Desktop System Video Card Hardware Guide - September 2008

Desktop System Video Card Hardware Guide - September 2008 

Video cards are produced by many manufacturers, but the chips controlling the cards, known as a GPU (Graphical Processing Unit), are manufactured by two main companies, NVidia and ATI.  These companies produce the chips, and tell the video card manufactures how to connect and use the chips into their own boards.  This means that the same GPU used on two separate cards can have noticeably different performance based on how the manufacturer configures it with memory or bus speeds.

Also, both NVidia and ATI follow a numbering scheme that lets consumers know which cards are more powerful.  The use of suffices for the ATI cards have been dropped, and instead all cards now have a number.  The higher the number, usually the higher the performance and more features.

*Pricing is based upon average AVADirect's pricing as of 9/22/08

Terms and Technologies

SLI (Scalable Link Interface):	NVidia's technology to link together two or more NVidia GPU based video cards and work in parallel.
CrossFireX:	ATI's technology to link together two or more ATI GPU based video cards and work in parallel.
Memory:	Video cards come equipped with all sorts of different memory technologies and sizes.  Usually, the higher the memory, the bigger the screen size that the video card is able to run with slowing down.  Most common memory types are DDR2, GDDR3, GDDR4, and now GDDR5.    Note on graphics memory.  GDDR and DDR are two separate types of memory standards.  While normal ram modules can be used, GDDR is designed to be better suited to type of data it will see.  Also, GDDR3 is about equal with DDR2 in terms of performance and speed.  Following that, DDR3 is about equal with GDDR4 memory.  It was designed to better handle the graphical data of the video card while still being based upon the same technology.
Unified Shader Architecture:	Instead of using different components to calculate 3D positions, pixels, textures, and etc., new video cards now use dozens to hundreds of Stream Processors instead.  These Stream Processors are simple math processors which run in parallel to achieve high data processing output with simple processing techniques.
Stream Processor or Shader:	A stream processor is a floating point math processor which performs vector and image processing on 3D objects.

NVidia GeForce Video Cards

GeForce 8 Series

 

The GeForce 8 series was a major overhaul of the GeForce architecture.  Instead of using separate functional units to do processing on pixels and vectors, Stream Processors are a unified type of functional unit that work on a more universal style of tasks.  Although not as powerful as specialized units, their extreme simplicity and much higher clock speeds allow a higher performance rating at a lower cost.

 

This series also saw the introduction and acceptance of DDR3 memory.  DDR3 memory allows higher bandwidth, which some of the higher-end cards need to operate effectively.  The GeForce 8 series also saw the introduction of DirectX 10 support.

 

Due to the aging structure of the GeForce 8 line, only the most current and non-discontinued video cards will be shown.

 

8800 GS - G92 Core                            Expected Performance:  8,000 - 11,000 3DMarks   Core Frequency: 550 MHz Shader Amount: 96 Memory Amount: 384 MB Shader Frequency: 1375 MHz Texture Units: 48 Memory Bus Width: 192-bit Memory Frequency: 1600 MHz ROPS: 12 Memory Type: GDDR3   With NVidia focusing on more powerful cards, and the disappointing performance of the 8600 line, NVidia introduced the 8800 GS to compete with the ATI Radeon cards that were dominating the mid-range market.  It has lowered speeds so that manufacturing costs are lower, but because it is based on the G92 GPU, its performance matches or beats the original 8800 GTS.

 

8800 GT - G92 Core                            Expected Performance:  9,000 - 12,000 3DMarks   Core Frequency: 600 MHz Shader Amount: 112 Memory Amount: 512 MB Shader Frequency: 1500 MHz Texture Units: 56 Memory Bus Width: 256-bit Memory Frequency: 1800 MHz ROPS: 16 Memory Type: GDDR3   This is the first 8800 series card to use the 65 nm technology.  This in turn allows the usage of a single slot cooler, and lowers the operating power.  It was also shown to beat the GTS (G80) and come very close to the GTX in terms of benchmarked performance.  This was a radically new departure as the GT was smaller, cooler, and half the price of a GTX.

 

8800 GTS - G92 Core                            Expected Performance:  10,000 - 13,000 3DMarks   Core Frequency: 650 MHz Shader Amount: 128 Memory Amount: 512 MB Shader Frequency: 1625 MHz Texture Units: 64 Memory Bus Width: 256-bit Memory Frequency: 1940 MHz ROPS: 16 Memory Type: GDDR3   A re-release of the original GTS, it is now using the same 65 nm core as the 8800 GT, except it is clocked faster.  It now offers better performance than either the 8800 GT or the 8800 GTX, and it also cost less than the 8800 GTX.  However, it had less memory than the GTX, so higher resolutions will affect the performance.

GeForce 9 Series

 

In keeping with the NVidia methodology, NVidia created the 9-series as a successor to the 8 series.  The 8 series had reach a limit with its current architecture, with only more ram and higher clock speeds offering more performance at too high a price.

 

9500 GT - G96 Core (55 nm)                            Expected Performance:  4,000 - 6,000 3DMarks   Core Frequency: 550 MHz Shader Amount: 32 Memory Amount: 256, 512 MB Shader Frequency: 1350 MHz Texture Units: 32 Memory Bus Width: 128-bit Memory Frequency: 1600 MHz ROPS: 8 Memory Type: GDDR3   NVidia created a low-powered core for the mainstream usage with this video card.  Very close in performance to the older 8600 GT/GTS video cards, but offers slightly better performance and lower heat/power due to 55 nm fabrication.

 

9600 GT - G94 Core                            Expected Performance:  7,000 - 10,000 3DMarks   Core Frequency: 650 MHz Shader Amount: 64 Memory Amount: 512 MB Shader Frequency: 1625 MHz Texture Units: 32 Memory Bus Width: 256-bit Memory Frequency: 1800 MHz ROPS: 16 Memory Type: GDDR3   The first of the 9-series cards, it offers twice the speed of the 8600 GT card, and is just 15% slower than the new 8800 GT.  It is a great buy for the value-gamer, as it is able to double its performance using SLI. This card has been re-released using the 55 nm core.  It now uses the G94b core instead of the G94a core.

 

9600 GSO - G92-150 Core                            Expected Performance:  7,000 - 10,000 3DMarks   Core Frequency: 550 MHz Shader Amount: 96 Memory Amount: 384 MB Shader Frequency: 1375 MHz Texture Units: 48 Memory Bus Width: 192-bit Memory Frequency: 1600 MHz ROPS: 12 Memory Type: GDDR3   Instead of this being a new 9-series card different from the 8-series, this card has exactly the same specs and performance as the 8800 GS.  However, it was re-released at a lower price point then the current 8800 GS's are currently selling for.

 

9800 GT - G92b Core (55 nm)                            Expected Performance:  9,000 - 12,000 3DMarks   Core Frequency: 600 MHz Shader Amount: 112 Memory Amount: 512 MB Shader Frequency: 1500 MHz Texture Units: 56 Memory Bus Width: 256-bit Memory Frequency: 1800 MHz ROPS: 16 Memory Type: GDDR3   Die shrink of the 8800 GT to 55nm, but otherwise then that, all the specs and performance are the same as the 8800 GT.

 

9800 GTX - G92 Core                        Expected Performance:  10,000 - 13,000 3DMarks            Triple SLI Capable   Core Frequency: 675 MHz Shader Amount: 128 Memory Amount: 512 MB Shader Frequency: 1688 MHz Texture Units: 64 Memory Bus Width: 256-bit Memory Frequency: 2200 MHz ROPS: 16 Memory Type: GDDR3   Like the 8800 GTX before it, this card is capable of triple SLI.  Unlike the 8800 GTX, this card is based upon the improved G92 core which uses less power to deliver better performance.

 

9800 GTX+ - G92b Core (55 nm)              Expected Performance:  11,000 - 13,000 3DMarks         Triple SLI Capable   Core Frequency: 738 MHz Shader Amount: 128 Memory Amount: 512 MB Shader Frequency: 1836 MHz Texture Units: 64 Memory Bus Width: 256-bit Memory Frequency: 2200 MHz ROPS: 16 Memory Type: GDDR3   This is the most recent 9 series card to date as it was produced to directly compete with the new video cards from ATI in June.  It uses the same GPU as the 9800 GTX, but has been fabricated at 55 nm unlike the 9800 GTX's 65 nm fabrication.  It also has higher clocked components to squeeze the most performance possible from the G92 chip.

 

9800 GX2 - G92 Core                        Expected Performance:  12,000 - 16,000 3DMarks               Quad SLI Capable   Core Frequency: 600 MHz Shader Amount: 2x 128 Memory Amount: 2x 512 MB Shader Frequency: 1500 MHz Texture Units: 2x 56 Memory Bus Width: 2x 256-bit Memory Frequency: 2000 MHz ROPS: 2x 16 Memory Type: GDDR3   The 9800 GX2 is a single slot dual-PCB graphics card.  It features two G92 based chips each on its own card, internally linked by SLI.  Performance had to be somewhat downclocked to keep heat down and reliability up.  Benchmarks have shown this card to be superior to the ATI Radeon X2, the card that took the title for most powerful card away from NVidia for a few weeks.\

 

GeForce 200 Series

 

These video cards were in response to ATI who were fast approaching the performance of the 8000 and 9000 series.  NVidia did not invent anything new by introducing the 9000 series except shrinking the die size and making them more efficient and less expensive by shrinking the memory bus and memory amounts.  The GT200 series so far has seen doubled amounts of shaders, bandwidth and memory to make NVidia once again king of the single GPU performance benchmark.

 

GTX 260 - GT200 Core                      Expected Performance:  12,000 - 14,000 3DMarks               Triple SLI Capable   Core Frequency: 576 MHz Shader Amount: 192 Memory Amount: 896 MB Shader Frequency: 1242 MHz Texture Units: 64 Memory Bus Width: 448-bit Memory Frequency: 1998 MHz ROPS: 28 Memory Type: GDDR3   Both the 260 and 280 use the identical core processor, but has cut down hardware features which saves on costs.  It has less memory, memory bus width, and component frequencies than the 280.  It however has a much better performance per value ratio than the 280 which exceeds all performance barriers with cost being a second objective.

 

GTX 260 216 - GT200 Core                 Expected Performance:  12,500 - 14,500 3DMarks         Triple SLI Capable   Core Frequency: 576 MHz Shader Amount: 216 Memory Amount: 896 MB Shader Frequency: 1242 MHz Texture Units: 72 Memory Bus Width: 448-bit Memory Frequency: 1998 MHz ROPS: 28 Memory Type: GDDR3   Due to better yields in the processing, the GTX 260 216 is a GTX 260 with less components de-activated compared to the normal GTX 260.  This has just been released, and currently either equal to or slightly above in price compared to the GTX 260.

 

GTX 280 - GT200 Core                       Expected Performance:  14,000 - 16,000 3DMarks             Triple SLI Capable   Core Frequency: 602 Shader Amount: 240 Memory Amount: 1 GB Shader Frequency: 1296 MHz Texture Units: 80 Memory Bus Width: 512-bit Memory Frequency: 2214 MHz ROPS: 32 Memory Type: GDDR3   The GTX 280 put performance first and value second just like the 8800 Ultra 2 years prior.  It features more processing power than any other single GPU solution on the consumer market.  Unless NVidia decides to make a dual-GPU based on this core, this will be the most powerful video card in the 200 series.

 

ATI Radeon Video Cards

****** Note on CrossFire for ATI video cards ********

 

Unlike the NVidia cards which rely on the video card for more than dual SLI, CrossFireX is capable of using up to four GPUs based on the chipset.  That means if the chipset supports four cards in CrossFire, then any current generation ATI card can be used in that CrossFire combination.  Currently, the Spider platform by AMD allows the usage of four GPUs as long as there is four PCI-E slots available.

 

Radeon 2000/3000 Series

 

While the 2000 series featured the first use of the Unified Shader Architecture, the 3000 series has replaced the 2000 series with a smaller die shrink from 65 nm and 80 nm to 55 nm fabrication.  Somewhat like the NVidia naming convention, ATI's GPUs go up in performance with increases in numbers.  However, ATI does not follow NVidia by making each video card its own number.  ATI has done away with suffices and prefixes for the most part.  However ATI's partners who produce the cards may offer several variations of the same card and may release them with different prefixes and suffices.

 

HD 2600 Pro - RV630 Core                             Expected Performance:  4,000 - 6,000 3DMarks             Core Frequency: 600 MHz Shader Amount: 120 Memory Amount: 256 MB Shader Frequency: 600 MHz Texture Units: 8 Memory Bus Width: 128-bit Memory Frequency: 1400 MHz ROPS: 4 Memory Type: GDDR3   Mainstream product which is aimed for only mid-level performance in games.  Is about equal in performance with the 8600 GTS.

 

HD 3450 - RV620 Core                             Expected Performance:  4,000 - 6,000 3DMarks             Core Frequency: 600 MHz Shader Amount: 40 Memory Amount: 256 MB Shader Frequency: 600 MHz Texture Units: 4 Memory Bus Width: 64-bit Memory Frequency: 1000 MHz ROPS: 4 Memory Type: GDDR3   Video card which is meant to replace the onboard graphic solution on the motherboard.  Its low amount of processing power makes it not able to have high framerates on all but the basic 3D application or game.

 

HD 3850 - RV670 Core                             Expected Performance:  8,000 - 10,000 3DMarks             Core Frequency: 666 MHz Shader Amount: 320 Memory Amount: 512 MB Shader Frequency: 666 MHz Texture Units: 16 Memory Bus Width: 256-bit Memory Frequency: 1656 MHz ROPS: 16 Memory Type: GDDR3   The 3850 along with the 3870 were close in performance to NVidia's 8800 GT.  They were also very close in price points as well.

 

HD 3870 - RV670 Core                              Expected Performance:  9,000 - 11,000 3DMarks             Core Frequency: 825 MHz Shader Amount: 320 Memory Amount: 512 Shader Frequency: 825 MHz Texture Units: 16 Memory Bus Width: 256-bit Memory Frequency: 1800 MHz ROPS: 16 Memory Type: GDDR3 / GDDR4   The 3870 is a higher clocked version of the 3850.  Everything else like number of components stayed the same.  Also this card first saw the introduction of GDDR4 memory.

 

HD 3870 X2 - RV670 Core                              Expected Performance:  12,000 - 14,000 3DMarks                       Core Frequency: 825 MHz Shader Amount: 2x 320 Memory Amount: 2x 512 MB Shader Frequency: 825 MHz Texture Units: 2x 16 Memory Bus Width: 2x 256-bit Memory Frequency: 900 MHz ROPS: 2x 16 Memory Type: GDDR3 / GDDR4   The Radeon 3870 X2 is a dual-GPU card solution.  It uses software CrossFire to link them together, sharing the PCI-express connection to communicate.  It was the first single card to break the 1 TFLOP barrier, something that NVidia could not do.  It is also possible to connect these cards in hardware CrossFire, and use up to four GPUs simultaneously.  This also stole the title of most powerful card away from the NVidia 8800 Ultra, but lost it within a few weeks of the 9800 GX2 coming out.

 

Radeon 4000 Series

 

ATI released the 4000 series just a few days after the 200 series from NVidia.  ATI really has come back by offering very good mid-range performance at prices that compete with or beat NVidia.  While NVidia has decided to conquer the very high end of performance, these cards from ATI conquer the mid-range.  What is also good is that these cards scale very well in CrossFire, comparing directly in performance to the high-end from NVidia for the same amount or less money.

 

HD 4650                              Expected Performance:  5,000 - 8,000 3DMarks             Core Frequency: 600 MHz Shader Amount: 320 Memory Amount: 512 MB Shader Frequency: 600 MHz Texture Units: 32 Memory Bus Width: 128-bit Memory Frequency: 1000 MHz ROPS: 8 Memory Type: DDR2   Entry level card designed to replace the 2600 and 3450 cards.

 

HD 4670                              Expected Performance:  6,000 - 9,000 3DMarks             Core Frequency: 750 MHz Shader Amount: 320 Memory Amount: 512 MB Shader Frequency: 750 MHz Texture Units: 32 Memory Bus Width: 128-bit Memory Frequency: 2000 MHz ROPS: 8 Memory Type: GDDR3   Entry level card designed to replace the 2600 and 3450 cards.

 

HD 4850                              Expected Performance:  11,000 - 13,000 3DMarks             Core Frequency: 625 MHz Shader Amount: 800 Memory Amount: 512 MB Shader Frequency: 625 MHz Texture Units: 40 Memory Bus Width: 256-bit Memory Frequency: 1986 MHz ROPS: 16 Memory Type: GDDR3   The first card to be released under the 4000 series.  This card was highly anticipated by many people who were wanting an alternative to NVidia.  While at first only suppose to be a direct competitor to the 8800 GT, it has surpassed many people's initial thoughts and performed better than the 9800 GTX.  However, due to the single slot design, it runs very hot.  First single GPU card to have a possible performance rating of 1TFlops

 

HD 4870                              Expected Performance:  12,000 - 14,000 3DMarks             Core Frequency: 750 MHz Shader Amount: 800 Memory Amount: 512 MB Shader Frequency: 750 MHz Texture Units: 40 Memory Bus Width: 256-bit Memory Frequency: 3600 MHz ROPS: 16 Memory Type: GDDR5   This card was released a week later after the 4850.  This card was slow to roll out due to the scarcity of GDDR5 memory modules.  It features a higher clocked core and almost double the bandwidth of the 4850, it puts it in the same performance bracket as the 9800 GTX and GTX 260.  Two of these cards in CrossFire can easily overcome the GTX 280 with the right driver support.

 

HD 4870 X2                              Expected Performance:  16,000 - 18,000 3DMarks             Core Frequency: 750 MHz Shader Amount: 2x 800 Memory Amount: 2x 1 GB Shader Frequency: 750 MHz Texture Units: 2x 40 Memory Bus Width: 2x 256-bit Memory Frequency: 3600 MHz ROPS: 2x 16 Memory Type: GDDR5   Utilizing the same technology as the 3870 X2, the 4870 X2 incorporates dual 4870 cores onto a single PCB.  The two cores are linked using the PCIe switch using a pseudo CrossFire setup.  Currently the most powerful single video card available.

For More Information

For more detailed information into the specific differences and actual performance of the different video cards, please follow these links and articles.

 

AVADirect Review of the VisionTek 4850

 

AMD 4870 X2 Performance Numbers Leaked

******* Note on expected performance numbers! **********

Our expected performance numbers are in now way guaranteed, and are only to be uses as reference numbers when looking at this article.  Actual performance not only relies on the video card itself, but your computer system as a whole.

The expected numbers are based on a simulated system with the following components.

• 
  

  Intel Core 2 Quad processor @ 2.4 GHz

  
  
• 
  

  Motherboard with a single PCI-Express x16 slot

  
  
• 
  

  4 GB of DDR2 memory @ 800 MHz

  
  
• 
  

  Standard size hard drive @ 7200 RPM