NVIDIA has held  a dominant position in the high-end GPU space ever since the introduction of the G80 architecture back in November of 2006.  When the G80 first arrived in the form of the GeForce 8800 GTX and the pared-down GeForce 8800 GTS, the cards were clearly superior to anything else at the time in terms of both performance and features.  As time has passed, rival ATI has closed the gap somewhat with the Radeon HD 3800 series, and has even surpassed NVIDIA in some ways, for example with  DX10.1 and native HDMI with audio support, but through it all NVIDIA offered enthusiast gamers what they arguably desired most - high performance.  And it shows in the discreet graphics card market share data.

NVIDIA eventually refreshed their product line-up with the 65nm G92 GPU, which incorporated support for their PureVideo HD video processing engine with some texture processing tweaks thrown in for good measure as well, but the G92 didn't offer significantly more performance than G80.  In reality, cards based on the G92, like the GeForce 9800 GTX, were actually inferior in some ways.  With today's launch of their GT200 GPU, however, NVIDIA has truly taken things to the extreme.  While the GT200 does borrow heavily from previous generation GPU architectures, it is simply in a class of its own at the moment.  And as you'll see, the jump in performance is similar to what we've all gotten use to and experienced again, when the GeForce 8800 GTX was first introduced.

The first two cards based on the GT200 GPU architecture are the GeForce GTX 280 and GeForce GTX 260.  As their names suggest, the GTX 260 is a pared down version of the flagship GTX 280.  We'll explain the differences in detail a little later on.  For now, take a look at the features and specifications posted below.

NVIDIA GeForce GTX 280 Graphics Card

NVIDIA GeForce GTX 280 and GTX 260 Specifications and Features

      
The GeForce GTX 280 and GTX 260 GPUs

Before we get down to the technical nitty gritty, you'll notice NVIDIA has a new naming convention for these cards.  GeForce GTX denotes the current high end, with the model number afterward its performance level in relation to other cards in the same family.

In regard to features, however, the GT200 GPU has a lot in common with the G80 and G92.  There is still no DX10.1 support, the video processing engine hasn't changed, and the GPUs are manufactured using TSMC's 65nm process node.  Of course there are some significant tweaks to the architecture that increase performance dramatically for not only 3D rendering but stream computing as well.

For more detailed explanations on many of the features shared between the GT200 and NVIDIA's previous GPU architectures, we'd suggest taking a look at a few previous HotHardware articles:

• GeForce 8800 GTX and 8800 GTS Launch
• GeForce 8800 GT: G92 Takes Flight
• GeForce 8800 GTS 512MB Refresh
• GeForce 9800 GTX Launch
• GeForce 9800 GX2 and nForce 790i SLI Launch

Our GeForce 8800 GTX launch article goes in depth on the G80 GPU architecture and explains NVIDIA's CUDA GPGPU technology.  Also, our GeForce 8800 GT, 8800 GTS 512MB, 9800 GTX and GX2 pieces encompass the majority of our G92 GPU coverage.

Unlike the initial G92 launch with the GeForce 8800 GT, where we were left with questions regarding the total number of stream processors and texture units inside the GPU, with today's GT200 launch there are no such questions.  The die shot below clearly shows the ten banks of 24 stream processor clusters.  Count them yourself if you'd like verification.

What you're looking at here is a 1.4 billion transistor GPU, produced on TSMC's 65nm process node. As the overlay shows, the stream processor clusters, texture units, ROPs, and frame buffer memory partitions are all segmented and spread out across the entire GPU. We should note, that the GT200 is the largest, most complex chip TSMC has ever manufactured.  The wafer shot to the left here shows just how big the GT200 is, and also hints to to fact this this is going to be an expensive part to produce.  If you count the number of full dies on that wafer, you'll see there are only 94 complete units.  In addition, the edge die that get chopped and wasted in the circular shaped of the wafer, are numerous.  Then account for defects and of course that number only goes down.  Although they haven't officially acknowledged it just yet, just looking at that wafer it's highly likely that NVIDIA has plans to shrink this GPU architecture ASAP.  45nm would do this chip a whole lot of justice.

And although the GT200 does borrow heavily from G80 / G92 architectures, there are some significant enhancements here, over and above the increased number of stream processing units.  The GT200 for example, supports three times the number of threads in flight as the G80 at any given time. It also has a new scheduler design this is up to 20% more efficient.  The GT200 also supports a wider 512-bit memory interface and has improved z-cull and compression technology.  This new GPU has more aggressive clock gating and power saving technology built-in too, which results in significantly lower idle power usage than you might expect and it support's NVIDIA's HybridPower feature as well.

Unlike the G80 which performed ROP frame buffer blending at half speed, the GT200 can perform the same tasks at full-speed.  The GT200 GPU also sports twice the number of registers for longer, more complex shaders.  The chip's output buffer size has been increased by a factor of six and it offers IEEE 754R compliant double precision for improved floating-point accuracy.  And although the first crop of cards based on the GT200 are equipped with DVI outputs that won't exploit it, the GPU has hardware support for 10-bit color scan out as well, but it'll need a DisplayPort interface to use it.

GT200 GPU Block Diagram - Graphics Mode Configuration

The high-level block diagram above represents the GT200 GPU operating in graphics mode on a GeForce 280 GTX. The shader thread dispatch logic is located at the top along with the setup and raster units. The ten thread processor clusters each include three groups of eight stream processors for a total of 24 processing cores per cluster.  Multiply that by ten and you have a grand total of 240 scalar processing cores. ROPs (raster operations processors) and memory interface units are located at the bottom.  The ROPs can deliver a maximum of 32 pixels per clock, which equates to 4 pixels per clock, per ROP partition, times eight partitions. We should also note that up to 32 color and Z samples per clock for 8X MSAA are supported per ROP partition as well.  The 512-bit wide memory controller is partitioned into eight segments, with two chunks of memory connected to each segment.

The GT200 is implemented differently on the new GeForce GTX 260, however.  On the GTX 260, two of the thread processor clusters and one of the memory partitions are disabled, for a total of 192 active SP units and a 448-bit memory interface, with 14 pieces of memory.

GT200 GPU Block Diagram - Compute Mode Configuration

The high-level diagram above is representative of the GT200 GPU operating in compute mode. The hardware-based thread scheduler at the top manages the scheduling of threads across the thread processor clusters. In compute mode, the texture caches are used to combine memory accesses for more efficient and higher bandwidth memory read / write operations. And while the ROPs are out of the picture, there are some new elements, called "Atomic", visible. The Atomics refer to the ability to perform atomic read-modify-write operations to memory. The Atomics provide granular access to memory and facilitate parallel reductions and parallel data structure management.

We should  note, that the GPU cannot operate in both modes simultaneously - it is either in compute mode or 3D graphics mode.  When PhysX support comes in a future driver release, for example, CUDA compatible GPUs like the GT200 (G80 and G92) can be used as a 3D accelerator or a PhysX accelerator at any given time, but not both simultaneously.  The GPU must switch states depending on the type of workload.

The actual GeForce GTX 280 and GTX 260 cards don't look much different than current GeForce 9800 GTX and GX2 cards, due to the shells surrounding the cards.

       
Click images for full view

The GeForce GTX 280 reference card picutred here has a GPU clock of 602MHz, with a Stream Processor clock of a1,296MHz.  The full 240 SP cores in the GT200 GPU are enabled, and the card sports a 1GB frame buffer consisting of 16 pieces of DDR3 memory clocked at 1.1GHz (2.2GHz DDR), connected via a 512-bit memory interface.  This configuration offers a peak texturing fillrate of a 48.2GTexels/s and over 141GB/s of memory bandwidth.

There are two dual-link DVI outputs on the card, along with an HD/TV output.  HDMI with audio is supported through the use of an adapter, and audio signals can be passed through the card by way of an SPDIF input, similar to the one found on the 9800 GTX and GX2.  HDCP is supported as well.

The GeForce GTX 280 required both a 6-pin and an 8-pin PCI Express power connector and max board power hovers around 236 watts.

       
Click images for full view

As we've mentioned, the new GeForce GTX 260 is essentially the same as the GTX 280, sans a couple of thread processor clusters and memory partitions. The GT200 GPU used on the GTX 260 has 192 stream processors enabled. The GPU is clocked at 576MHz on the reference card, with a 1,242MHz stream processor clock. 866MB of on-board DDR3 frame buffer memory is clocked at 999MHz (1.99GHz DDR), connected to the GPU over a 448-bit interface.  The GeForce GTX 260 offers a peak texturing fillrate of 36.9GTexels/s with 111.9GB/s of memory bandwidth.

The output options are the same on the GTX 260, but power requirements are more modest.  The GeForce GTX 260 requires two 6-pin PCI Express power connectors and has a max power of 182 watts.

Something you don't see in these pictures (but you will see on the next page) is that both the GeForce GTX 260 and GTX 280 sport of pair of SLI edge connectors and both support two-card and three-way SLI configurations.

For the purpose of this review, we got our hands on a couple of retail-ready GeForce GTX 280 cards for testing. A third card we used for 3-way SLI was an engineering sample provided by NVIDIA, as was the GeForce GTX 260.

The two retail-ready cards came by way of EVGA and ASUS.  Below are EVGA's GeForce GTX 280 FTW Edition and ASUS' ENGTX280 TOP.

With the exception of the complex graphics affixed to the card's outer shell, EVGA's GeForce GTX 280 is physically identical to NVIDIA's reference design.  EVGA, however, chose to up the card's default GPU and memory clocks.  Whereas NVIDIA's reference design calls for a 602MHz GPU clock with 1.1GHz memory, EVGA has configured their FTW Edition card with a 670MHz GPU and 1.21GHz memory.  The frame buffer size is 1GB, and all outputs and connected are identical to the reference design.

When we cracked open the box, we found a typical assortment of accessories and software.  In the box, we found the obligatory driver CD, a user's guide, a dual 6-pin to 8-pin PCI Express power adapter, a dual-Molex to 6-pin PCI Express power adapter, dual DVI to VGA adapters, an HD component output dongle, and an EVGA case badge. We should also note, that the driver CD contained a copy of EVGA's "PRECISION" overclocking utility, and this card, like other EVGA cards, carries a lifetime warranty.
 

The Asus ENGTX280 TOP was also based on NVIDIA's reference design, and also had a 670MHz GPU clock with 1.21GHz memory.  The card is emblazoned with pink, purple, and brown camouflage, with a warrior chic smack dab in the center.  ASUS too bundles in the usual assortment of goodies with the ENGTX280 TOP, including driver and utility discs, a user's guide, a dual-Molex to 6-pin PCI Express power adapter, a DVI to VGA adapter, an HD component output dongle, and a "pleather" CD / DVD case, with matching mouse pad.

Much of NVIDIA's recent marketing has centered around the notion that GPUs aren't just for gaming any longer, which you'd expect considering they've got a GPU architecture in their arsenal capable of 933 GFlops.  Of course, we've known this to be true for quite some time, but there definitely seems to be more happening in the GPGPU arena as of late.  During our briefing for the GeForce GTX 200 series architecture, for example, NVIDIA showed off a number of applications that all benefited from the power of a GPU, and none of them were games.

The application you see pictured above is a beta, pre-release version of Elemental Technologies’ BadaBOOM.  BadaBOOM takes advantage of ETI’s GPU-powered RapiHD Video Platform to offload video encoding duties from the CPU, onto the GPU, to accelerate the process of converting standard-definition video from any format to H.264 for portable media devices, like an iPod, Zune, or iPhone.  Using BadaBOOM and a GeForce GTX 280, we were able to encode an MPG of the digital short "The Plush Life" in only 24 seconds, at a rate of about 140 - 150 FPS.  To give you a point a reference, it took an 8-Core Skulltrail rig almost twice as long to complete a similar encoding process using Nero Recode, at a rate of about 85 - 95 FPS.

NVIDIA, along with representatives from Stanford University, also took the opportunity to showcase a brand new version of the Folding @ Home client which used the GPU for its calculation. This version of the Folding @  Home client running on a GeForce GTX 280 can processes roughly 500ns / day. That is a massive speed increase compared to existing CPU and previous GPU architectures.  A typical CPU can do about 4ns / day, a PS3 about 100, and Radeon HD 3870 approximately 170.

Of course we can't forget PhysX.  If you remebmer, NVIDIA acquired AGEIA not too long ago and plans to incorporate PhysX support into all CUDA capable (GeForce 8, 9, and GTX 200 series) graphics card.  While out an NVIDIA's Editor's Day even, we saw numerous demos in action and heard from representatives from NVIDIA and AGEIA.  By all accounts, the relationship is going well and according to information given to us at the event, having NVIDIA's marketing muscle behind the technology has resulted in a number of new developers signing on to use the technology.  Support for PhysX should be coming in a future driver revision, due out in a few weeks time.

Many other types of complex mathematical calculations are also well suited to GPU acceleration.  To further demonstrate, we enlisted the help of a small application dubbed "GPUQuant" that performs Black & Scholes or Monte Carlo calculations on either a CPU or GPU.  The results above speak for themselves.  The new GeForce GTX 200 series architecture offers significantly more compute performance than any existing CPU or GPU for this type of calculation.  We should point out, however, this application was NOT multi-threaded when running on a CPU, so theoretically similar calculations could perform at approximately 4x the rate shown above on a quad-core CPU, but even then it would still offer only a fraction of the throughput of the GPUs.

HOW WE CONFIGURED THE TEST SYSTEMS: We tested all of the graphics cards used in this article on either an Asus nForce 790i SLI Ultra based Striker II Extreme motherboard (NVIDIA GPUs) or an X48 based Asus P5E3 Premium (ATI GPUs) powered by a Core 2 Extreme QX6850 quad-core processor and 2GB of low-latency Corsair RAM. The first thing we did when configuring these test systems was enter their respective BIOSes and set all values to their "optimized" or "high performance" default settings. Then we manually configured the memory timings and disabled any integrated peripherals that wouldn't be put to use. The hard drive was then formatted, and Windows Vista Ultimate was installed. When the installation was complete we fully updated the OS, and installed the latest DX10 redist and various hotfixes, along with the necessary drivers and applications.

HotHardware's Test Systems Intel and NVIDIA Powered

Hardware Used: Core 2 Extreme QX6850 (3GHz) Asus Striker II Extreme (nForce 790i SLI Ultra chipset) Asus P5E3 Premium (X48 Express) Radeon HD 3870 X2 (x2) GeForce 9800 GTX (x3) GeForce 9800 GX2 (x2) GeForce GTX 260 GeForce GTX 280 (x3) 2048MB Corsair DDR3-1333 C7 (2 X 1GB) Integrated Audio Integrated Network Western Digital "Raptor" 74GB (10,000RPM - SATA)

Relevant Software: Windows Vista Ultimate SP1 DirectX June 2008 Redist NVIDIA Forceware v177.34 ATI Catalyst v8.5 Benchmarks Used: 3DMark06 v1.0.2 3DMark Vantage v1.0.1 Unreal Tournament 3 v1.2* Crysis v1.2* Half Life 2: Episode 2* Enemy Territory: Quake Wars* * - Custom Demo

Futuremark 3DMark06 Synthetic DirectX Gaming

3DMark06

3DMark06 is a synthetic benchmark, designed to simulate DX9-class game titles. This version differs from the earlier 3Dmark05 in a number of ways, and includes not only Shader Model 2.0 tests, but Shader Model 3.0 and HDR tests as well. Some of the assets from 3DMark05 have been re-used, but the scenes are now rendered with much more geometric detail and the shader complexity is vastly increasedl. Max shader length in 3DMark05 was 96 instructions, while 3DMark06 ups that number to 512. 3DMark06 also employs much more lighting and there is extensive use of soft shadows. With 3DMark06, Futuremark has also updated how the final score is tabulated. In this latest version of the benchmark, SM 2.0 and HDR / SM3.0 tests are weighted and the CPU score is factored into the final tally as well.

3DMark06's overall score from the default benchmark test doesn't tell use very much, mostly because it is CPU bound - even with a Core 2 Extreme QX6850 at the heart of our test system.  The new GeForce GTX 260 puts up a score well ahead of the 9800 GTX, but behind the GX2 and Radeon HD 3870 X2. The GTX 280 fares better but is still limited due to the CPU bound circumstance of this test.

3DMark06's individual Shader Model tests tell essentially the same story.  The new GeForce GTX 260 and GTX 280 put up scores measurably better than any other single GPU, but the differences aren't very significant. Thankfully, we can now retire 3DMark06, because Futuremark's latest version, 3DMark Vantage, has hit the scene.

Futuremark 3DMark Vantage Synthetic DirectX Gaming

3DMark Vantage

The latest version of Futuremark's synthetic 3D gaming benchmark, 3DMark Vantage, is specifically bound to Windows Vista-based systems because it uses some advanced visual technologies that are only available with DirectX 10, which y isn't available on previous versions of Windows.  3DMark Vantage isn't simply a port of 3DMark06 to DirectX 10 though.  With this latest version of the benchmark, Futuremark has incorporated two new graphics tests, two new CPU tests, several new feature tests, in addition to support for the latest PC hardware.  We tested the graphics cards here with 3DMark Vantage's Extreme preset option, which uses a resolution of 1,920 x 1,200, with 4X anti-aliasing and 16X anisotropic filtering.

As the graph above show, results generated with 3DMark Vantage's Extreme preset testing option produce much more diverse results than the again 3DMark06. According the 3DMark Vantage, the GeForce GTX 280 is easily the most powerful graphics card we tested.  The GeForce GTX 260 also surpassed all other singe-GPU powered cards, but it wasn't quite on the level of the dual-GPU powered GeForce 9800 GX2.  Vantage also shows the excellent multi-GPU scaling offered by the NVIDIA powered cards - the quad-SLI and two- and three-way SLI GeForce GTX 280 setups simply crush the competition.  Please note, however, the Radeon HD 3870 X2 CrossFireX configuration had scaling issues with the Catalyst 8.5 drivers.  Future driver revisions should help its score here significantly.

3DMark Vantage's individual GPU tests also demonstrate the excellent performance of the GeForce GTX 280 and 260.  Once again, both cards were clearly superior to any other single-GPU powered configuration and the GTX 280 was faster than all of the dual-GPU setups as well.  Standard SLI and 3-Way SLI with the ASUS and EVGA cards we tested also showed excellent scaling here, producing results that were leaps and bounds ahead of anything else.

Half Life 2: Episode 2 DirectX Gaming Performance

Half Life 2: Episode 2

Thanks to the dedication of hardcore PC gamers and a huge mod-community, the original Half-Life was one of the most successful first person shooters of all time. And courtesy of an updated game engine, gorgeous visuals, and intelligent weapon and level designs, Half Life 2 became just as popular.  Episode 2 - the most recent addition to the franchise - offers a number of visual enhancements including better looking transparent texture anti-aliasing. These tests were run at resolutions of 1,920 x 1,200 and 2,560 x 1,600 with 4X anti-aliasing and 16X anisotropic filtering enabled concurrently.  Color correction and HDR rendering were also enabled in the game engine as well.  We used a custom recorded timedemo to benchmark all cards for these tests.

Our custom Half Life 2: Episode 2 benchmark corroborates what 3DMark Vantage told us on the previous page, for the most part.  Here, the GeForce GTX 280 and GTX 260 are clearly superior to any other single-GPU out there.  The GTX 280 also outperformed the dual-GPU setups represented here, and even managed to outpace the 3-way 9800 GTX SLI and dual-9800 GX2 quad-SLI rigs at the lower resolution.

When paired up in a duet or trio of GTX 280 cards, NVIDIA's latest flagship shows good scaling with two cards, but the three-way SLI setup was CPU bound and actually performed lower than standard SLI setup due to increased CPU overhead.

Unreal Tournament 3 DirectX Gaming Performance

Unreal Tournament 3

If you're a long-time PC gamer, the Unreal Tournament franchise should need no introduction.  UT's fast paced action and over the top weapons have been popular for as long as Epic has been making the games.  For these tests, we used the latest addition to the franchise, Unreal Tournament 3.  The game doesn't have a built-in benchmarking tool, however, so we enlisted the help of FRAPS here.  These tests were run at resolutions of 1,920 x 1,200 and 2,560 x 1,600 with no anti-aliasing or anisotropic filtering enabled, but with the UT3's in game graphical options set to their maximum values, with color correction enabled.

We saw more of the same with Unreal Tournament 3. In this game, most of the high-end cards produce somewhat similar results at 1,920 x 1,200, with only the GeForce 9800 GTX trailing considerably. At 2,560 x 1,600, however, the deltas get larger and the GeForce GTX 280 and GTX 260 show their mettle.  The GTX 280 was once again the fastest of the single-GPU powered card. It wasn't quite as fast as a 9800 GTX SLI setup, however.  And scaling with standard SLI and 3-way GTX 280 SLI looked much like HL2:EP2 - because the test is CPU bound with that much GPU horsepower in the system, performance doesn't increase all that much, if at all.

 

Enemy Territory: Quake Wars OpenGL Gaming Performance

Enemy Territory: Quake Wars

Enemy Territory: Quake Wars is Based on a radically enhanced version of id's Doom 3 engine and viewed by many as Battlefield 2 meets the Strogg, and then some.  In fact, we'd venture to say that id took EA's team-based warfare genre up a notch or two.  ET: Quake Wars also marks the introduction of John Carmack's "Megatexture" technology that employs large environment and terrain textures that cover vast areas of maps without the need to repeat and tile many smaller textures.  The beauty of megatexture technology is that each unit only takes up a maximum of 8MB of frame buffer memory.  Add to that HDR-like bloom lighting and leading edge shadowing effects and Enemy Territory: Quake Wars looks great, plays well and works high end graphics cards vigorously.  The game was tested with all of its in-game options set to their maximum values with soft particles enabled in addition to 4X anti-aliasing and 16x anisotropic filtering.

The new GeForce GTX 280 and GTX 260 performed very well in our custom Enemy Territory: Quake Wars test. Once again, both cards were clearly superior than any other single-GPU based product we tested.  The GeForce GTX 280 was faster than all of the dual-GPU setups as well; it took the 3-way GeForce 9800 GTX and dual-GeForce 9800 GX2 quad-SLI rigs to take it down.  Scaling in a two card GeForce GTX 280 SLI configuration was also quite good.  But 3-way GTX 280 SLI was limited at 1,920 x 1,200.  At 2,560 x 1,600, 3-way offered up the best performance of the bunch, but it was only slightly faster then the two-card setup.

Crysis v1.2 DirectX 10 Gaming Performance

Crysis

If you're at all into enthusiast computing, the highly anticipated single player, FPS smash-hit Crysis, should require no introduction. Crytek's game engine produces some stunning visuals that are easily the most impressive real-time 3D renderings we've seen on the PC to date.  The engine employs some of the latest techniques in 3D rendering like Parallax Occlusion Mapping, Subsurface Scattering, Motion Blur and Depth-of-Field effects, as well as some of the most impressive use of Shader technology we've seen yet.  In short, for those of you that want to skip the technical jib-jab, Crysis is a beast of a game.  We ran the full game patched to v1.2 with all of its visual options set to 'High' to put a significant load on the graphics cards being tested  A custom demo recorded on the Island level was used throughout testing.

Crysis tells essentially the same story as all of our previous in-game tests.  Here, the new GeForce GTX 280 and GTX 260 best all other single-GPU based rigs. They both fall victim to the GX2 and 9800 GTX SLI setups, however.  Multi-GPU scaling with Crysis was also good with the new GeForce GTX 280, putting up scores far superior to anything else we've ever tested.  3-way SLI, however, didn't offer up much of an improvement over two cards, at least not with our quad-core powered test system and the current 177 series drivers.

We also did some quick testing of the GeForce GTX 280's video processing engine, in terms of both image quality and CPU utilization with some HQV and H.264 playback tests.

Both ATI's and NVIDIA's latest GPUs have no trouble with SD video playback.  All three of the cards put up near perfect scores in the HQV test.  In case you're not familiar with HQV, 130 points is the maximum score attainable.  At 128 points, a PC equipped with either of these graphics cards plays back DVD video at quality levels better than the vast majority of set-top DVD players on the market.  We should note, however, the GTX 280 produced noticeably better results in the Jaggies 1 test and somewhat better results in the Jaggies 2 test than the others, but it wasn't enough of a difference to affect the scoring due to HQV recommended scoring system.

Next we conducted a test using an H.264 encoded movie trailer clip for "Beowulf" which is available for download on Apple's QuickTime HD website.  The CPU utilization data gathered during these tests was taken from Windows Vista's built-in Performance Monitor. The graphs show the CPU utilization for a GeForce 9800 GX2 and a Radeon HD 3870 X2 using PowerDVD HD to playback the QuickTime clip.

 
GeForce 9800 GX2



Radeon HD 3870 X2



GeForce GTX 280

With a fast quad-core processor powering our test system and an unencrypted HD video clip being played back, all of the cards we tested had low CPU utilization in this test.  The GTX 280 didn't fair quite as well as the GX2, however, which is confusing because they both sport the same video engine.  Perhaps with future drivers, the GTX 280's performance here will improve a bit. We should note that with hardware acceleration disabled, playing this video clip results in about 12% - 15% average CPU utilization, so there is a marked improvement with both PureVideo HD and UVD.  Also note that with encoded content, like an off the shelf Blu-Ray disc for example, CPU utilization will be measurably higher that what you see here.  However, both platforms should have no trouble playing back high def digital video.

We'd like to cover a few final data points before bringing this article to a close. Throughout all of our benchmarking and testing, we monitored how much power our test systems were consuming using a power meter. Our goal was to give you all an idea as to how much power each configuration used while idling and under a heavy workload. Please keep in mind that we were testing total system power consumption at the outlet here, not just the power being drawn by the motherboards alone.

Total System Power Consumption Tested at the Outlet

As you can see, the GeForce GTX 280 and GTX 260 fare very well in the power consumption department while idling, with both cards consuming the least amount of power. Tax them with a heavy workload, however, and power consumption goes way up.  The GTX 260, for example, consumed 31 more watts than a 9800 GTX under load, and the GTX 280 used over 100 more watts.  Start doubling, or tripling up on the GTX 280s, and consumption hits insane levels.  The standard SLI and 3-way GeForce GTX 280 configuration consumed more power than any other graphics setup we have tested to day.

You'd think with all that power being consumed, the GeForce GTX 280 and 260 would be annoyingly loud, but thankfully that is not the case. At idle, the cards are virtually silent.  Under load, they are considerably louder of course, but it is not unbearable. They are definitely louder than a GeForce 9800 / 8800 GTX, but are about on par with the GX2.  We did consider our two- and three-card GeForce GTX 280 SLI setups to be relatively "loud", however.

We should also note that despite the GPU's large size, it didn't operate at temperatures much higher than existing card. While idling, we reported a GPU temperature on a  GTX 280 of 66'C.  Under load, that temperature went up to about 85'C.

Performance Summary: Summarizing the performance of NVIDIA's new GeForce GTX 280 and GeForce GTX 260 cards is quite easy.  The GeForce GTX 260's overall performance falls somewhere in between the single-GPU based GeForce 9800 GTX and dual-GPU powered GeForce 9800 GX2, and it is usually faster than the dual-GPU powered Radeon HD 3870 X2 as well.  The flagship GeForce GTX 280, however, was overall  the fastest single graphics card we have ever tested.  There were a couple of instances when the GeForce 9800 GX2 pulled ahead of the GTX 280, but in the vast majority of our testing, no other single graphics card could match the performance of the GeForce GTX 280.

NVIDIA has done it again and raised the bar for that can be considered an ultra high-end GPU.  The GT200 series GPU at the heart of the GeForce GTX 280 and GeForce GTX 260 offers more brute force performance than anything else to come before it.  Though producing such a massive GPU, does have its drawbacks, which are evident when you consider its die size and when looking at power consumption characteristics, the fact remains NVIDIA has produced the most powerful and fastest graphics card we have ever tested - yet again.

Suggested retail pricing for the GeForce GTX 280 is set at $649 and the GeForce GTX 260 checks in at a more palatable $399.  NVIDIA has informed is that the GeForce GTX 280 will be available in quantity tomorrow ( June 17th) and the GeForce GTX 260 is slated to arrive Thursday of next week, on June 26.  At those prices, the GeForce GTX series does not come cheap.  But enthusiasts have always had to "pay to play" so to speak.  We suspect there may be some wiggle room in these prices, however, as partners push clock speeds higher than NVIDIA's reference specifications and as ATI pushes out their next-gen GPU architecture, which is rumored to offer strong performance and a relatively low price point.  As usual, we'll know more in the coming weeks.

For now though, NVIDIA has further cemented their position atop the 3D graphics food chain.   They have executed once again with a new line of graphics cards that not only offer significantly higher frame rates, but large performance gains for gamers, game developers and researchers looking to exploit both GPU and GPGPU capabilities of NVIDIA's massive new multi-purpose graphics and compute engine.

Extreme Performance PhysX Support Coming Large Frame Buffers Low Idle Power Consumption	High Peak Power Consumption Expensive Can Be Somewhat Loud