by Alexey Stepin , Yaroslav Lyssenko
12/03/2007 | 03:09 PM
Using top-performance GPUs on rather inexpensive graphics cards is a widespread practice in the consumer 3D graphics industry when there are no alternatives available. A well-known example was the announcement of the world’s first DirectX 9 compatible GPU, the R300 chip developed by ATI Technologies, which used to be installed on the flagship Radeon 9700 Pro as well as on the mainstream Radeon 9500. Later on, it was replaced with the R360 and R360 cores in the top-performance segment and with the RV350/360 in the mainstream sector.
There are factors that justify this approach. First, it helps utilize defective cores that haven’t passed a frequency test and/or have defective units that can be turned off. Second, it doesn’t require developing a new PCB which is a costly and long process. A good illustration of this development approach was shown by ATI with its Radeon X1800 and X1900 series that had in fact one and the same PCB design.
The downside of this approach is obvious, too. As the tech process is getting optimized, the chip yield is growing up and the share of defective cores is shrinking, and the manufacturer finds himself having to install full-featured top-class GPUs on inexpensive graphics cards, which is of course not optimal in terms of profit margin. The use of expensive and complex PCBs has a negative effect on the manufacturing cost as well as the end price of a product, so the developer has to diversify his GPU range sooner or later.
Being quite urgent even at the times of Radeon 9700, this problem is especially sharp today when GPUs have evolved into immensely complex devices. For example, Nvidia’s G80 chip incorporates as many as 681 million transistors, but the company has to install it on a $299 card because the G84 core, the basis of the GeForce 8600 series, proved to be unable to deliver performance sufficient for that price category, especially in new games that use Direct X10 capabilities.
The GeForce 8800 GTS 320MB was competitive but it used full-featured G80 cores and an expensive 12-layer PCB that had been originally developed for a $400 graphics card. Of course, this compromise product had to be replaced and Nvidia was working on such a replacement. October 29 the company officially introduced the new G92 graphics processor and the GeForce 8800 GT graphics card series. We’ll tell you how much of a success the new product is and if it brings a new level of performance into the $199-259 sector.
Nvidia developed a new graphics processor, codenamed G92, for use in the GeForce 8800 GT graphics card series. Let’s see what technical characteristics it is endowed with.
Click to enlarge
Click to enlarge
First of all, we should note the thinner tech process in comparison with the G80 chip (GeForce 8800 Ultra/GTX/GTS). Theoretically, this should keep the die area, power consumption and heat dissipation within reasonable limits, which is important considering the increased transistor count. The new chip surpasses both Nvidia G80 and ATI R600 from this point of view, being currently the most complex GPU in the world.
It may be partially explained by the integration of the NVIO unit, previously a standalone chip, into the G92 die as well as by the more advanced PureVideo HD video decoder and post-processor. Some general improvements on the micro-architecture level and the PCI Express 2,0 interface may have contributed to the increased number of transistors as well, yet we don’t think that all those innovations are indeed worthy of 72 million transistors. Perhaps the G92 contains more shader and texture processors than declared but some of them are just disabled for now. If so, Nvidia will be able to easily introduce new G92-based solutions with higher performance or guarantee a 100% yield of chips with the existing configuration.
The new core represents a modified G80 architecturally, the revision having concerned texture processors mostly. The G80 has two filter units per each texture address unit, which ensured a high performance with anisotropic filtering, but the G92 is configured like the G84 in this respect with two filter units for each two address units.
Theoretically, the new GPU has 56 rather than 28 texture units, but practice suggests that this approach is not justifiable due to the wide use of anisotropic filtering today. We can recall those quite numerous cases when the GeForce 8600 GTS is slower than the Radeon X1950 Pro: although it seems to have more TMUs, its 16 texture processors turn out to be only 8 actually. This is not so critical with the G92, though. Even if the performance of the texture processors of the GeForce 8800 GT will correspond to the G80’s 28 texture units, the new GPU will still be superior to the GeForce 8800 GTS in this parameter, also because of the higher frequency of the main GPU domain.
Besides the differences in the design of the texture processors, the G92 is almost the same as the G80 on the architecture level. The number of functional subunits is different, though. The new chip has only 4 raster processors instead of 6. Each of them can process 4 pixels with 16 sub-pixel samples per clock cycle, which gives you a total of 16 pixels per clock cycle for color and Z values. If only the Z-buffer is processed, the processing speed is increased to 128 pixels per clock cycle in ordinary mode and to 32 pixels per clock cycle with 4x FSAA. The algorithm of multisampling of transparent textures has been improved. According to Nvidia, it now provides about the same quality as super-sampling but without a great loss of performance. This may mean some changes in the driver, however, because the company used to tout transparency antialiasing as a unique feature of the GeForce 7 series but later turned it on in the GeForce 6, too.
The reduction of the number of raster operators shouldn’t be regarded as a bottleneck in the G92 architecture because the performance of modern GPUs with a unified architecture depends largely on the number and design of shader processors and on the efficiency of the driver’s shader code translator. The GPU’s 16 ROPs can only have a serious effect on performance if you use a very high resolution together with full-screen antialiasing, but the memory bandwidth is going to be a narrower bottleneck then. Moreover, the GeForce 8800 GT is not actually meant for such extreme operation modes which are a prerogative of the most expensive single cards and multi-GPU configurations.
The G92’s memory controller has 4 sections instead of 6 as in the previous chip and the total memory interface width is 256 bits now (64x4). That’s a recession in comparison with the GeForce 8800 GTS, but it is partially made up for by the increased memory frequency. Some improvements may have been introduced into the memory controller architecture to boost its efficiency. Don’t forget about the capacity: we know that 320 megabytes of memory on board the GeForce 8800 GTS 320MB is sometimes not enough in modern games and the GeForce 8800 GT, at least in the senior configuration, has an advantage here. 512 megabytes of graphics memory should be a positive factor under real-life gaming conditions.
When it comes to shader processors, the GeForce 8800 GT is undoubtedly superior to the GeForce 8800 GTS. It not only has more ALUs but also has them clocked at a frequency which is higher than that of the GeForce 8800 GTX. Of course, this should have a positive effect on the card’s performance in modern games which make wide use of visual effects created by means of complex math1ematical computations. The configuration of the shader processors hasn’t changed: 112 superscalar ALUs are grouped into 7 units with 16 ALUs in each (as mentioned above, there is probably a disabled eighth unit). One shader code branch can vary from 16 to 32 pixels like with the G80 chip.
Nvidia’s new GPU is one of the first top-end GPUs with a full-featured HD video decoder. Another such chip is the ATI Radeon HD 3800. The G92 employs the video-processor introduced in the Nvidia G84 with all its highs and lows. Thus, the GeForce 8800 GT can do all HD decoding on the GPU, including decoding of data that use the entropic compression algorithms CABAC and CAVLC (at least for the H.264 format).
VC-1 is probably supported on the same level as before, but our earlier test proved that the average CPU load when playing this format was not much worse than with the most advanced video processor of today which is incorporated into ATI’s Radeon HD 2600 and 3800 series (for details see our article called The Look of Perfect: Analysis of Graphics Accelerators Performance during Media Playback). So, the GeForce 8800 GT claims to be the first graphics card to combine good 3D performance with superb multimedia capabilities.
Besides offloading the CPU when decoding high-definition video, the new PureVideo HD features a more advanced post-processor for such content. This means you can expect a high visual quality when watching Blu-ray and HD DVD movies on a PC equipped with a GeForce 8800 GT.
ATI’s Radeon HD 2000/3000 series cards still hold a monopoly on the audio-over-HDMI feature, though. The GeForce 8800 GT lacks this ability because the G92 doesn’t incorporate an audio core. This is not much of a drawback considering the limited capabilities of the audio solution offered by ATI’s Radeon HD 2000/3000 cards, yet the idea of using only one cable for connecting your multimedia system to an LCD or plasma panel is appealing anyway.
Among innovations that are not directly related to the graphics architecture the G92’s support of the PCI Express 2.0 but that delivers two times the bandwidth of the widespread PCI Express 1.x must be mentioned. When using an x16 connection, the bandwidth of PCI Express 2.0 is 16GB/s (8GB/s in each direction) as opposed to 8GB/s (4GB/s in each direction) with PCI Express 1.x. Moreover, version 2.0 has a higher load capacity of the slot (150W instead of 75W) but this doesn’t concern the GeForce 8800 GT which should be able to work on both new and older mainboards. On the other hand, the user may have compatibility problems that have been reported by some mainboard manufacturers, for example ASUS. This problem may occur to owners of mainboards that support PCI Express 1.0/1.0a whereas mainboards with PCI Express 1.1 should be perfectly compatible with the new graphics cards.
Nvidia’s new card will be available in two versions, with 256 and 512MB of memory, at a recommended price of $199 and $259, respectively. Thanks to the new GPU and simplified PCB these prices can be reduced later on to $199 for the 512MB version. This appealing pricing should make the new card a real bestseller especially if it does well in gaming tests.
In this review the Nvidia GeForce 8800 GT 512MB will be represented by Leadtek’s WinFast PX8800 GT Extreme. We’ll use it to check out the design features and gaming performance of the new graphics card from Nvidia.
This product is packed into a box of a standard size and shape. The box is rather small and lacks a handle – you can easily put it into a bag or something. The package is painted orange and blue, which is a very gaudy and eye-catching, yet perhaps not very delicate, design. The face side of the box shows a picture of yet another war robot.
The word Extreme in the graphics card’s name is marked with a green oval. It means that this version of PX8800 GT has frequencies higher than the reference card’s. Basic technical characteristics like the amount and type of the graphics memory are listed at the bottom of the box. You can also learn that the box contains a full copy of Neverwinter Nights 2, a popular role-play game.
We can’t say this package design is better or worse than what we could see reviewing earlier products from Leadtek. It is just different. It is good, but not extraordinary.
The contents of the box are neatly laid out into the cardboard compartments while the graphics card itself is wrapped into an antistatic pack. Besides the graphics card, the box contains the following:
We guess that’s good enough for the $199-259 price category. It is nice to see a copy of Neverwinter Nights 2 included with the WinFast PX8800 GT Extreme. Besides an exciting plot, this popular RPG features good visuals and can put a load even a top-end graphics card at its highest graphics quality settings.
The GeForce 8800 GT will be mostly manufactured for Nvidia at contracted facilities and delivered to the company’s partners in ready-made form. So, besides the standard versions of the card, you can only see pre-overclocked versions and/or versions with a unique cooling system, but most often it will be a standard version with a different sticker on the cooler. We don’t dismiss the possibility of such graphics cards coming out with an original PCB design, though. Major graphics card makers are quite capable of developing an eight-layer PCB with a 256-bit memory bus.
The new GPU from Nvidia has a 256-bit external memory interface and doesn’t require an additional NVIO chip, so its PCB design is simpler than that of the GeForce 8800 GTS, with 8 layers instead of 12.
Both PCBs have the same length of 22.8 centimeters but the components are placed quite differently on the GeForce 8800 GT just because the memory chips are installed in a semicircle around the GPU like on Nvidia’s GeForce 7 series cards. Being optimal for a 256-bit memory bus, this solution is employed by both Nvidia and ATI Technologies.
The power section resides at the back of the PCB, as usual. It is about as complex as on the GeForce 8800 GTS, but some elements are not installed, which indicates a low power draw of the GPU even when it is pre-overclocked by the manufacturer. This PCB may also be used for more advanced G92-based graphics cards where the eighth unit of texture and shader processors is unlocked. The power circuit is governed by a four-phase digital PWM controller Primarion PX3544, which is alike to the PX3540 chip employed on GeForce 8800 GTX/GTS. Four out of the controller’s three phases are used here: the elements of the fourth phase are not installed on the card. A simple PWM controller Intersil ISL6549CBZ is responsible for power supply of the memory chips. A piezoelectric speaker is located in the top right corner of the PCB – it informs the user with a warning signal about problems, for example when the additional power is not attached to the card. The card has a single power connector, a 6-pin rather than an 8-pin PCI Express 2.0 one, which is yet another indication of a relatively low level of power consumption of the GeForce 8800 GT.
The card carries eight GDDR3 chips (Qimonda HYB18H512321BF-10) in FBGA-136 packaging. These chips have a capacity of 512Mb (16Mbx32), a VDD and VDDQ voltage of 2.0V, and a rated frequency of 1000 (2000) MHz. The eight chips provide a total of 512 megabytes of graphics memory accessed across a 256-bit bus. Although the standard memory frequency of the GeForce 8800 GT is 900 (1800) MHz, it is pre-overclocked on this card to 1000 (2000) MHz, which provides the same memory bandwidth of 64GB/s as on the GeForce 8800 GTS.
As opposed to the G80, the G92 uses the traditional open type of packaging with a metallic protective frame but without a heat-spreading cap. There is a place left around the chip for a protective frame like on the GeForce 8800 GTX and GTX, though.
The die is relatively small. It is comparable to the G70 and much larger than the G71 although incorporates 2.5 and 2.7 times as many transistors as these previous-generation cores. Such a compact core is manufactured using the thinner 65nm tech process. The marking says that this is a third revision chip manufactured at the 40th week of 2007, which is early October. The main domain frequency is increased from the standard 600MHz to 680MHz and the shader domain is clocked at 1700MHz, which is 200MHz higher than the standard value and should ensure a good performance gain in games.
The left part of the PCB is interesting, too. You can see that the GeForce 8800 GT allows for the installation of a HDMI connector instead of the top DVI port. There is also an empty seat for a rather large chip here. We don’t know the purpose of it. Perhaps this is a place for the converter that supports the DisplayPort interface or Nvidia may be planning to endow the GeForce 8800 GT with the audio-over-HDMI feature to catch up with ATI’s products.
In the most widespread version the GeForce 8800 GT has two DVI-I connectors with support for dual-link mode and a universal 7-pin video output that supports Composite, S-Video and YPbPr output. The card can also be used in SLI configurations and has a MIO interface connector for that. Configurations with more than two cards, available with GeForce 8800 GTX and Ultra, are not supported because the card has only one such connector.
Nvidia developed a new reference cooler for the GeForce 8800 GT. It is a single-slot device with a complex internal design that should be capable of cooling the 65nm G92 chip.
We took the cooler apart to check out in interior, which was not an easy task. The casing seems plastic but is really made from metal. It is fastened to the base not only with the side locks but also with a sticker to the heatsink (probably to avoid rattling due to the vibration from the operating fan).
The photo shows three heat pipes that transfer the heat flow from the copper sole that contacts with the GPU die to the heatsink. The diagonal placement of the heatsink is logical since the cooler occupies only one slot and the hot air cannot be exhausted through slits in the mounting bracket. With the heatsink positioned like that, the hot air is thrown off the graphics card towards the interface connectors. This is no good for the temperature inside the system case, of course, yet shouldn’t be a problem for a well-ventilated case.
The cooler’s sole is made from copper while the rest of the base is aluminum. A generous amount of dark-gray thermal grease is used as a thermal interface between the GPU and the cooler. Less of it should have been quite enough for the relatively small G92 core. Perhaps a die package with a heat-spreading cap was planned originally.
There are special juts with pads soaked in white thermal grease at the spots where the cooler’s base touches the memory chips and power circuit elements. The whole arrangement is secured on the PCB with 12 spring-loaded screws. The cooler has a small radial fan that is very noisy at the max speed, which is not surprising considering its consumption current of 0.8A.
The fan uses a four-pin connection with outputs for a velocity sensor and a PWM-based speed regulation system. It works at a reduced speed most of the time – you can refer to the next section for exact numbers.
The new reference cooler from Nvidia is good overall, yet its efficiency is lower than that of the GeForce 8800 GTS due to its compact size. Despite the low heat dissipation of the new core, the system may prove unable to cool pre-overclocked GeForce 8800 GT, like the Leadtek WinFast PX8800 GT Extreme, and remain quiet. That’s why we can expect to see quite a lot of GeForce 8800 GT with a non-standard cooler.
Although the WinFast PX8800 GT Extreme comes pre-overclocked by the manufacturer, we still tried to speed it up further.
We managed to increase the GPU frequency to 720MHz with the default cooler – the shader domain was clocked at 1800MHz at that. That doesn’t seem much in terms of frequency growth, yet you should keep it mind that the default GPU frequencies of this GeForce 8800 GT model are already as high as 680/1500MHz. Nvidia’s new graphics core seems to have a good frequency potential: the growth from 600/1500MHz to 720/1800MHz with an air cooler is quite an achievement.
The memory chip refused to overclock just as we had expected and the graphics card was only stable at a memory frequency of 1050 (2100) MHz. That’s normal considering that these memory chips have a rated frequency of 1000 (2000) MHz and that we didn’t change the cooler or volt-mod the card. A better cooler may help reach higher frequencies, of course. Since we had a very modest frequency growth, especially with the memory chips, we didn’t test the Leadtek WinFast PX8800 GT Extreme at the overclocked frequencies.
We measured the level of noise produced by the Nvidia GeForce 8800 GT card with a digital sound-level meter Velleman DVM1326 using A-curve weighing. The level of ambient noise in our lab was 36dBA and the level of noise at a distance of 1 meter from the working testbed with a passively cooled graphics card inside was 43dBA. We got the following results:
The fan was very loud when we started up the system, but its speed then lowered considerably to a comfortable level. The card was practically inaudible among the noise from the HDDs and the fans of the PSU and CPU cooler. We should note that our testbed is configured like a top-end gaming system that is rather noisy by itself. In a quieter environment, for example in a multimedia system, the cooler of the GeForce 8800 GT is going to be more distinct. The option of manual control over the fan speed is currently blocked on the GeForce 8800 GT, which is quite right because an inexperienced user could reduce the speed too low, putting the card under the risk of overheat. The single-slot cooler works at its limit already.
Unfortunately, we couldn’t measure the power consumption of the Nvidia GeForce 8800 GT. Despite the declared compatibility with mainboards that do not support PCI Express 2.0, the new card does have some issues in this respect. Particularly, the Leadtek card proved incompatible with our Intel Desktop Board D925XCV (Intel 925X) and EPoX EP-9NPA+ SLI (Nvidia nForce4 SLI) that are based on chipsets that support version 1.0a of the PCI Express standard.
Although PCI Express 2.0 cards are compatible only with mainboards with PCI Express 1.1 and higher, our GeForce 8800 GT worked on an ASUS A8R32-MVP Deluxe (ATI CrossFire Xpress 3200) which only supports PCI Express 1.0a, according to the official PCI-SIG website. Moreover, the Leadtek WinFast PX8800 GT Extreme worked well on an experimental ATI RD480 (ATI CrossFire Xpress 1600) mainboard.
So, the GeForce 8800 GT has some incompatibility issues and we recommend you to check this out before you purchase that card. The problem may be corrected by modifying the graphics card’s BIOS but that’s a risky matter. We’ll be keeping an eye on that subject and informing you until the situation is clear.
By the way, this is not the first case of incompatibility of Nvidia’s graphics cards with some mainboards. We can recall the GeForce 7950 GX2 and some models of GeForce 8800 GTX as examples. Now there is the same problem with the GeForce 8800 GT and the users have to hope that Nvidia will see to this issue and take measures to correct it.
To test the performance of Nvidia GeForce 8800 GT we assembled the following standard test platform:
Since we believe that the use of texture filtering optimizations is not justified in this case, the ATI and Nvidia graphics card drivers were set up to provide the highest possible quality of tri-linear and anisotropic filtering. Also, to ensure maximum image quality, we enabled transparent texture filtering - Adaptive (multi-sampling) antialiasing for ATI Catalyst and Antialiasing – Transparency: multisampling for Nvidia ForceWare. As a result, our ATI and Nvidia driver settings looked as follows:
For our tests we used the following games and benchmarks:
First-Person 3D Shooters
Third-Person 3D Shooters
We selected the highest possible graphics quality level in each game using standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way. Games supporting DirectX 10 were tested only in this mode. Performance was measured with the games’ own tools or, if not available, manually with Fraps utility version 2.9.1. We also measured the minimum speed of the cards where possible.
With a few exceptions, the tests were performed in the following standard resolutions: 1280x1024/960, 1600x1200 and 1920x1200/1440. We used “eye candy” mode everywhere, where it was possible without disabling the HDR/Shader Model 3.0/Shader Model 4.0. Namely, we ran the tests with enabled anisotropic filtering as well as MSAA 4x antialiasing. We enabled them from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of Catalyst and ForceWare drivers
Leadtek WinFast PX8800 GT Extreme will be competing against the following graphics accelerators participating in our test session:
Since Leadtek solution runs at increased GPU and memory speeds, we tested it in two different modes: as is and with the frequencies dropped down to the reference values for Nvidia GeForce 8800 GT.
The game doesn’t support resolutions of 16:10 format, so instead of 1920x1200 we used a resolution of 1920x1440 pixels that has an aspect ratio of 4:3.
The Nvidia GeForce 8800 GT provides a nice addition to the minimum of speed. At a resolution of 1920x1440, when the amount of graphics memory has an effect on performance, the 8800 GT also enjoys a 26-27% lead over the GeForce 8800 GTS 320MB in terms of average speed.
That’s not surprising since the GeForce 8800 GT is slightly inferior to the GeForce 8800 GTS 320MB in one parameter only, namely in memory bandwidth, but surpasses it in the rest of the technical characteristics. The pre-overclocked version supplied by Leadtek is only really good at 1920x1440. Anyway, the average speed of over 60fps with slowdowns to 55-60fps will surely satisfy even a most demanding gamer.
BioShock doesn’t support FSAA when running in Windows Vista’s DirectX 10 environment. That’s why we benchmarked the cards without FSAA.
Both versions of the Nvidia GeForce 8800 GT – the ordinary and the pre-overclocked one – look superb in this test thanks to the 112 shader processors clocked at 1.5GHz and 1.7GHz, respectively. It is a question whether the GeForce 8800 GTS 320MB and the Radeon HD 2900 XT can be used for playing this game at 1920x1200, but the GeForce 8800 GT easily provides a high speed in that mode, especially in the Leadtek version that is always faster than 40fps.
At a recommended price of $259 the Nvidia GeForce 8800 GT leaves no chance to the ATI Radeon HD 2900 Pro and even beats the ATI Radeon HD 2900 XT!
ATI’s Radeon HD 2900 solutions used to be better in Call of Juarez, but the Nvidia GeForce 8800 GT has come to end this tradition. The version with standard frequencies outperforms the Radeon HD 2900 Pro while the pre-overclocked version from Leadtek is as fast as the Radeon HD 2900 XT 512MB while having a lower price. We guess the increase in the number of shader processors and their frequency relative to the GeForce 8800 GTS is the main factor contributing to this success.
Both versions of GeForce 8800 GT allow playing at a resolution of 1600x1200 with enough comfort. The owner of a WinFast PX8800 GT Extreme can even try to launch the game at 1920x1200 knowing that the speed will never drop below 25fps. You don’t need as high an average frame rate in Call of Juarez as in most other 3D shooters due to the specifics of its gameplay.
New in this review is our use of the final version of Call of Juarez with the DX10 Enhancement Pack.
Performance in DirectX 10 mode is as low as in the non-interactive demo we used earlier. You can only have a speed of near 20fps at 1280x1024, and the GeForce 8800 GT is 15% and 5% slower than the GeForce 8800 GTX in its ordinary and pre-overclocked versions, respectively.
The new card from Nvidia seems to be limited by its memory performance at higher resolutions as it falls far behind the flagship model. Perhaps the reason is in the architecture of the G92’s texture processors which are less efficient at performing complex filtering than the G80’s ones. Well, that’s rather a theoretical problem since every card is too slow in this test for normal gaming.
The GeForce 8800 GT is somewhat slower than the GeForce 8800 GTS 320MB in open scenes because of differences in the architecture of the TMUs or lower memory bandwidth or both. But on the Research level the new card goes ahead as it boasts as many as 112 shader processors clocked at a very high frequency. The pre-overclocked version from Leadtek is the leader in both cases although by a narrow margin on the Pier map.
It’s different in the HDR mode: the Nvidia GeForce 8800 GT is unrivalled everywhere, enjoying a 25-27% lead over the GeForce 8800 GTS 320MB. The Leadtek WinFast PX8800 GT Extreme adds a few percent more at high resolutions.
The frame rate is fixed at 30fps in this game as this is the rate at which the physical model is being updated at the server. Thus, this 30fps speed is the required minimum for playing the game.
The Nvidia GeForce 8800 GT with the reference frequencies is slower than the GeForce 8800 GTS 320MB at high resolutions in this game. As we know, Quake Wars is a rather simple game graphically, but it does employ huge-resolution textures (for details see our article called Unreal Battlefield: Enemy Territory Quake Wars Performance Preview). That’s why the memory bandwidth is a serious factor especially when FSAA is turned on. Indeed, the overclocked GeForce 8800 GT offered by Leadtek matches the GeForce 8800 GTS 320MB in this parameter and the two cards have identical results from the resolution of 1600x1200. The new card is unrivalled at the lower resolutions.
Like Battlefield 2142, this game does not support resolutions of 16:10 format. So, we use 1920x1400 (4:3 format) instead of 1920x1200 in this test.
The Nvidia GeForce 8800 GT is about 5% slower than the ATI Radeon HD 2900 XT in terms of average performance but has a twice higher minimum of speed. ATI’s solution loses its ground at the higher resolutions notwithstanding its 512-bit memory interface whereas the Nvidia ensures a comfortable speed even at 1920x1440. The overclocked version from Leadtek is 3-5% ahead of the reference card.
The reference and pre-overclocked versions of GeForce 8800 GT are both far ahead of every other graphics card but Episode Two is a heavier application than Episode One and none of these cards, including the Leadtek, cannot provide the desired frame rate of 55-60fps at 1920x1200 with enabled 4x FSAA. Both versions of the new card from Nvidia are slower than 55fps at 1600x1200 but the game seems to be playable anyway.
The game doesn’t support FSAA when you enable the dynamic lighting model, but loses much of its visual appeal with the static model. This is the reason why we benchmarked the cards in S.T.A.L.K.E.R. using anisotropic filtering only.
The ATI Radeon HD 2000 series still has serious problems in this game: the Radeon HD 2900 XT can’t provide a comfortable speed even at 1280x1024. This is not a problem for the Nvidia GeForce 8800 GT, though, as it makes the resolution of 1600x1200 playable as well (its minimum speed is never lower than 25fps). The reference and pre-overclocked versions of the card are 25% and almost 40% ahead of the GeForce 8800 GTS 320MB!
The overclocking is most beneficial at 1280x1024 where the frequency of the shader processors affects the performance most while the influence of the memory subsystem is minimal. Here, it is about 17% at first and then lowers to 11% at 1600x1200 and to 9% at 1920x1200.
None of the cards, including the Leadtek WinFast PX8800 GT Extreme, can provide a comfortable speed in this game at the highest level of detail. As we found out earlier, you can drop some of the game settings from High to Medium to improve its speed while losing little in visual quality.
Anyway, the title of the winner in the mainstream category goes to the Nvidia GeForce 8800 GT. This card easily beats the more expensive solutions from the opposing camp such as ATI Radeon HD 2900 XT 512MB. On the other hand, the PC version of Lost Planet was optimized for Nvidia’s hardware under Nvidia’s program “The Way It’s Meant to be Played”.
The GeForce 8800 GT profits a lot from overclocked frequencies in this test, yet its standard version is also good at every resolution, providing very comfortable conditions for the gamer. Its results are quite acceptable for any dynamic first-person shooter, let alone a third-person one whose requirements to the average and minimum speed are lower. ATI’s Radeon HD 2900 series cards have a good average but a low minimum speed which makes the game barely playable on them.
The current version of the game doesn’t support FSAA, so we performed the test with anisotropic filtering only.
The Nvidia GeForce 8800 GT isn’t far better than the GeForce 8800 GTS 320MB at 1280x1024 but then the gap between them grows to 20% at 1600x1200 and to 26% at 1920x1200.
The new card features the best result in the category of mainstream cards priced at $259 and delivers acceptable performance at every resolution, including 1920x1200, although the Gothic 3 engine is quite a resource-hungry one. For some reason, the overclocked version of the GeForce 8800 GT from Leadtek is always inferior to the reference card in terms of minimum speed here.
The game loses much of its visual appeal without HDR. Although some gamers argue that point, we think TES IV looks best with enabled FP HDR and test it in this mode.
Outperforming the rest of the participating cards, the Nvidia GeForce 8800 GT looks splendid in this test, but the walk along the dungeons of the empire’s capital is just a warm-up. The real test will follow next. Like in some other cases, the GeForce 8800 GT benefits the most from overclocking at high resolutions only: there is only a minor increase in the minimum speed at 1280x1024.
The Nvidia GeForce 8800 GT passes the open-scene test just as brilliantly. Its minimum speed is never lower than 30fps even in the worst case – you can have such a large reserve of speed only from a Radeon HD 2900 XT 512MB. The new card has no rivals in terms of average performance: the GeForce 8800 GTS 320MB is 20% behind at 1920x1200 while the other cards, including ATI’s Radeon HD 2900 XT 512MB and Radeon HD 2900 Pro, are slower yet. Such a high level of performance could only be achieved previously by means of expensive, uneconomical and very hot multi-GPU tandems, but now it is available for every gamer who can afford a $259 graphics card.
The new add-on to Company of Heroes is tested in DirectX 10 mode only since it provides the highest quality of the visuals.
Like the other graphics cards, the Nvidia GeForce 8800 GT has a low minimum of speed, although it is better than, for example, the ATI Radeon HD 2900 XT and Pro in this respect. But the average frame rate provided by both versions of the new graphics card is far better than the results of their opponents, including the Nvidia GeForce 8800 GTS 320MB. The latter just lacks more memory to perform faster with enabled FSAA, so it is natural to see the Nvidia GeForce GT with 512MB of memory on board, in the lead. Save for occasional performance hits, the game seems playable on the GeForce 8800 GT at resolutions up to 1600x1200 and 1680x1050 inclusive. The average frame rate is quite high.
ATI’s Radeon HD 2900 cards with their good technical characteristics suffer from insufficiently optimized drivers here. AMD/ATI didn’t have an access to the game that was developed with Nvidia’s participation and had to perform the optimization after the release of the game. Perhaps the performance of these cards will be improving with newer versions of the Catalyst driver.
The Nvidia GeForce 8800 GT scores only 328 points more than the ATI Radeon HD 2900 Pro in 3DMark05. The pre-overclocked version from Leadtek increases the gap by 362 points more. 3DMark05 defaults to the resolution of 1024x768, so these numbers are not reflective of the new card’s real performance. The overall scores suggest that the new card is better than the GeForce 8800 GTX, yet we know that it’s not true from the gaming tests. So, let’s better see what we have in the individual tests.
The results of the first test are logical since the GeForce 8800 GT is better than the GeForce 8800 GTS 320MB in every parameter including the frequency of the main domain and the number of TMUs. The speed of the shader processors is not crucial in this test and the effect from overclocking can only be observed at high resolutions.
The second test puts a stress on math1ematical computations and the Leadtek WinFast PX8800 GT Extreme is everywhere ahead of the reference card. The Nvidia solution is also ahead of the Radeon HD 2900 XT/Pro that has 320 ALUs, although the gap is smaller here.
The Nvidia GeForce 8800 GT and its overclocked version from Leadtek are also superior in the third test. This resembles the first test except that ATI’s Radeon HD 2900 are not so far behind the leader.
Judging by the overall scores, the Nvidia GeForce 8800 GT feels at ease in 3DMark06 than in 3DMark05. The new card is slightly better than the ATI Radeon HD 2900 XT 512MB and far superior to the ATI Radeon HD 2900 Pro.
As was to be expected, the GeForce 8800 GT wins the SM2.0 tests since it has 28 TMUs clocked at 600MHz. In the SM3.0/HDR tests it is the speed of processing complex shaders that becomes the crucial factor and the new card is slightly slower than the ATI Radeon HD 2900 XT 512MB. You should be aware, though, that the latter is a top-end product that costs much more money.
The GeForce 8800 GT enjoys a big lead in both SM2.0 tests. It can only be rivaled by Nvidia’s more expensive GeForce 8800 GTX that has 32 texture processors.
The new card from Nvidia is not that brilliant in the individual SM3.0/HDR tests, yet it has a 15% advantage over the ATI Radeon HD 2900 XT 512MB in the first test and equals it in the second test while having a lower price.
The new GPU from Nvidia is not just a success. We can say that it pushes the performance bar in the $199-259 sector up much higher, and it is in this sector that there have been serious problems with gaming performance until the arrival of the GeForce 8800 GT. The new card is a kind of revolution since the GeForce 8800 GTS 640MB/320MB become outdated while the GeForce 8800 GTX is not as much better than the new card as to justify its price of $599.
The release of the GeForce 8800 GT cannot but provoke serious changes in Nvidia’s current product line-up. The GeForce 8800 GTS 320MB will be discontinued, while the GeForce 8800 GTS will be replaced with an updated version that has 112 active shader processors to fill in the gap between the GeForce 8800 GT and GeForce 8800 GTX. Perhaps, the latter card will become cheaper because it doesn’t seem right to buy a $599 card while you can get almost the same performance from a $259 product unless you want the support of triple- or quad-SLI technologies for further upgrade.
Although belonging to the mainstream category, the Nvidia GeForce 8800 GT 512MB delivers fantastic performance in most of today’s games, being just a little inferior to the monstrous GeForce 8800 GTX. That’s the effect of the tech process that helped increase the frequency of the shader processors greatly. Moreover, the Leadtek WinFast PX8800 GT Extreme sometimes had a higher performance than the GeForce 8800 GTX in our tests! The new card was only occasionally inferior or equal to the ATI Radeon HD 2900 XT 512MB. So, Nvidia hasn’t developed a new GPU for the high-end market in the whole year since the release of the G80, but it now offers a product that makes the speed of the GeForce 8800 GTX affordable for buyers of mainstream graphics cards.
It is yet too early to claim that the Nvidia GeForce 8800 GT is the best graphics card of this winter, though. The recently announced ATI Radeon HD 3850 and HD 3870 are targeted at the below-$239 price sector but support DirectX 10.1, a HDMI interface and a better implementation of multi-GPU technologies.
We should also note certain drawbacks of the new product, the most important of which is its incompatibility with some mainboards that support the older PCI Express 1.0a standard. Perhaps this can be corrected by BIOS updates, yet the problem exists and many potential users can be taken aback at this news. We recommend you to make sure your mainboard supports PCI Express 1.1 before purchasing a GeForce 8800 GT. The new card has no problems with that standard as well as with PCI Express 2.0.
Another drawback that needs mentioning is the design of the cooling system. It’s not quite clear why the developer limited himself to the single-slot form-factor if dual-slot coolers have long become a typical attribute of top-end graphics cards. Perhaps they wanted to simplify the installation of the GeForce 8800 GT into small system cases such as cases for multimedia systems. So, the cooler is indeed compact, but this compactness means a reduced efficiency since the hot air is not exhausted out of the system. We guess there’ll be quite a lot of GeForce 8800 GT models with more efficient coolers on the market, though.
The Leadtek WinFast PX8800 GT Extreme doesn’t differ from Nvidia’s reference card in anything except for the GPU and memory frequencies. This product indicates that G92-based solutions feature excellent scalability and can easily reach the performance of a GeForce 8800 GTX and even higher.
The Leadtek product deserves our praises for its performance as well as accessories, particularly the copy of Neverwinter Nights 2. Of course, the pre-overclocked version of GeForce 8800 GT is going to cost more than the version with the reference frequencies, yet still much less than $599, which is the price of the 8800 GTX. So, if you are searching for a compromise between top performance and price, the Leadtek WinFast PX8800 GT Extreme may be just what you need.
Nvidia GeForce 8800 GT 512MB
Leadtek WinFast PX8800 GT Extreme