Page 2:The GeForce 200 GTX
Page 3:Arithmetic Power (Tests)
Page 5:What about Direct3D 10.1?
Page 6:The Architecture in Detail
Page 7:Scalable Processor Array
Page 8:Reworked Streaming Multiprocessors
Page 10:Fillrate Tester Results
Page 11:Vertex/Pixel Shading Performance
Page 12:Specifications: Better!
Page 13:GTX 280 or GTX 260?
Page 14:The Test
Page 15:Flight Simulator
Page 16:Call of Duty 4
Page 17:Test Drive Unlimited
Page 19:World in Conflict
Page 20:Supreme Commander
Page 21:Unreal Tournament 3
Page 22:Mass Effect
Page 23:Race Driver: GRID
Page 24:BadaBOOM Media Converter, Folding@Home
Page 27:Temperatures, Overclocking
Page 28:Bottom Line
Page 29:Performance Recap
As is often the case, Nvidia is making two cards available for its launch – a very-high-end version, the GeForce GTX 280, and a slightly more affordable but still high-end card, the GeForce GTX 260. Let’s look at the other characteristics of these cards compared to their competitors.
|GPU||HD 3870 X2||9800 GX2||8800 Ultra||GTX 260||GTX 280|
|GPU frequency||825 MHz||600 MHz||612 MHz||576 MHz||602 MHz|
|ALU frequency||825 MHz||1500 MHz||1512 MHz||1242 MHz||1296 MHz|
|Memory frequency||900 MHz||1000 MHz||1080 MHz||999 MHz||1107 MHz|
|Memory bus width||2x256 bits||2x256 bits||384 bits||448 bits||512 bits|
|Memory quantity||2 x 512 MB||2x512 MB||768 MB||896 MB||1024 MB|
|Number of ALUs||640||256||128||192||240|
|Number of texture units||32||128||32||64||80|
|Number of ROPs||32||32||24||28||32|
|Shading power||1 TFlops||(1152) GFlops||(581) GFlops||715 GFlops||933 GFlops|
|Memory bandwidth||115.2 GB/s||128 GB/s||103.7 GB/s||111.9 GB/s||141.7 GB/s|
|Number of transistors||1334 million||1010 million||754 million||1400 million||1400 million|
|Die surface area||2 x 196 mm²||2 x 324 mm²||484 mm²||576 mm²||576 mm²|
|Shader Model supported||4.1||4.0||4.0||4.0||4.0|
With 1,400 million transistors, and especially at 576 mm², Nvidia has created another monster – the biggest GPU ever produced, breaking the record, impressive as it was, of the G80 (16% smaller)! Yet this value ought to have remained constant over the generations (it has even been diminishing in recent times for “general-public” CPUs). Clearly production of the GT200 is extremely expensive for Nvidia, even if the conservative engraving depth is what explains the existence of a chip like this today.
Another notable point is that by continuing to use GDDR3, Nvidia is now not one but two generations behind in this department, since GDDR5 will make its appearance with the soon-to-be-released Radeon HD 4870. But though that is worth mentioning, thanks to the 512-bit bus, the increase of memory bandwidth is still 64% compared to the 8800 GTX’s 86.4 GB/s. And we’re finally seeing the debut of a high-end card with more than 512 MB of usable memory (not counting the relatively old and not widely distributed 8800 Ultra)! With 1 GB (and 896 MB for the GTX 260 – a good thing), performance at 2560*1600 resolution should finally be able to hold up!
Finally, the frequencies used are fairly conservative, especially with the ALUs, which are slower than on the 8800 Ultra, among others.
- The GeForce 200 GTX
- Arithmetic Power (Tests)
- What about Direct3D 10.1?
- The Architecture in Detail
- Scalable Processor Array
- Reworked Streaming Multiprocessors
- Fillrate Tester Results
- Vertex/Pixel Shading Performance
- Specifications: Better!
- GTX 280 or GTX 260?
- The Test
- Flight Simulator
- Call of Duty 4
- Test Drive Unlimited
- World in Conflict
- Supreme Commander
- Unreal Tournament 3
- Mass Effect
- Race Driver: GRID
- BadaBOOM Media Converter, Folding@Home
- Temperatures, Overclocking
- Bottom Line
- Performance Recap