Preceding the launch of the GeForce 8800 series, NVIDIA was having great trouble with “hard launching” its products. For many of its new graphics products the company simply ended up with humble “paper launches”—a launch in which there is no actual hardware available for sale. A number of factors contributed to these troubled launches. NVIDIA might have been facing manufacturing and fabrication problems, translating to a limited of supply of hardware for manufacturers. Manufacturers might have also received actual working samples just a few days in advance of the launch, meaning they weren’t able to get a product out in time.

Whatever the reason behind the paper launches, NVIDIA placed a lot of hard work into making sure its GeForce 8800 launch didn’t suffer from the same problems. This time around, manufacturers received working product samples weeks in advance of the actual product launch. These extra few weeks provided ample time for manufacturers to assemble the cards and create a large stock, effectively taking them out of a position in which they would be rushing up to the last minute to get a finished product out the door. The 8800 launch also focused on having products shipped to retailers in large quantities before the actual launch. This move meant that consumers were able to purchase cards immediately on launch day. For the GeForce 8800, NVIDIA had a total of 15 retailers, online and offline, which were launch partners.

Despite the nearly flawless launch, one thing that was missing from store shelves was an overclocked GeForce 8800GTX/GTS. Although the absence of an overclocked 8800 video card is not a very large problem for consumers with knowledge of overclocking, factory overclocked video cards are a good choice for people with little overclocking knowledge and experience. In addition, since manufacturers usually certify their factory overclocks, they fall under the manufacturer’s warranty. This gives the best of both worlds. A consumer can receive overclocked performance while still being protected by warranty. Manually overclocking, on the other hand, will usually void your manufacturer warranty and manufacturers rarely, if ever, refund you if you damage your video card attempting to overclock it.

The reason no manufacturers were selling factory overclocked cards at launch was simply because NVIDIA forbade them from doing so. As of the end of December NVIDIA has, however, changed its stance and begun allowing manufacturers to sell overclocked versions of the GeForce 8800 GTX and GTS.

There are a number of strong reasons behind why NVIDIA chose to initially ban manufacturer overclocking of the 8800 series. A lot of NVIDIA’s decision was made to prevent its newly released GPUs from malfunctioning. This is based off of some problems the company experienced with overclocked 7900 video cards. In the past couple of months, in fact, many owners of overclocked G70 and G71 GPU-based video cards have been complaining of erratic behavior of their factory overclocked video cards. The cards continue to perform poorly at their overclocked levels despite the overclocked frequencies being certified by their manufacturers. Because of this, NVIDA decided to allow a couple of months to pass before allowing the overclocking of its new series. This decision has helped separate overclocking problems to get mixed up with launch problems; NVIDIA didn’t want consumers mistaking overclocking issues with launch issues. In addition, the decision provided manufacturers with more time to choose chips with the greatest overclocking ability.

XFX is one of NVIDIA’s most active partners in overclocking its products, most of which are usually targeted at the high-end market segment. With NVIDIA now allowing factory overclocking, it is only expected for XFX to be among the first manufacturers to offer an overclocked GeForce 8800 product. The new overclocked series by XFX is indicated by “XXX”. Products are available in both GTX and GTS versions. The new “XXX” series by XFX comes at frequencies 10% higher than NVIDIA’s default 8800 speeds. As can be expected, the overclocked frequencies are covered by warranty, a point we touched on earlier.

We will be analyzing two XFX cards today that come factory overclocked. One is based on an NVIDIA GeForce 8800 GTX GPU and the other is based on an 8800 GTS GPU. The two overclocked video cards will be compared against currently available ATI and NVIDIA solutions. We will also be comparing the two cards against a single 8800 GTX that runs at default frequencies from Foxconn. The card’s exact name is the Foxconn FV-N88XMAD2-ON.

There are currently two versions of the GeForce 8800 available for sale. The 8800 GTX is the high-end version of the 8800 series, and as of right now is also NVIDIA’s flagship product. A slightly downscaled version is also available in the form of the 8800 GTS. Certain features and characteristics have been cut to reduce the cost of the 8800 GTS. Nevertheless, it still brings all the improvements of the G80 architecture including DirectX 10 support along with a unified shader architecture.

NVIDIA will be launching new versions of the 8800 series before AMD releases the R600 in an attempt to counteract the future AMD competitor. This week, actually, NVIDIA has launched a 320 MB version of its GeForce 8800 GTS. All aspects of this video card, including PCB layout and cooling remains the same. The only major difference is that the total amount of memory has been halved. The release of this video card will effectively reduce the cost of enthusiast-level DirectX 10. We will most likely have a review of this video card posted within this week.

The following table are the characteristics while the GPUs are under SLI.

The G80 is the first NVIDIA unified shader architecture. In a unified shader architecture, there are no longer any conventional vertex shader and pixel shader units. Instead, they have all been replaced with general purpose floating-point processors, which NVIDIA likes to call unified stream processors. The stream processors are extremely flexible in terms of the tasks they can handle. They are able to process vertices, pixels, and geometry or physics. In short, a large variety of workloads can be assigned to them.

By using stream processors, NVIDIA hopes to improve the efficiency of its video cards. This is primarily because stream processors can be dynamically assigned to tasks depending on the scene at hand. For pixel shader-intensive scenes, more stream processors can be assigned to processing pixel data. If the scene is more vertex intensive, more stream processors can be assigned to vertex calculations. This method of processing data is a large improvement over the traditional GPU architecture in which there are a definite number of vertex shading units and pixel shading units.

The 8800 GTX has a total of 128 stream processors, all of which operate at a speed of 1,350 MHz. In the G80 architecture, stream processors operate at a frequency that is completely independent of the core clock. For example, the GeForce 8800 GTX has a core frequency of 575 MHz. The stream processors do not run calculations at these speeds but instead at their own independent 1,350 MHz. On the other hand, texture filtering units, which are fully decoupled from the stream processors, along with ROPs operate at the core clock.

The GeForce 8800 GTS has 96 stream processors, 32 less than the 8800 GTX. These 96 stream processors operate at a slightly lower frequency of 1,200 MHz. The GTS version also has a slightly lower core clock of 500 MHz, losing 75 MHz in the change from GTX to GTS.

The GeForce 8800 GTX uses 768 MB of GDDR3 memory, currently the memory type that is most widely used in the graphics industry. Although the architecture does support use of GDDR4, the new memory technology will be used on later cards. The 768 MB of memory runs at a clock frequency of 1,800 MHz. The GeForce 8800 GTS also uses GDDR3 memory. The amount of memory, however, has been slightly reduced to 640 MB. Clock frequency has also been decreased by 200 MHz to 1,600 MHz.

The 8800 GTX holds a total of 12 memory chips. Each chip has a bus width of 32bits, adding up to a 384-bit bus width. The maximum bandwidth of the GeForce 8800 GTX is 86.4 GB/s. The 8800 GTS has a 320-bit memory interface with a total bandwidth of 64 GB/s running at 1,600 MHz. In the case of the GeForce 8800 GTS, the lower bus width is a result of only 10 memory chips coming installed on the card.

The below table lists the technical characteristics of the XFX GeForce 8800 GTX/GTS XXX video cards we will be reviewing. We have placed their values side by side with the default NVIDIA reference card’s characteristics for easy comparison.

The only differences between the standard 8800 GTX and 8800 GTS compared to the “XXX” versions are the frequencies they run at. The XFX GeForce 8800 GTX XXX has a core clock of 630 MHz, which is 55 MHz faster than the default NVIDIA values. Memory for the overclocked GTX is also 200 MHz faster, coming in at a total of 2,000 MHz. The XFX GeForce 8800 GTS XXX core has been overclocked a total of 10% to 500 MHz. The 640 MB of GDDR3 memory has also been factory overclocked by XFX 200 MHz to 1,8000 MHz.

Although the two XFX cards are overclocked, they still use the standard NVIDIA reference design cooling and PCB size. Just like the reference design, the video cards still use a large cooling solution that covers most of the cards’ surfaces. The heatsink is in contact with both the GPU and GDDR3 video memory, meaning that it cools both. Since the cooling solution is quite large, the video card is not able to fit into a single-slot and instead requires two slots for proper installation. The higher clock frequencies don’t add another PCIe power input to the GeForce 8800 GTS, and it still ends up requiring only a single power input. The faster and bulkier GeForce 8800 GTX, as can only be expected, still requires two PCIe power inputs.

A rear view of the card clearly shows that both video cards are identical to their NVIDIA reference counterparts as far as PCB size goes. The longer NVIDIA GeForce 8800 GTX is equipped with two NVIDIA SLI connectors while the downscaled GeForce 8800 GTS only comes with one. For both video cards the SLI connectors are located near the top right of the PCB.

The accessory bundle that XFX includes with the video cards contains nothing very extraordinary. Since the price of these video cards is already quite high, XFX chose on including a light bundle of accessories so as to not increase the final price of the product even further. The bundle includes a driver CD-ROM, a simple but clear video card installation manual, two DVI to VGA adapters, a composite video dongle, and an S-Video connector.

Product: XFX
XFX NVIDIA GeForce 8800 GTX 768MB DDR3 XXX Version
XFX NVIDIA GeForce 8800 GTS 640MB DDR3 XXX Version

Moving on to the Foxconn GeForce 8800 GTX, we again see very little difference between the Foxconn GeForce 8800 GTX and the NVIDIA reference 8800 GTX board. The only difference is the sticker which is pasted onto the cooling solution. Foxconn’s 8800 GTS, a review of which can be accessed here, actually had an LED system integrated into the cooling solution. This feature, however, is not present in the “GTX” version.

The back of the Foxconn GeForce 8800 GTX is identical to the NVIDIA GeForce 8800 GTX reference board, and for the most part any 8800 GTX available for sale. Again, two SLI connectors are placed on the top right of the PCB. Both connector, however, don’t need to be used to have a working SLI configuration. NVIDIA claims the additional connector will be used in future versions of SLI technology.

The accessory bundle consists of a USB gamepad which has a design similar to that of the Xbox 360 controller, except that it is slightly smaller and based on darker colors. Foxconn also provides a simple installation handbook along with two DVI to VGA connectors, an external component video dongle, and two molex to 6pin PCIe connectors.

Product: Foxconn
Foxconn FV-N88XMAD2-ON

We tested power consumption in the same way we always do. We measured the power draw of the entire system using each of the cards installed in two different states; idle and full load. In order to make sure our results were accurate, we used the same exact hardware wherever possible.

• scheda madre: abit AW9D-Max (chipset Intel 975X)
• memoria: Corsair CM2X1024 6400 (3-4-3-9) @ 800 MHz; 2x1 Gbyte
• hard disk: Western Digital WD1600JS - Serial ATA - 7.200 rpm, 160 Gbytes
• processore: Intel Core 2 Extreme X6800
• alimentatore Tagan TurboJet TG1100-U95 (1.100 Watt)

There is a small hardware difference in our SLI tests. In order to properly utilize SLI, we had to use the EVGA nForce 680i motherboard. This is because the Abit motherboard we used for single-card tests supported Crossfire only. For our idle tests, we had an empty desktop with Windows XP booted. Under “full load”, we had the Sciencemark 2.1 benchmark running at a resolution of 2560x1600 pixels, effectively stressing the video subsystem.

Both overclocked 8800s end up consuming more power than their default clocked counterparts. The heat generated by these cards is also higher when overclocked, however, altogether it is still very similar with the default clocked cards. For both the overclocked XFX GTS and GTX video cards we recorded a temperature of 83 degrees Celsius when they were under stress.

Adequately testing the performance of next generation video cards requires that we use a system that presents the video cards with as little bottleneck as possible. For this reason we chose to use the Core 2 Extreme X6800, currently the fastest dual-core CPU that can be bought. We chose to use this processor over the Core 2 Extreme QX6700 because it is able to perform faster in games because of its higher clock speed of 2.93 GHz. The Core 2 Extreme QX6700 doesn’t get all cores 4 completely taken advantage of during games, and so it is simply treated as a 2.67 GHz dual-core processor. In light of this, the Core 2 Extreme X6800 turned out to be the more preferable solution for our tests.

The motherboard we chose for our test platform was the Abit AW9D-Max, which is an LGA Socket 775 solution based on the Intel 975X chipset. The system ran with 2 GB of Corsair XPS CM2X1024-6400C3 memory which operate at timings of 3-4-3-9 at a clock frequency of 800 MHz. The hard disk used for the system was a 160 GB Western Digital Caviar WD1600JS which uses an SATA interface and spins at a speed of 7200 Rounds Per Minute (RPM).

For SLI tests we used an EVGA nForce 680i SLI motherboard, based on the NVIDIA nForce 680i chipset. For Crossfire configurations we used the Abit AW9D-Max motherboard.

The test platform’s operating system was Microsoft Windows XP Service Pack 2, loaded with all updates as of Octboer 2006. The NVIDIA drivers we used for the testing was Forceware 96.94 while for ATI products we used Catalyst 6.10 drivers. Our SLI tests did not use Forceware 96.94 tests, but instead they used Forceware 97.02. To maintain consistency with our previous reviews, we chose to not use the latest drivers available for download by ATI or NVIDIA. The reason behind this is that both companies are currently focused on optimizing their drivers for Windows Vista. NVIDIA and ATI’s most recent Windows XP drivers, on the other hand, have been focused on increasing the compatibility of their video cards with high-definition movie formats. As this is not the focus of today’s we review, we chose to use the older drivers.

The Dell 3007WFP display

One thing we always stress that is a fundamental point of any high-end video card review is the display. There is not much reason to buy a $500 video card if it will only be running low resolutions. For this reason, we bought the largest display we could find for this review so as to test the high-end video cards at extreme resolutions. The Dell 3007WFP is by far one of the largest LCD displays currently available for purchase, and supports a native maximum resolution of 2560x1600.

To effectively stress the video subsystem and to push the video cards to their limits, the lowest resolution we use in this review is 1600x1200. In addition, piling an even larger load on the video cards, anisotropic filtering does not ever get set under 16x. Again, this choice is influenced by our desire to push these video cards to their maximum.

We will be comparing the NVIDIA GeForce 8800 video cards against the NVIDIA GeForce 7950 GX2, NVIDIA GeForce 7900 GTX, and ATI Radeon X1950XTX. The 7950 GX2 was chosen because it is currently the fastest single-card solution, the GeForce 7900 GTX is NVIDIA’s previous top of the range video card solution, and the Radeion X1950 XTX is currently AMD/ATI’s direct contender to the GeForce 8800 series until the R600 finally makes its appearance.

The top of the chart is dominated by the various multi-GPU solutions we tested. The overclocked XFX 8800s are successfully able to perform faster than their counterparts.

The general trend of the tests is that the XFX GeForce 8800 XXXs turn out faster than the default clocked 8800s. Multi-GPU systems are naturally still the fastest, and their tests were mostly included for easy performance comparison.

Shadermark results have two interesting elements to them. For the most part, the highest performers are video cards based on the GeForce 8800 architecture, with slightly higher performance for the overclocked XFX solutions. In addition, we see a breakdown in SLI and Crossfire as they clearly end up performing less than single card solutions during the test.

F.E.A.R. is a game we often used to test the performance of last-generation video cards. It is a title that is rather stressful on the video subsystem, and enabling soft shadows directly stresses the GPU even further. Our copy of the game was patched to Version 1.08. We tested using resolutions of 1600x1200, 1920x1200, and 2560x1600 coupled with 4x anti-aliasing and 16x anisotropic filtering. In the last test we have 4x AA with soft shadows enabled.

The overclocked 8800s have appreciable gains in performance compared to the 8800 video cards running at stock speeds. The XFX GeForce 8800 GTS XXX has the more notable improvement, and has gains of around 9 frames per second at 1600x1200 and 1900x1200. The XFX GeForce 8800 GTX XXX is able to achieve performance that comes in right behind the Radeon X1950 XTX. All of the video cards see a huge drop in performance at resolutions of 2560x1600. We are not sure whether this is a driver issue, or if this is the result of a bottleneck in the video memory.

Again the XFX 8800s are faster than their stock speed counterparts and are able to perform 10% faster than them.

With soft shadows enabled, the GeForce 8800 architectures are clearly able to outperform previous generation architectures, which have always suffered in performance with this setting enabled.

Oblivion is without doubt one of the most demanding PC game titles of 2006. It is able to effectively stress every component of the PC, and can bring even the best of systems to their knees. Despite how useful this video game title is in calculating the performance of video cards, there is no built-in benchmarking utility. Therefore we decided on using Fraps to record the system performance. All settings in the game were maxed out and set to their highest. HDR, however, was the only exception, which we decided to disable. The test only lasted 30 seconds, in which we walked around performing various actions.

The test shows a clear advantage for the XFX overclocked 8800s.

The above graph shows the behavior of the video cards over the length of the 30 second test. It is important to remember; since this is not a timedemo it is nearly impossible to repeat the exact same actions for each video card. Thus, there are minor differences between the tests, although looking at the big picture we performed the actions nearly exactly the same each time around.

Episode 1 is the last expansion of Half-Life 2 which Valve currently distributes through Steam. In this test we have used the resolutions of 1920x1200 and 2560x1600, along with 4x anti aliasing and in the other test HDR enabled. We always keep anisotropic filtering set at 16x. The demo we use is trdem1 which is provided by The TechReport.

Lost Coast is one of the first expansions of Half-Life 2 that was released by Valve. The short level is more of a technological showcase than of an actual game. The tests with Lost Coast were executed using hocbench at resolutions of 1600x1200, 1920x1200 and 2560x1600 with anti aliasing at 4x and anisotropic filtering at 16x.

There is without doubt a large performance advantage for the overclocked XFX cards. Performance scalability moving to higher resolutions is also quite impressive, but we still see drops in performance as moving to higher resolutions directly affects video cards and increases their workload.

Prey is a title that was recently released in 2006. The game is based on a modified version of the Doom 3 engine. For our testing purposes, our version of the game was a retail version, and we tested using the hwzone demo from hocbench. Our test resolutions were 1600x1200 and 2560x1600 paired with 4x anti aliasing and 16x anisotropic filtering.

Using high test resolutions has one clear advantage; it is able to isolate the video card performance from the system performance. Here, again, the overclocked XFX 8800 GTX/GTS perform quite well, effectively beating the stock speed 8800 GTX/GTS video cards. 

For our Serious Sam 2 benchmark we used the “Branchester” scene in the game. Our retail version of the video game was patched to Version 2.070. To benchmark, we used the Hochbench utility and ran tests at resolutions of 1600x1200 and 2048x1536 along with 4x anti aliasing in one test and HDR in the other. Both tests had anisotropic filtering set at 16x.

As has been the general trend for all tests we ran, the overclocked XFX video cards have considerable performance gains over the GeForce 8800s running at stock speeds. Serious Sam 2, however, is not a title that is very demanding on the GPUs, and for this reason as we pass to higher resolutions we don’t see much of an impact on performance compared to how much we have seen with other titles we tested today.

Throughout our tests, the XFX GeForce 8800 GTX XXX along with the XFX GeForce 8800 GTS XXX both obtained overall performance levels that were always faster than the standard clocked GTX and GTS video cards. This therefore indicates that the XFX XXX series is currently the fastest GTX and GTS video cards available for sale. There are, however, some drawbacks of the video cards.

There is no doubt in saying that at the moment the GeForce 8800 series are the fastest video card solutions currently available. The XFX XXX series is able to push the performance of the G80 even further by increasing the GPU and memory clock speed. The increase in speed, however, also results in an increased price. The XFX XXX series, in fact, establishes a new price point for the 8800 series.

The XFX GeForce 8800 GTX XXX is available at an average price of around $639.00 online. The average price of a standard clocked GeForce 8800 GTX, on the other hand, is around $560. This means that you are paying around $80 more for a 10% overclock. Going to the GeForce 8800 GTS XXX solution, we again see similar pricing. The video card can be found for, on average, $450. The average price of a standard 8800 GTSin contrast, is $390, meaning you are paying $60 more for the overclocked version.

In both cases, the only reason behind why XFX has increased the prices of the video cards is because of their higher clock speeds. The accessory bundle comes with equipment that can’t be rated as much more than standard, and the video cards’ designs and cooling solutions are both clones of the NVIDIA reference boards. This means that the only grounds XFX has for increasing the price is the faster clock speed of its XXX series.

The margin of advantage the overclocked cards hold is on average 10% faster higher than a standard clocked 8800 GTX/GTS. At the present moment, this makes XFX’s XXX series the fastest single-slot GeForce 8800 GTX/GTS video cards available. We do not feel, however, that their performance completely justifies the cost. We would have preferred seeing XFX personalizing the XXX series and making the cards more unique. The cooling solution is one place where we feel XFX didn’t place enough emphasis on. Although we faced no temperature issues, the fact that the card is exactly the same as the reference board makes it harder to pay the extra money for it.