NVIDIA GeForce 7800 GTX – new races, or a 24-key chord

Have you ever flown large passenger airplanes? If yes, you must remember the moment the pilots are just engaging the turbines, they start picking up and gaining speed. That sound which grows from din to whizz is full of power and speed, and with every second it seems to you that it is the utmost limit but the sound pitch is growing higher and more powerful. Amusingly, the acceleration normally stops beyond the anticipated human limits, and this stop is definitely made at the pilots' will, not because of attaining limitations of the turbine RPMs. This analogy works very well for video cards. With the release of every new model, those who are professionally involved in modern graphic accelerators are unanimous in saying "that can't be faster, it's impossible to make it faster as it is because so many obstacles are in the way..." But a year afterwards there is released a new video card that beats all the previous records, and all the talk about attaining a performance limit are repeated all over again. In this material, we decided not to make previous mistakes and better concentrate at the current coil of performance as a middle of the way aimed at infinity, and not like about the end of the world – so, today we are talking about the most recent graphic processor G70 and NVIDIA GeForce 7800GTX video card built on its base, which opens a new page in the history of GeForce series 7 video cards. Let's look at the table of specifications and talk about the latest NVIDIA's product in more detail.

As we can see, the shader capabilities in the new chipset have remained unchanged and are at the level of version 3.0. This version is more than enough for the nearest future because games that offer support for these shaders are just appearing, their number will grow continuously, which is beyond any doubt. But at that the technical similarities to NVIDIA GeForce 6800 Ultra (NV45) are almost over. In NVIDIA GeForce 7800GTX, there have been implemented 8 vertex and 24 pixel shaders, which means 1 vertex shader is available to each 3 pixel pipeline. The chip contains 300 million transistors, which is 80 million greater than in its predecessor, and at the same time it produces much less heat, consumes less power because is made following the 0.11 mk process technology. Original clock speeds which are now 430 MHz for the chip and 1200 MHz for the memory subsystem have also gone up. Finally, NVIDIA GeForce 7800GTX has inherited the possibility to assemble a SLI system, because with two NVIDIA GeForce 7800GTX video cards you get an unprecedented gaming performance. But all what was said above is almost an advert, so let's better analyze each item in more detail to find out what NVIDIA really managed to attain.

As many anticipated, the progress has indeed followed the line of least resistance, and the rise of performance in the new solution by NVIDIA has been achieved due to the increased number of pipelines. NVIDIA found it impossible to raise performance due to the increase in the core and memory frequencies the way ATI did in its X850XT, because the yield of chips produced following the 0.13 mk process technology with 220 mln transistors running at frequencies higher than 500 MHz is very unlikely, and even if it were possible, it would hardly be of interest. It is hard to imagine how much power this graphic processor would consume and emit heat, because even at 400 MHz the GeForce 6800 Ultra put more than enough load to a 400W PSU. That is why NVIDIA decided first to migrate its new processor to the 0.11 mk process technology and due to that reduce the power consumption. The task is not only to improve the old but to increase performance to a new level. That is why they decided to add 8 pixel pipelines to the existing 16, with 2 vertex pipelines to the existing 6. Logically, this should increase performance of the whole video system at least by 1.5 times or even more, at best. There are things like optimization and debugging of the chip which were revealed and mastered still in NV40. Moreover, NVIDIA's specifications for GeForce 7800GTX delightfully and proudly state about the improved bandwidth in each block inside the chip, including pixel and vertex pipelines, as well as creation of a new texture engine and many other modifications made, of which you will soon read on NVIDIA's web site yourselves. Our task is to verify all these improvements in action.

Of course, the functional part of the processor has undergone innovations. There has been added the functionality of new full-screen antialiasing – Transparent Antialiasing, and other effects included in the new specification of CineFX 4.0 products. There has also appeared an analog of 3Dc normal compression by ATI dubbed V8U8 that allows a twofold compression of normal maps.

The new rendering technique by NVIDIA called High Dynamic-Range (HDR) allows using 16-bit floating-point components to implement more complex illumination. Also, it's enough using accelerators for games only. NVIDIA suggests using NVIDIA GeForce 7800GTX  in applications which have nothing to do games, because floating-point computational capabilities of this accelerators are much more serious than in the regular Pentium 4. What matters is that the program requiring serious computations should be able performing parallelization. The areas of application include computer vision, fluid modeling, image and volume processing, queries to databases, gravitational field computations, Monte Carlo techniques, computational geometry, and even global techniques of illumination like Ray Tracing and Photon Mapping. In applications like these, the performance gain achieved through using NVIDIA GeForce 7800GTX may be more than 10-fold.

Amusingly, without games the NVIDIA GeForce 7800GTX is able being a serious aide to the processor, but if you are up to playing at it, you will definitely not get by without a powerful processor. Reminding it that the problem of CPU-boundedness has been looming for quite a long time, but recently it is getting increasingly topical and serious. The idea behind it is that most games require not only the power from the graphic processor, but also that of the CPU, and if the latter runs much slower than the former, the graphic accelerator starts getting idle and no longer gives all of its 100% power. Remember that while testing NVIDIA GeForce 6800GT SLI with AMD Athlon 64 4000+ we talked much about the CPU-boundedness. That time, almost all the tests showed impossibility to overcome a certain FPS bar regardless of resolution and AA settings because of the CPU performance shortage. As the GPU power goes up, this problem should manifest itself even more. While in the single mode Pentium 4 starting from 3GHz manages to cope with NVIDIA GeForce 7800GTX, in the SLI mode the 7800GTX will definitely run idle because to date there is no processor on the market that is able feeding enough data to this GPU. That equally applies to ATI's CrossFire and the new R520 chip. But this problem either should be solved in future, because with the release of dual-core processors the manufacturers will start optimizing their games for parallelized computations using two CPUs, which should make it possible to feed full amount of data to the pipelines of accelerators thus making them run at full load.

But let's turn back to the video card in question - NVIDIA GeForce 7800GTX.

As can be seen from the photo, not only has the board acquired a powerful engine made up of 24 pixel pipelines but a fairly stylish look. NVIDIA GeForce 7800GTX, like its predecessor GeForce 6800 Ultra, is equipped with two DVI-I outputs and a connector for plugging in a video input and video output, and has green color of textolyte. Due to the narrow cooling system, the board takes up one slot inside the housing. The cooling system has been evidently reduced in size due to the dropped heat emission, but we'll talk about that in a separate section.

NVIDIA GeForce 7800GTX offers a native support for PCI-Express 16x bus, so there aren't many chances for release of this card aimed at the AGP bus. The 75W applied to the PCI-Express bus is definitely not enough for the new accelerator, so the board has a 6-pin connector for additional power supply. There is also indeed a standard connector for merging a NVIDIA GeForce 7800GTX into the SLI system - the specs are honest :). All the memory modules are covered by a massive radiator that unites the graphic processor as well. The power conditioners also have their own radiator, and it has turned much massive versus GeForce 6800 Ultra.

From the rear side of the board, the memory modules are also cooled, although its efficiency is disputable because the metal radiator is very thin and serves more for fixing the radiator at the outer side of the board rather than for heat rejection. Having turned off eight screws that fix radiators from the rear side and two small screws fastening the plastic with the fan on the external side, we were able to view the main radiator.

As can be seen, the graphic accelerator and memory are indeed cooled by its different parts. So you don't have to worry about the overheating of memory modules because of the thermal behavior of the graphic processor. But let's go on talking about the cooling system in the respective section and now turn to describing the PCB.

By the PCB design, it slightly reminds the GeForce 6800 Ultra, although the design has been essentially reworked. Similarities are seen only in the approximate match of positioning the main onboard components like the graphic processor and memory modules, and in the positioning of the video capture chip. The power conditioner module has been completely reworked and is now based on 9 conditioners.

Eight memory modules are positioned over both sides of the PCB along the perimeter of the GPU, 4 on each side. However, it is well seen that there welded points for additional 8 modules. Since there is 256 MB of memory on our specimen in question, that means NVIDIA GeForce 7800GTX can be equipped with 512 MB of memory as well. In all the other respects, the board is of a bit larger size than that of GeForce 6800 Ultra and has a bit fewer small components.

The graphic processor G70 installed on NVIDIA GeForce 7800GTX was made on the 16th week of 2005 which means late April or early May and is not yet a batch-produced specimen, but a test sample. Although it is merely a test sample, the chip is already of stepping A2. It looks like engineers succeeded to make it all after the second try. The graphic processor runs at 430 MHz. We don't mean to say that the frequency is too high, but is quite enough for the first revision of the chip. In view of the heat emission falling within the norm and the 0.11 mk process technology, we can safely state that G70 will be the base for not only GTX-versions but also for much more powerful ULTRA and other versions since the chip can be definitely scaled in frequency due to improved cooling.

The board uses absolutely classical memory modules. Perhaps in the near future it will become possible to give up marking the nominal memory capacity values and mentioning their manufacturer since all the boards on the base of GDDR-3, regardless of the GPU manufacturer, are equipped with SAMSUNG modules marked as K4J55323QF-GC16. I am not going to list boards using the same memory because they are too many. What I mean to say that these modules, albeit overclockable, do not demonstrate outstanding results. The memory subsystem runs at 600 MHz or 1200 MHz, considering the doubling of frequency due to the DDR. These modules offer 1.6 ns access time and judging by the manufacturer's specifications, run at their maximum speed without margin of safety. So it is quite difficult to talk about the overclocking capabilities of these modules - it is poor rather than good.

TV features

“To us, movies are the most important of all the arts”, so in NVIDIA GeForce 7800GTX a lot of innovations have been introduced in terms of TV features. To start with, the NVIDIA® PureVideo™ Technology as a combination of high-quality hardware and software is able making the image clear, smooth, with precise color rendition and precise image scaling for the whole video content. The GPU is able adapting to new video-encoding techniques for hardware-driven decoding and imaging smoothing. There is also the HW decoder for MPEG2 and WMV including the WMV-HD standard, a HW system for deinterlacing and scaling, color correction, and a HDTV-coder integrated into the GPU. Clearly, all that is aimed at making the on-screen image on modern TV maximum correct and adequate, but in reality the HDTV is not hasty at coming to Russia, and all the devices which support it normally have a regular D-Sub input nearby through which the image is displayed no worse than via the composite video output.

The video input on NVIDIA GeForce 7800GTX has been implemented with the aid of the Philips chip marked as SAA7115HL. Unfortunately, there are too few comments on this board, so it is difficult to talk about its qualities, but what is certain is that the TV-in will work.

Cooling system

The whole cooling system has already been described above in its most parts, but some issues require a separate description. For instance, the cooler controlled with four contacts. In some aspects, NVIDIA runs into extremes. Until recent times, all its cards have been equipped with coolers having two wires with the RPM rate controlled performed with variable-voltage by components on the main PCB. A system like that was unable to detect failure of degradation of the fan. Now it is all on the contrary. Normally, the cooler is controlled with three wires, but now on NVIDIA GeForce 7800GTX there are four wires running to the fan.

There is also one more questionable thing. In our package bundle the board was equipped with an aluminum radiator, but on the photos in press releases a similar model of NVIDIA GeForce 7800GTX was equipped with a copper radiator. It is hard to say with which radiator the NVIDIA GeForce 7800GTX was sent to the batch-production or whether it will sell because we've got only a test sample of the board. However, we can state with confidence that even an aluminum radiator is enough for normal functioning of the board without bugs and artifacts because of overheating. Moreover, even at times of intensive load the board was running quietly and was not making a roaring noise which is typical of most top-end video accelerators, and in the stand-by mode was absolutely noiseless.

Finishing with the examination of the board, let's look at the radiator from the reverse side. In general, the radiator produces a very serious impression - that is not a rough piece sawn of aluminum ingot but a sophisticated engineering appliance. The main heat-spreading surface from the GPU is linked to the main radiator with a heat pipe. In so doing, the memory modules are as if cooled by a separate radiator. Of course, the unified structure of the radiator brings the heat of the core over to the memory modules, but there is a firm hope that memory modules will not be affected by that - moreover, they have their own system of active cooling.

Power supply

The power consumption of top-end products is a pain for all the worshippers of NVIDIA's graphic accelerators. To be honest, according to the declared specifications the NVIDIA GeForce 7800GTX  consumes as much as NVIDIA GeForce 6800 Ultra, that is, 110W at most. Having improved the process technology to 0.11 mk, NVIDIA added another 80 mln transistors inside the core, thus returning the power consumption to the previous value. An amusing fact: the specifications now state that the minimal power of the PSU is 350W for normal functioning of the board, not 450W as it was shown for GeForce 6800 Ultra. Through experience, we can say that NVIDIA GeForce 7800GTX indeed consumes less power, but we should be aware that buying NVIDIA's best accelerator we have to take care about the right PSU.

Overclocking

Contrary to all expectations, overclocking the NVIDIA GeForce 7800GTX  has shown quite positive results which easily move the card to the next performance class. In our test lab, we were able raising the frequencies to the following values.

Therefore, the frequencies increased by 60 MHz for the chip and by 100 MHz for the memory. The overclocking capability like that is a sign of high-quality process technology and high chip yield. We shouldn't forget that our board uses a pretty middling cooling system. Through increasing its size it is possible to attain operation stability at higher frequencies.

Tests

You can get the idea of all the testing methodology in our article devoted to this issue. First off, we should make a reservation – in this part of the review, we'll be referring to the article on methodology, since we believe it does not make much sense in describing the purposes of all the tests, methods of testing as well as the bottlenecks typical of all modern graphic accelerators regardless of the manufacturer and model all in one article. The article on methodology is continuously being updated and complemented. Nevertheless, I'd like to outline the major points:

Our test configuration is made up of the following components:
• Intel Pentium 4 3.4 GHz  Extreme Edition Hyper-Threading;
• Motherboard Intel D925XCV (1 GB DDR2 533 MB DualChannel, PCI-Express 16x);
• HDD Seagate Barracuda ATA IV 7200 rpm;
• PSU - PowerMan 520 W.

Our test lab appreciates Intel, Gigabyte, and Patriarch for the hardware presented for tests.

We ran the tests using the following benchmarking applications:

1. 3D Mark 2001 SE Ver330 (DirectX 8.1)
2. Codecreatures Benchmark Pro (DirectX 8.1)
3. Quake 3
4. SeriousSam 2 (OpenGL)
5. Unreal Tournament 2004
6. Unreal 2 (DirectX 9.0b)
7. Tomb Raider Angel Of Darkness (DirectX 9.0b)
8. FarCry(DirectX 9.0b)
9. Doom III (OpenGL)
10. Half-Life 2 (DirectX 9.0c)
11. Aquamark. (DirectX 9.0b)
12. 3D Mark 2005 (DirectX 9.0c)
13. SpecViewPerf 8 (DirectX 9.0b)

For our tests, we used reference versions of the drivers for both types of video cards. We used all the default settings but for the V-sync which we disabled. The AA and AF modes were activated via the control panels of the drivers. For NVIDIA cards we used the most recent ForceWare 77.62 drivers, and for ATI cards - Catalyst 5.4 WHQL

In all the graphs, the data for NVIDIA GeForce 7800GTX card being tested is depicted in red. In the graphs for AA and AF modes, the dark vinous color stands for the data for NVIDIA GeForce 7800GTX  overclocked to 490/1310 MHz. The yellow color stands for data for the current competitor of 7800GTX – X850 XT Platinum Edition. The green color depicts data for the predecessor to today's video card - Leadtek 6800Ultra. For comparison, we also brought in the results for ATI Sapphire X800XT which are depicted in violet color, for NVIDIA GeForce 6800GT – dark green, with X800 by Gigabyte having a passive cooling system closing all the graphs.

3D Mark 2001 SE Ver330 (DirectX 8.1)

As usual, we start with outdated synthetic benchmarks. Although they are no longer topical today, the results allow looking at the performance of solutions being tested with old games which are still topical beyond any doubts.

In this test, shaders of the first version and a great number of polygons are used, so the raw power of the accelerators does determine something. Of course, NVIDIA GeForce 7800GTX  is ahead of the whole planet, but the real performance gain versus ATI X850XT PE is not significant. This can be explained by the adherence of the test to ATI's boards. Look at the results for other cards by NVIDIA - they lag well behind their ATI's analogs. We enable the FSAA.

At the lower resolutions, NVIDIA GeForce 7800GTX  showed to all the participants who is the king on the market, and on getting the higher resolutions the difference almost smoothed out. Note how positively the overclocking of the card being tested affected the performance.

Codecreatures Benchmark Pro (DirectX 8.1)

We are bring in this test in contrast to 3Dmark 2001, since under absolutely similar requirements it favors the products by NVIDIA.

The gap between ATI X850XT is much wider than in the previous time and is 20 FPS on average. By their specifications, this test and 3Dmark Nature are similar like twins, but they produce different results. Let's look at the results produced with FSAA:

In its overclocked state, NVIDIA GeForce 7800GTX takes a noticeable lead at all the resolutions, and even without overclocking it is an absolute leader. Let's make a minor conclusion and move on. Although NVIDIA GeForce 7800GTX is a leader at applications using DirectX 8, it definitely doesn't reveal all of its speed capabilities.

Quake 3 (OpenGL 1.1)

Then there go old OpenGL games, tests with which have long been tests for CPU-boundedness. The CPU power is so important to them for greater FPS. This test has been brought in for edification of those who are up to buying a powerful video card but a 2GHz processor.

As can be seen, NVIDIA GeForce 7800GTX hasn't lost its CPU-boundedness up to the highest resolutions. The same applies to both Leadtek 6800Ultra and NVIDIA GeForce 6800GT. Let's enable the FSAA.

With FSAA enabled, most video cards demonstrate less CPU-boundedness, but NVIDIA GeForce 7800GTX  is uncompromising and demonstrates practically identical results at all the resolutions taking an almost 1.5 times lead over ATI X850XT PE. But NVIDIA GeForce 6800 Ultra has shown a much smaller lag behind the new leader. Somehow or other, NVIDIA offers a powerful OpenGL driver. Let's make the task a bit more complicated through adding polygons to the frame.

SeriousSam 2 (OpenGL)

On the one hand, Serious Sam is no longer Q3 or RTCW, because the number of polygons in the frame is greater and the effects are more formidable. On the other hand, this is the same OpenGL 1.1 that is on par with DirectX 8, thus Q3 and Serious Sam have something in common.

We see a dreadful CPU-boundedness at all the resolutions. That is normal, and we did not expect anything different. By the way, NVIDIA GeForce 7800GTX  is anyway the leader.

Use of antialiasing hasn't changed the picture. To make the video card a determining factor, the CPU has to be replaced with a much more powerful one. We are now moving on to examining DirectX gaming tests.

Unreal Tournament 2004 (DirectX 9.0)

UT2004 is the first in this series. The game is fully optimized for attaining a maximum FPS, so during its development most focus was given to the compensation of CPU-boundedness. The game uses only one shader, so the result in many ways depends on the raw power of accelerators.

Wow, what an impressive result! Lower resolutions certainly demonstrated CPU-boundedness, but higher show a twofold superiority of NVIDIA GeForce 7800GTX over ATI X850XT, with the others running slower than the latter. Even SLI did not give such a performance boost in this game. Switch on the AA.

Enabling the antialiasing has slightly smoothed out the gap between NVIDIA GeForce 7800GTX and all the others, but anyway left it at a really essential level. The first three resolutions require a more powerful CPU, but the higher resolution even showed a difference between overclocking and its absence.

Unreal 2 (DirectX 9.0b)

Unreal 2 makes quite a big difference from UT2004, although their engines are similar. This is related to that UT2004 is aimed at the multiplayer mode, whereas Unreal 2 is a classical single-player that features in use of extended scripts for building a plot etc.

Again a complete CPU-boundedness for almost all the cards. No comments.

A nice graph, isn't it? NVIDIA GeForce 7800GTX  shows a 90% CPU-boundedness and takes a stable lead over the other, albeit minor.

Tomb Raider Angel Of Darkness (DirectX 9.0b)

The course of slightly outdated tests is closed by a game that has no special liking of NVIDIA's products. TROAD uses many complex shaders of the second version. The test scene was selected in a way to make the accelerator's job maximally difficult.

In fact, TRAOD did not like the fifth series of GeForce, but with series 6 and 7 the "fondness" seems to be for good. Again, NVIDIA GeForce 7800GTX  showed the same results for all the resolutions - perhaps enabling the AA will change something?

No. It won't. The CPU-boundedness is also seen like in all the other tests without AA enabled, although the processor is as powerful as Pentium 4 Extreme Edition 3.4. Make you own conclusions.

FarCry (OpenGL)

The game is built on its own engine, offers simply superb graphics coupled with advanced AI of the game characters.

Although the result with which ATI X850XT PE and Sapphire X800XT have taken a lead over the new NVIDIA offspring looks strange, but that is a fact. The cause of that is in that ATI drivers are less CPU-bounded in some tests, and once conditions of rigid CPU-boundedness appear they rush forward by 2 FPS. Nevertheless, such a victory can hardly be regarded as fair, because if the load is increased, the following happens:

The overclocked version is definitely a sure leader, but in the nominal state it is a real challenge for NVIDIA GeForce 7800GTX  to compete with ATI X850XT. But look at how performance in Leadtek GeForce 6800 Ultra and NVIDIA GeForce 6800GT has dropped. Compared to these, NVIDIA GeForce 7800GTX rules. Probably, installation of FarCry patch 1.3 that supports version 3.0 shaders will solve performance issues for NVIDIA cards in this game.

Doom III (OpenGL)

While FarCry is nobody's game, DooM III is a definite follower of NVIDIA accelerators. It is no wonder - these days it is increasingly trendy among hardware monsters to trample down hit games before their release and optimize for their products.

We see a sure victory of NVIDIA cards in general and NVIDIA GeForce 7800GTX, in particular. Compared to Leadtek GeForce 6800 Ultra, the performance in the upper resolutions has gone up by 12 FPS, which is good in itself. ATI X850XT loses even to NVIDIA GeForce 6800GT. Again, note the CPU-boundedness of the results.

Enabling the AA hasn't changed the relations of speeds for all the participants, but the CPU-boundedness started appearing less. The results for NVIDIA GeForce 7800GTX in the overclocked state look impressive.

Half-Life 2 (DirectX 9.0c)

ATI boards can be rehabilitate only by a game, competitor to DooM III. That is Half-Life 2. In fact, HL 2 is a competitor to DooM III not only in terms of the gaming process. That is a field of competition for DirectX versus OpenGL, and thus ATI versus NVIDIA. It would be logical for ATI to give shelter to a game that is natively benevolent to their products from the very start. We can state it with confidence that the picture of speeds as compared to DooM III would be different.

At the lower resolutions where all cards proved CPU-bounded, ATI accelerators rule. But as the resolutions go up, NVIDIA GeForce 7800GTX starts proving its leadership to all the others. The remaining products by NVIDIA a definite losers to ATI boards. Enabling the FSAA, and ...

We see an oil painting. The power of NVIDIA GeForce 7800GTX was enough for a substantial and serious win at resolutions from middle and higher. It is always a pleasure to beat the rival at his field.

Aquamark. (DirectX 9.0b)

Again we can observe a one-one victory of NVIDIA GeForce 7800GTX  in all the resolutions, although not considerable but stable. The test shows a strong CPU-boundedness, which is seen through the similarity of results for almost all the participating boards.

With the AA enabled, the approximate relation of speeds has preserved, but the CPU-boundedness remained where it was. Nothing to add.

3D Mark 2005 (DirectX 9.0c)

This test is one of the most recent. It offers support for all the latest technologies like version 3.0 pixel and vertex shaders etc.

As we see, NVIDIA GeForce 7800GTX  wins at all the resolutions, and as the resolution goes up the gap between the remaining cards is growing wider.  While Leadtek GeForce 6800 Ultra and ATI X850XT run somewhere at about the same level, NVIDIA GeForce 7800GTX takes a big leap forward.

Enabling the AA finally beats all the other rivals of NVIDIA GeForce 7800GTX completely and lets it take the distinguished title "the king of the hill". Surprisingly, at 1600x1200 the card in question takes almost a twofold lead over its predecessor.

Since the upper two graphs displayed the index component of the test, we are bringing in the results produced with the game parts of 3Dmark 05 for your judgment. However, because the picture in them will repeat the above graphs, we are not commenting on them but present them in the "figures only" mode.

SpecViewPerf 8 (DirectX 9.0b)

Finally, we are bringing in the results produced for professional applications.

It is well seen that at this area there are still no serious changes - the gaming accelerators should remain belonging to games, whereas professional to professional. All the graphs have been partitioned into two groups. Some are fond of NVIDIA drivers, others feel a liking to those by ATI. Little depends on the real performance of the boards.

Final Words

Summing it all up, we say that the new GeForce 7 series offers quite a promising start and a bright future. Reworking the chip has brought fairly good results and allowed NVIDIA to finally grab the leadership crown in its hands. All the innovations in the new product have practically no

shortcomings and offer only the positive to the end user. The only problem is still the high CPU-boundedness that determines the performance of the video subsystem, but that should be solved in the nearest future with the release of dual-core processors. Today's top-end rival product ATI X850XT is no longer a competitor to NVIDIA GeForce 7800GTX and currently this accelerator can be regarded as the most powerful solution on the market, especially in that these cards are able being merged into the SLI-mode thus doubling the performance. The only competitor to this board may be a new product by ATI built on the base of the new R520 chip which so far has not been announced. Therefore, we can congratulate NVIDIA on the release of the most recent product. Welcome to the era of new-generation accelerators.