact to say thatit's a method of two or more identical (or different) graphics chipsets used simultaneously. A bit later we willspecify this slightly vague statement.
3dfx's competitors couldn't put up with its laurels received after 3dfx had successfully provided the users with thisnew possibility of their graphics system modernization. The first official attempt to develop something similar belongedto Metabyte (represented by Wicked 3D department). They created PGC (Parallel Graphics Configuration) technology, whichin short implied that there were two absolutely identical or absolutely different graphics cards installed in the system,one of which formed and displayed the upper half of the frame and the other - the lower half. As a result, you could add aPCI graphics accelerator to your AGP one and get a tangible performance gain.
PGC solution also has a few lows:
We would like to note that despite its evident attractiveness and Alienware's license PGC technology failed to winthe market.
ATI Company chose a different way, slowly moving in two directions simultaneously. Concerning the first performancerising method, everything is very simple and understandable here: it implies the development of new graphics chips. Notso long ago they announced RAGE 128 PRO chip, which showed much better parameters and higher performance than those bytheir competitors. The solution provided for the second method they worked on was a new technology known under the codenameAurora, which was later renamed into MAXX (Multiple ASCI Technology). Just in case you forgot: ASIC is Application SpecificIntegrated Circuit.
You may find the whole bunch of reasons to mock at the inventiveness of ATI's marketing specialists, who decided to givetheir new technology such a name: MAXX. The first associations that come to our mind are BOSCH washing machine and a certainMen's magazine. In our opinion, this is exactly the case when the code name appears much more attractive than the onefinally chosen for the product. However, it is not the names and promotion strategies, we are talking about in thisarticle but the technology. So, let's not veer away and find out what MAXX actually is.
First of all MAXX technology is a purely software solution, which allows involving two graphics processorssimultaneously. In fact, it means that the idea of parallel data processing has finally found its way into life. Eachof the graphics processors is responsible for one frame from the very beginning to the very end. And then the readyframes are displayed on the monitor in turn. Theoretically, MAXX technology may be capable of involving more than twographics processors, however we doubt if it is really expedient to introduce configurations like that into life at leastin the near future. And the technique, which allows two graphics chips installed on one and the same graphics card to displaythe completed frames by turns, is called Alternate Frame Rendering (AFR).
We tried to simplify our explanation of the main idea of MAXX technology, and now we suggest dwelling on some particulardetails of this interesting solution. We will take a look at the first graphics card supporting this technology - ATI RAGEFURY MAXX.
In fact, RAGE FURY MAXX graphics card is none other than two RAGE FURY PRO cards placed on the single PCB.
As you can see from the photo above, there are two RAGE 128 PRO chips on one card. Then there are also 64MBlocal graphics memory, the card has AGP 2x/4x form-factor and a standard VGA-out. In reality the mentioned 64MB oflocal SDRAM graphics memory are just a combination of two frame buffers - 32MB for each of the two RAGE 128 PRO chips.Of course, it is not only the frame buffer that uses the local graphics memory. It would be more correct to say that onlya certain part of the local graphics memory is used for the frame buffer (with double or triple buffering). The major partof the local graphics memory serves to store textures.
There is one very interesting question here. Since we know that RAGE FURY MAXX is actually a combination of two absolutelyfully-fledged RAGE FURY PRO graphics cards on one PCB, maybe it will be possible to divide the existing VGA-out into twoparts as well and to arrange two monitors support. However, MAXX technology will support dual monitor configuration onlyin 2D-graphics because of certain inner restrictions. At least as far as we know there are RAGE FURY MAXX graphics cardswith two VGA-outs and ATI is rumored to be introducing dual monitor support in the future. That's why now we can only guesshow it will work, in what regimes and if it will work at all. Hopefully our questions won't remain unanswered.
And now we would like to slightly veer away from our topic. Please take a look at the photo. Do you see two wiressoldered to the PCB over the VGA connector? Frankly speaking, you won't be surprised to see such wires if it is not afinished off-the-shelf product but the so-called sample. All the samples serve to check if everything is working properlyand to try all sorts of modifications. However, it looks as if these two small wires are to blame for the fact that wefailed not only to get this card for testing but even to make normal pictures of it. Well, actually we can understandATI here, however, we don't quite understand their biased attitude towards certain editions and press reps. It turnedout that ATI has a number of "favorites", whom they take into their confidence. Well, if it prohibited to test andphotograph RAGE FURY MAXX, it's OK, but be honest and follow the rules of game yourself! Don't make any exceptionsthen! And in fact this ban appeared valid only for the Russian press. :-( This is what we call "clever marketing policy".Of course, what's the use of paying any attention to Russia if the sales volumes of ATI products are very small here? If theRussian market is of no interest to ATI then the situation is pretty clear. If ATI regards the Russian market as quitepromising then it is the best marketing strategy they could ever think of. They should provide the Russian press with aslittle information as possible, give no samples for testing, promise wonders and keep as few promises as they can. And inthe end a lot of Russian users will soon have ATI products in their systems. We can only sympathize with ATI's competitors,who have chosen an absolutely different strategy and will very soon lose their positions in the Russian market.
OK, enough for our sarcastic comments, back to the point of our discussion. We won't give you the specs of RAGE 128 PROchip. If you want to have a look you may check our RAGE FURY PRO Review for that. And we will start speaking about MAXXtechnology and its practical peculiarities.
In this regime only one RAGE 128 PRO chip is activated, the other one remains uninvolved. And there are hardly anydifferences between this card and an ordinary RAGE FURY PRO. We don't consider it a drawback or an unnecessary restriction,that there is only one graphics processor working in 2D graphics, it would be incorrect. We think that one RAGE 128 PRO ismore than enough for 2D regime. Besides, ATI positions its RAGE FURY MAXX graphics card as the one for dedicated gamers that'swhy if you have bought a gaming card then why not use it for its direct purpose?
ATI recommends RAGE FURY MAXX for those gamers, who own larger monitors and usually play at the resolutions startingfrom 1024x768. Of course, if you look through the reviews on various graphics accelerators, you will easily notice thatat higher resolutions the bottleneck of almost all of them is the local memory bus bandwidth, which tangibly reduces theperformance. Even in 16-bit color depth the amounts of data to be transferred through the local graphics memory bus arejust enormous. So, now let's find out something concrete about the performance of RAGE FURY MAXX in 3D-graphics.
In fact, each of the RAGE 128 PRO chips installed into RAGE FURY MAXX card is used for its direct purpose: it forms pixelscomposing the final image displayed on the monitor screen. MAXX technology allows each graphics chip of the pair to utilize thetriangle setup engine and texture caches in two independent frame buffers to the full extent. That is why when the images aregenerated on the monitor the card works in the most optimal way.
We would like to draw your attention to a very important detail. The today's realization of MAXX technology, i.e. RAGE FURYMAXX, helps to solve the problem of the insufficient local graphics memory bus bandwidth and hence the problem of low fillrate.However, this problem is typical only of higher resolutions. In other words, it means that when working at the resolutions under1024x768, MAXX technology will lose its effect and will provide no performance gain.
Each RAGE 128 PRO chip has its frame buffer, which carries out the rendering of every image frame. And each of the chipsalso has its own integrated RAMDAC. The software, and to be more precise, the operating driver defines, which frame bufferwill provide the frame to be displayed on the screen. And it can be done not necessarily in strictly defined turn: up tothree frames running may be taken from one and the same frame buffer, and only after the third one the second frame buffercan take its turn.
Note that the necessity to switch between different frame buffers can cause inevitable delays, which may lead to a certainlag. The lag is a time period between the user's actions and the graphics processor reaction to them. In case of RAGE FURYMAXX this lag between the user's action and the reaction to it may be long enough to finish displaying a full frame and eventhe following one as well. This is at the best. And at the worst the lag may appear enough for four or even six frames (incase of triple buffering). The thing is that when the frame is transferred to the screen, the next one is usually ready orgetting ready. That's why the reaction to the user's actions may be introduced in the newly formed frame, i.e. in the thirdone only. Unfortunately, only the practical experiment will show whether all this goes this particular way or not.
The operation driver is responsible not only for switching between the frame buffers and for RAMDAC synchronization. Italso controls AGP bus to make sure that each RAGE 128 PRO chip receives the corresponding data from AGP memory. Again, it ispretty evident that all the shifts require additional time and hence the delays will be inevitable. That is why the performancewill hardly double. Nevertheless ATI supposes that the performance of its RAGE FURY MAXX will be 1.5-1.8 times higher thatthat of the usual RAGE 128 PRO based graphics card.
Proceeding from the fact that the clock frequency of RAGE 128 PRO chip equals 140MHz, RAGE FURY MAXX graphics card will beable to form up to 210-252 million pixels per second with multitexturing enabled. Although there are almost no games withoutmultitexturing today, all the graphics chips developers usually mention the potential performance of their products in regimeswithout multitexturing, for marketing purposes probably. In this respect, the potential performance of RAGE FURY MAXX will beabout 420-504 million pixels per second, and the ad says about 500 million pixels per second. The facts coincide. We can'treproach ATI for using these figures, because everybody does the same thing now. However, it will always give us a point fordiscussion.
Note that we used the frequency values of real RAGE 128 PRO chips and the clock frequency of the local graphics memorybus taking into account the parameters of RAGE FURY PRO graphics card we had for testing not so long ago. The graphicscards RAGE FURY MAXX tested by our overseas colleagues have a bit lower clock frequencies: 125MHz graphics core and 143MHzlocal memory bus frequency. We are not sure if those RAGE FURY MAXX cards we will see in retail one day will work at thesame frequencies. ATI may be forced to reduce the frequencies only for some technical reasons or in case problems arise,however, we don't yet see any causes for concern. Now we believe that the final version of RAGE FURY MAXX cards will workat the frequencies not lower than those of RAGE FURY PRO ones.
Well, let's suppose that the potential fillrate of RAGE FURY MAXX graphics card in multitexturing regime is 250 millionpixels per second. Is it much or little? For a better idea you can compare this value with the parameters of competitorproducts: with multitexturing enabled GeForce 256 GPU from NVIDIA has the potential fillrate of 240 million pixels/sec.(with 120MHz graphics core frequency), the competing chip, Savage2000 from S3, has about 250 million pixels/sec. (with125MHz graphics core frequency). You see that RAGE FURY MAXX is not a bit worse than its competitors in terms of fillrate.But there is another very important thing to remember. GeForce 256 and Savage2000 chips utilize only one 128bit local graphicsmemory bus and use SDR SDRAM memory. In our case the situation is a bit different.
In fact RAGE FURY MAXX card has two local graphics memory buses 128bit each with the bandwidth of 2.5GB/sec (at 159MHzfrequency), which makes the total peak bandwidth equal to 5GB/sec. We would like also to point out that ATI declared thistotal bandwidth equal to 4GB/sec, they probably proceeded from the memory bus frequency of 143MHz. Now we will try toexplain everything with the help of numbers.
Suppose, we want to play at 1600x1200 in 32-bit color with 32-bit z-buffering and multitexturing enabled (with thesuperposition of 2 textures, each of 32-bit precision). The average overdraw is 4, and the fps rate - 30fps. With allthis data we get that we need to transfer 16B data through the local graphics memory bus. Take a look at the calculationsbelow. They are very simple. At this resolution and with these fps our graphics processor has to form 1600x1200x30=230400000pixels per second. To do it 230400000x16=3686400000 or about 3.7GB/sec have to be transferred through the local graphicsmemory bus.
Each of the two RAGE 128 PRO chips should create only a half of the required frames, which makes only 15 frames in ourcase. And the utilization of the memory buses (we have two of them 128bit each) is also twice lower than the required one.All in all, 3.7GB/sec have to be transferred through two buses instead of one, which makes 1.85GB/sec through each of them.We think that both 128bit buses of our RAGE 128 PRO chips can easily transfer such amount of data. This is the key advantageof this graphics card over competitors, which provides the performance increase we mentioned.
Of course, these were rather rough calculations since we didn't take into account the delays caused by switching betweenRAMDAC and during AGP access separation. However, the general picture is distinct enough to understand the main principles ofMAXX technology and the factors influencing the performance increase.
One indisputable advantage of MAXX technology is its so-called "transparency" for applications as well as for Direct3D andOpenGL interfaces. In fact it means that the user doesn't need to care whether this technology is supported by the game he isplaying or API he is using. It is the software, in other words MAXX technology, which is in charge of solving problems likethat.
MAXX technology is organized in such a way that the local graphics memory appears divided into two separate parts, whichare practically not connected to each other. Sure if it were possible to use the whole graphics memory as one single texturestorage, it could be really useful. Actually we can't call it a real drawback, because this restriction is made by thetechnology itself. But on the other hand, when the texture memory is limited the specially developed texture compressiontechnology comes to rescue. That is why if you lack memory use texture compression. However, there are still a few questionshere as well. The specs list for RAGE 128 PRO chip says that it supports DXTC, and in fact with the drivers version wecurrently have at our disposal there is no hardware support for this technology.
Unfortunately, we have no chance to test RAGE FURY MAXX graphics card, however, Alex "Sharky" Ross from www.sharkyextreme.comhas already done it and the results he obtained beautifully prove our theoretical ideas.
To illustrate what has been said above we would like to offer you the results of one test (you may find the testing systemconfiguration details here, and the testingresults are available hereand here):
As you see, separate frame buffers and separate local graphics memory buses helped RAGE FURY MAXX to show highperformance at such a high resolution with extreme memory bus utilization.
Besides, it is also worth mentioning that two RAGE 128 PRO chips used in RAGE FURY MAXX graphics card are optimizedespecially for 32-bit color depth regime. The info proving this point is available in out ATI RAGE FURY PRO Review.
So, we have to admit that ATI found a really good way of obtaining higher fillrate. And the undeniable advantage ofMAXX technology is, of course, its universality, since it is a purely software solution. It doesn't actually matterwhat graphics chips will be placed onto one graphics card, the technology will never fail, and the user will enjoythe performance increase. It should be also noted that you will be able to notice this performance gain in all theexisting applications and the only thing you have to do is just to buy a RAGE FURY MAXX graphics card and to installit into your PC. And after that you have to run all your games at 1024x768 and up. However, if you have a smallermonitor (less than 15") and you usually play at 800x600 then buying RAGE FURY MAXX will hardly make much sense. Youwon't get any new advantages, that's for sure.
Speaking about fillrate we can say that RAGE FURY MAXX leaves behind GeForce 256 based graphics cards with SDR SDRAMlocal graphics memory. But GeForce 256 possesses integrated geometric coprocessor, and RAGE 128 PRO doesn't have hardwareT&L support. Well, there is something to talk about here. RAGE FURY MAXX graphics cards with 64MB local graphics memory onboard are expected to cost about $250. This is very close to the price of the graphics cards based on NVIDIA GeForce 256with 32MB SDR SDRAM local graphics memory. What to decide on then?
Actually there are no games today, which could make real use of hardware transformation and lighting. Of course,everybody keeps promising that the games like that will come out by the end of December already. Suppose it's true.Anyway, in spring the number of games supporting hardware T&L will undoubtedly surpass 30. As a result, those of youwho will buy or have already bought a GeForce 256 GPU based graphics card with SDR SDRAM memory will gain an advantageover the owners of RAGE FURY MAXX cards. The latter are expected to appear in the market only in December. And if wetake into account the fact that ATI will probably fail to develop acceptable drivers in time (which has already happenedseveral times), the launching of RAGE FURY MAXX graphics cards may get postponed till January of the year 2000 or even tillspring. And nearly at the same time there must appear the graphics cards based on NVIDIA GeForce 256 using DDR SGRAM, and thecards on Savage2000 will be also already selling. Note that the cards on GeForce 256 equipped with DDR SGRAM local graphicsmemory will go for $250-300 (and maybe even cheaper) for a 32MB version. In this respect a question occurs: who will decideon RAGE FURY MAXX, which only advantage is high fillrate? As to ATI, they consider this card will last for 8-12 months, whilewe think it will be 6-8 months at the most.
If RAGE FURY MAXX had come out earlier, in spring 1999, then it could have had a really impressive success in the marketbecause it would have been a serious competitor to Riva TNT2 Ultra based graphics cards. And now the things get changed soquickly that RAGE FURY MAXX will have to compete with the graphics accelerators known not only for high fillrate but alsofor their integrated geometric coprocessor, which is absent by RAGE 128 PRO. We will have to make a choice between thegraphics cards with and without the coprocessor, since their prices are almost same. As for us, we see no sense in buyingRAGE FURY MAXX, at least for $250. Despite our extremely negative attitude towards graphics cardson GeForce 256 with SDR SDRAM memory, we would agree exactly to these cards ifwe have to choose. And if you buy a GeForce 256 based graphics card instead of RAGE FURY MAXX one, you may be absolutelysure that this card will serve you well at least for a year until it gets outdated.
Well, we can state that ATI was again late for the give-away show. Only if ATI develops a highly aggressive pricing policy,RAGE FURY MAXX will get a chance to occupy a certain market sector at least till the amount of games optimized for hardware T&Lreaches the critical point and the geometric coprocessor becomes a must. According to the most pessimistic forecasts it mayhappen by the end of the year 2000, and as to the most optimistic ones, - by the middle of the year 2000. This info actuallyseems rather credible because the mentioned forecasts were made by game developers, who should have enough evidence to provetheir point.
By the way, there is an opinion that a powerful CPU can make up for the absence of T&L hardware support by RAGE 128PRO chips, and hence by RAGE FURY MAXX graphics cards. We don't agree with this statement. We won't give any detailedexplanation to avoid repeating the already mentioned things. In brief the thing is that a geometric coprocessor not onlyreduces the system CPU utilization but also that of the system memory bus. If the game is optimized for T&L hardware support,we have every reason to say that the CPU will hardly stand idle, because the game developers can always think of something tooccupy the system CPU. That is why if you run a game intended for T&L hardware support in a system with RAGE FURY MAXX graphicscard hoping that a powerful CPU will cope with all the coordinates transformation and lighting calculations, then you will besurprised to notice the whole bunch of unpleasant effects. We are sure that AI will work much slower and the real worldphysics won't be supported at all. By the way, you will anyway need a powerful CPU with a RAGE FURY MAXX graphics card,because the performance of RAGE 128 PRO chip greatly depends on the system CPU computing capacityand the tests we carried out for RAGE FURY PRO prove it.
And one more idea on the matter. ATI is fond of stressing that their chips support hardware motion compensation andinverse Fourier transformation (iDCT). This provides high image quality during DVD-video playback and the CPU utilizationin this case hardly gets over 50%. However, we don't think that you urgently need a RAGE FURY MAXX graphics card if you wantto watch DVD-movies, an ordinary RAGE FURY PRO will be more than enough for that. That's why this feature shouldn't beregarded as the major argument for buying this card. Since we came to speak about it. In the regime of DVD-video softwaredecoding, i.e. carried out solely by the system CPU (if it is a Celeron 366MHz) Cinemaster'99 player provides ideal imagequality during DVD-video playback, not a bit worse than in case of MPEG2 video playback with hardware (full or partial)support. The only difference makes the CPU utilization, but on the other hand we can't think of a situation when you watcha DVD-movie while playing Quake2 through the network.
No doubt, MAXX technology is very interesting as an alternative solution for performance increase. The coolest thingabout it is its hardware independence, i.e. the graphics cards can be manufactured with any two graphics processors onboard, each of which forms a complete frame from the very beginning up to the end. The ready frames are displayed on themonitor in turn. Each graphics processor can display up to three frames running not more (in case of triple buffering),and while one chip is displaying its frames the other one is working on the next ones. Everything is operated via software.Frankly speaking we don't see any serious obstacles preventing the developers from using more than two graphics processorson one and the same board within MAXX technology. However, it is not quite expedient for several reasons. First of all themaximum performance gain is possible only in case of two-chip solution. The third chip will increase the system cost andcomplexity, while the performance gain will be relatively low. Second, there may appear some serious problems with powersupply and overheating.
Does MAXX have any future at all? It probably does. It would be very interesting to see two graphics chips with integratedgeometric coprocessors working together. Besides, ATI will perhaps continue working on MAXX technology, and one of thepossibilities here may be the ideas of T-Buffer technology. Practically MAXX technology allows forming scenes in differentframe buffers, so now they need to work out a mechanism to combine the frame buffers content and then to display the resulton the monitor. Isn't it a good new solution for special effects realization? Well, these are just our personal suggestions,but maybe...
All we can do now is just patiently wait for the new RAGE FURY MAXX graphics cards to appear in the market in orderto finally take a closer look at them. Besides, we should also pay special attention to the drivers quality. And only afterwe are through we can correctly estimate the marketing prospects for RAGE FURY MAXX and only then we will be able tounderstand if this product appeared at the right time or if it came too late.
If you belong to the team of ATI dedicated admirers and have a really big monitor, then you can certainly buy a RAGEFURY MAXX. It is most likely that you will remain satisfied with your purchase, its performance and image quality in32-bit color. However, you should bear in mind that this card may get outdated in 6 months, when new games require ageometric coprocessor. If you have already made your mind to buy a new graphics accelerator and don't know what tochoose we would be glad to give you another piece of advice.
First, you'd better wait for the graphics cards based on S3 Savage2000 and look at their practical value.
Second, you may wait for the graphics cards on GeForce 256 with DDR SDRAM graphics memory, which are also expected inDecember. Take a look at their prices. If a card like that costs about the same as RAGE FURY MAXX, then we would recommendbuying the one on GeForce 256.
And finally, you may give up everything and wait for the new solutions from 3dfx, ATI, Matrox or NVIDIA.
We have an idea of what can we expect from 3dfx, but anyway, on 15 November we should already know all the details.As for NVIDIA, we are likely to get acquainted with its new more powerful GeForce 256 manufactured with 0.18 microntechnology in spring of the year 2000. And what about ATI? Are they preparing us something new? Yes, they are, andthey have been doing it for a long time already. This is a new chip known under the codename Rage5. We suppose that it willbe also equipped with an integrated geometric coprocessor. But this time its future will solely depend on the time it islaunched. If ATI doesn't feel like losing its positions in the market it will have to work as hard as possible to launch anew generation graphics accelerator not later than in spring of the year 2000. Besides, they should also make sure that thedrivers are correctly done and that the parameters of the graphics card are at least up to the competitors' mark. Anyway,time will show if ATI has learnt its lesson...