Editor's Note: ChrisRay, a contributive member at Rage3D since 2003, brings to the table the insight gleaned during his many years on the enthusiast scene. ChrisRay is administrator at Nvidia's SLI Zone, an Nvidia User Group Member, and a valued member of the video hardware community.

Its been several years since the Geforce 8800GTX first took the graphics industry by storm. We have seen incremental upgrades, such as the 9800GTX, which brought 8800GTX performance to the masses as well as multi GPU cards, such as the 9800GX2, which brought SLI to one board. However each improvement seemed to be accompanied by new drawbacks. For example, the 9800GX2 delivered great performance but relied on AFR for performance as well as being limited to 512MB of memory, a step back from the 768MB available on the 8800GTX. Today Nvidia launches the Geforce GTX 260 and Geforce GTX 280. Are they the true heir apparents to the 8800GTX?

What's in a name?

Nvidia took some criticism for its confusing naming convention with the G92 cards, namely the Geforce 8800GT and Geforce 9800GTX. Going by names alone, it was difficult to determine what card you were actually getting. To simplify this, Nvidia has adopted a new naming scheme: Geforce GTX XXX. The idea is that GTX represents the high end and the numbering convention represents the performance expectations. We'll see if this new naming convention helps the decision making process for both consumers and OEMs.

GTX280

GTX280

GTX280

GTX260

GTX260

GTX260

A cursory look at the cards reveals them to be visually identical, with the same cooler and VGA/DVI connectors. But, just like a hotrod, under the hood is where to find the differences. The GTX260 has part of its shader/ROPS/Texture units cut off in comparison to the GTX280 and only requires two 6 pin connectors; the GTX280 requires both 1ea 6 pin and 1ea 8 pin PSU connectors. Lets next take a closer look at the technical specs to see the differences in detail.

The GT200 Architecture

Specs Explored

At a glance you can see why these chips carry 1.4 Billion transistors. The shader count has been massively increased and the ROPS have been doubled compared to the Geforce 9800GTX. The GTX280 features 1GB of available memory. The GTX260 has had 2 of its clusters disabled. As a result the stream processors, Texture Map units, and ROPS and memory bus have been cut slightly while only featuring 896MB of memory. The GTX260 is very similar to the 8800GTS 640 that was featured with the Geforce 8800GTX launch. The Geforce GTX280 has 141GB of bandwidth while the Geforce GTX260 carries a more conservative 112GB of bandwidth.

Dual Issue Shader

With the Geforce 8800 and Geforce 9800 series, each shader was capable of performing 1 MADD per clock. While under certain circumstances it could perform a "Mul", most of the time this Mul function went unexposed. With the GTX200 series, the Mad + Mul functions for each unit is fully exposed under all circumstance. This can lead to a performance improvement from the shaders clock per clock. The range seems to vary from 3-15% dependent on the type of shader being rendered, keeping mind that those numbers are just based on a few limited tests. Its difficult to quantify how much this will matter over a wide range of titles.

Geometry Shader

Internal Buffer structures have been upsized by 6x. This should provide a significant improvement to streamout performance.

Register Space and Double Precision

The register space on the G200 shader array has been doubled, which should allow for better performance in heavy shader utilization situations as well as open up support for Double Precision.

Video

The Geforce GTX 260/280 cards carry forward the same PureVideo HD functions introduced with the G92 architecture, and include a special power saving mode for DVD video playback.

The GTX280 requires 1ea 6 pin connector and 1ea 8 pin connector. Using an adaptor is not recommended unless you can ensure your PSU molex is capable of delivering 150 watts. The GTX260 requires 2ea 6 pin connectors similar to prior high end cards. The GTX280 consumes a maximum of 240 Watts while the GTX260 consumes 192 Watts of Power at full load.

The GTX 200 series features advanced power saving tech allowing the cards to power down in 2D/Desktop mode. When running idle modes, both clocks and vcore are significantly throttled, allowing for significantly lower power consumption while idling than previous high end Nvidia hardware.

2D Clocks

GTX 280 2D Clocks

GTX 260 2D Clocks

These idle power strides are significant for pretty much any gamer who uses there computer for other tasks besides gaming. High End SLI users will really appreciate this as 3 way SLI setup will only consume 75 watts at idle verses the 140-200 on other 3 way SLI setups. That's a significant amount of heat NOT dumped into your room environment. The GTX 200 series also supports Hybrid Power like the 9800GX2 and 9800GTX cards which can completely power down the GPU with a supporting Nvidia chipset.

Note: The GTX260 used in this review was an engineering sample. Newer models of the cards contain a bios with a slightly faster fan RPM and thus will effect idle/load temperatures.

CUDA, What does it mean for me?

CUDA applications remain sparse, making it difficult to discuss CUDA in depth. We've been hearing about GPU computing for awhile, with CUDA being a topic of conversation since the Geforce 8800GTX was first announced. Since it didn't impact gaming in any real fashion, CUDA was remained mostly a yawner. Things are changing now however, with dozens of apps now supporting CUDA. While few of them will likely impact Joe Gamer, CUDA potentially provides an alternative to Super Computing Solutions.

Most of these apps probably aren't going to be of much interest to the gamer or most end users. One notable app was the ability to encode 1080 Videos to IPod video format at 18x the speed of a Quad Core CPU. Nvidia demoed encoding a 2 hour movie in 38 minutes. The most notable thing CUDA will probably bring is ...

GTX260 Folding@Home

GTX280 Folding@Home

Folding@Home: CUDA Client

At long last there's a Folding@Home client for Nvidia hardware. Testing of this new beta client is still in its infancy, but our early tests show the GTX260 averaging about 450 NS while GTX280 averages about 510 NS. Both took about 26-32 minutes to complete a project. Once the new CUDA client emerges from closed beta, all DX10 Geforce owners will have the opportunity to participate in the Folding@Home. This has been a long time coming and a great step for showing the potential of CUDA.

PhysX

Nvidia's recent buyout of Ageia puts them in an interesting position. Subsequent to the buyout, the Ageia API has been converted over to CUDA in a matter of months, versus the year and half Ageia took to write it all in Assembly. All Geforce cards since the 8 series include PhysX support.  While no PhysX supported games are currently available, Nvidia demos presented many PhysX enabled software that should hit the scene in the coming year. Some of it looked very interesting and some impacted game play in such manner that the game just wouldn't have been the same without it. The PhysX API is compatible with XBox360, PS3, and the PC and it looks to ultimately have a positive influence on the gaming industry.

Final Thought

Both CUDA and PhysX have the potential to make a GPU purchase more than just a GPU purchase. It's one of those things that we will watch closely these upcoming months, because they potential could change the way we perceive graphic cards.

GTX 280 SLI

GTX260 3Way SLI

Crysis Testing Medium

Crysis Medium

Performance Thoughts: There is nothing particularly of interest here. The GTX260 and GTX280 both pull ahead of the 9800GX2. Crysis is not particularly the best scaler for SLI as the 9800GX2 doesn't perform very well here. Unfortunately Crysis also seems heavily CPU dependent at these settings and SLI reached a hard CPU bottleneck. It should be noted that this game's performance with SLI will scale better as clock speeds increase on your CPU.

Crysis Testing High

Crysis High

Performance Thoughts: These settings tell a slightly different story. The 9800GX2 and 9800GTX solutions fall behind both the GTX260 and GTX280 rather significantly, most likely due to the aforementioned running out of video memory due to their 512MB memory limitation. At these settings, with the memory limitations, significant stuttering was observed from the 9800GTX and 9800GX2 configurations. As with Medium settings, Crysis SLI scaling can be very CPU dependent; raising your clocks will increase the relative performance of your Crysis performance. GTX260 SLI and GTX280 and GTX280 SLI provide most the consistent experience, while GTX260 3 Way SLI provides very little benefit in Crysis

Bioshock Testing

Bioshock

Performance Thoughts: The GTX260 and GTX280 do exceptionally well in 16xAA and 16xQ settings; the superior bandwidth of the GTX280 allows it to pull significantly ahead of the 9800GX2 with 16xQ enabled, while the GTX260 is able to match the 9800GX2 in raw FPS value. The super dual issue shader capabilities may also be playing a role in the performance here. SLI scaling also couldn't be better with the GTX 260 3 way SLI setup holding a significant lead over the other solutions.

Unreal Tournament 3 Performance

Unreal Tournament 3

Performance Thoughts: Using the Deimos map we can see decent multi GPU scaling, though no where near as impressive as in Bioshock. Regardless of scaling, the GTX280 is able to pull ahead of the 9800GX2 by a few percentage points. Dual GTX280s in SLI are also able to maintain their lead over the Quad 9800GX2 setup while GTX260 SLI drags a little behind. Once 3 way SLI is enabled, the GTX260 pulls ahead of all configurations tested. Due to the constant crashing of the 512MB cards with 16xQ enabled, 16xQ results were excluded from this comparison.

Lost Planet Results

Lost Planet

Performance Thoughts: First the demo did not like the HDTV used for testing, so a standard 1680x1050 widescreen resolution was used. To increase GPU load 16xQ was selected. Also, unlike prior tests, these results were not recorded with fraps but rather the built in benchmarking utility. The Lost Planet benchmark has built in multi GPU support and scales exceedingly well with SLI configuration. The GTX280 is once again able to pull ahead of the 9800GX2 cards while the GTX260 stays a little behind. While in SLI mode, the GTX280 holds the same lead over Quad SLI. The GTX260 comes up behind the 9800GX2 configuration in both single and SLI mode, but pulls ahead of the Quad 9800GX2 and GTX280 SLI setup with 3 way SLI enabled.

FEAR Results

FEAR

Performance Thoughts: We were unable to provide 16xQ results because, like UT3, the game kept crashing on the 512MB solutions when enabled. Nonetheless, we set maximum details with soft shadows enabled for this testing. The GTX260 and GTX280 follow similar trends to Lost Planet and Bioshock as overall performance goes. Nothing particularly interesting or out of the ordinary all things considering. The GTX260 3 way solution once again provides the highest framerates.

Neverwinter Nights 2

Performance Thoughts: Neverwinter Nights also shunned the HDTV resolution, so 1680x1050 was the resolution of choice. Also limiting were the in-game AA settings, limited to only 8x Multisampling. With this title, two interesting things can be noted. First, it is largely shader bound title where normal maps are applied to every surface in game. Next, the game seems to suffer memory limitations at maximum settings, particularly with high shadows enabled. This allows the GTX260 and GTX280 SLI solutions to pull above the Quad 9800GX2 in performance. The GTX260 3 way SLI also shows strong signs of the game becoming CPU limited with so much GPU power.

Age of Conan

Performance Thoughts: Age of Conan is the latest MMORPG on the market, making use of multi core SM 3.0 and HDR. Future revisions are supposed to support DX10 but these tests were included with the launch DX 9.0 renderer. This title shows similar performance to Unreal Tournament 3. Though the performance spread between cards is small. The 9800GX2, GTX260 and GTX280 are able to provide playable performance with a single card. Truly comfortable performance only comes from the SLI solutions, with the GTX280 pulling ever so slightly ahead of Quad 9800GX2. Like previous tests the 3 way GTX260 SLI solution pulls ahead of the other SLI solutions.

Note: 16xQ results were not included due to the 512 meg cards being unable to perform with texture errors and crashes.

With the plethora of configurations were tested here, with multiple levels of SLI, we will conclude by discussing them at their various price and performance levels.

GTX280 Comparisons

GTX280

GTX280 vs. 9800GX2: No doubt about it, and for multiple reasons, the GTX280 is the better card card here. Not only does it provide 1GB of pure framebuffer memory for the card to access, it also consistently outperforms the 9800GX2 without the need for AFR rendering. This is important for those sensitive to AFR latency issues between inter-frame delays. When the geometry shader becomes more of an issue, the GTX280 will have a significant advantage as well. The GTX280 also does not suffer from artificial performance barriers such as 512MB of memory hamstringing it from achieving its full potential. Finally, the GTX280 consumes significantly less power while idling. The downside of the GTX280 vs the 9800GX2 includes higher temps and increased power consumption at load. In summary, this is the card for those wanting the single most powerful GPU Nvidia has to offer. With a price tag of $650US, it is not a cheap investment.

GTX280 SLI:  The GTX280 SLI verses Quad 9800GX2 follows the same storyline. Both solutions use AFR and profiles to provide performance boosts. However, with the power behind the GTX280 SLI, it never find its rendering performance hamstringed by the 512MB limit imposed upon the 9800GX2. This allows GTX280 SLI to pull far ahead of the 9800GX2 with 16xQ and higher levels of AA/resolutions enabled. The 9800GX2 also occasionally interferes with LED/PWR Connectors on the bottom slot, which the GTX280 cooler is designed to avoid. The $1300 dollar asking price, however, introduces a new competitor: GTX260 3 way SLI, Which we'll cover in a moment.

GTX280 3 way SLI: We were unable to acquire a third GTX280 for this preview, so are unable to provide any meaningful feedback on GTX280 3 way SLI.

GTX260 Comparisons

GTX260

GTX260 vs. 9800GX2: The GTX260 finds itself in an interesting position. While it usually falls slightly behind the 9800GX2 in raw FPS, one must not forget that the 9800GX2 relies on profiles and AFR for its performance. When you take these things into account, it becomes difficult to recommended the 9800GX2 over a single GTX260. Other advantages offered by the GTX260 include  being equipped with the more reasonable 2ea 6 pin connector configuration as well as the reduced power requirements. While short of the GTX280's 1GB configuration, the GTX260's 896MB of memory is far above the 512MB limitation suffered by the 9800GX2, which will allow it to pull ahead in certain situations, such as Crysis, at High Settings. The GTX260 retails for $400US, which further puts the 9800GX2 in a difficult place. With all these factors in mind, we will not be at all surprised to see the 9800GX2 quickly phased out.

GTX260 vs. 9800GX2 Quad SLI: The recommendation in this case echoes the last. With only requiring 6 pin connectors, the GTX260 solution will be open to more power supplies. Like the 9800GX2 Quad solution, it does rely on AFR rendering and SLI profiles. While the GTX260 is a little slower in some situations, the additional power features and shader capabilities should make it a more future proof solution. As with the GTX280 cooler, you will never have to worry about the second bottom card hitting the LED/PWR connectors that the 9800GX2 would tend to squish up against.

GTX260 3 way SLI vs. GTX280 SLI: This choice is quite a bit more interesting. At $400US versus $650US for the GTX280, a GTX260 3way SLI configuration will cost you a net total of $1200US, providing you superior performance in most situations to a $1300US GTX280 SLI setup. The GTX260 2ea 6 pin configuration also comes in to play again here, competing against the 8 pin requirement of the GTX280.  The downsides, of course, include the GTX260 3 way SLI's higher power consumption at both idle and load, making chassis airflow a larger factor, as well as the requirement of a 3 way capable SLI motherboard.

Overall Thoughts and Conclusions

The GT200 is an interesting architecture. This 1.4 Billion transistor beast provides us with a single GPU capable of delivering performance beyond the previous generation's multi GPU solutions. That alone is an impressive feat and should not be ignored. However, other than performance enhancement features and improved power efficiency, there isn't a whole lot of "new" to the GTX 200 series over prior DX 10 Geforce cards. 9800GX2 users are unlikely to run out and replace their cards immediately, unless they are aiming for a 3 way SLI solution for highest level of performance. And, of course, more performance isn't a bad thing. As newer titles come out and start really stressing the current cards, the GTX200 series cards are really going to help keep the frame rates and resolution up.

As we have seen, there are other considerations to look at when looking at an Nvidia card purchase, considerations not limited to the GTX200 series. Nvidia is clamoring hard about the important of PhysX and CUDA, and there are some interesting apps taking advantage of CUDA now. With PhysX its slightly different. While no current games currently support PhysX, there are several upcoming A titles and B titles with PhysX support built in. Nvidia demoed several of these games at editors day. If Nvidia can deliver upon the PhysX promises they are making, gaming on a DX10 Geforce card could become quite interesting in the future.

Non GPU Related CPU Note

Nvidia has been making a big deal about the "optimized" PC lately. Stating that you should buy a PC with a good GPU AND good CPU for optimal performance. In this case I have to give a small nod to AMD. While the Phenom isn't delivering the fastest CPU experience, it was more than capable of providing an excellent gaming experience even on a 3 way GTX260 solution. While this may not be the most "Balanced" approach. It does lend some credibility the idea that a decent CPU paired up with powerful graphic cards can provide an excellent gaming solution. While some may question the choice of using a Phenom in this investigation I felt it gave an interesting perspective.