When NVIDIA launched two 55nm solutions based on the overhauled GT200, AMD did not respond with its own cards, as it often happened, although rumors about some improved RV770 were already circulating the web. However, the company announced a new graphics card based on the improved RV7x0 GPU in early April. NVIDIA did not want to surrender the initiative, so it launched another model -- GeForce GTX 275. This graphics card is a direct competitor to the new solution from AMD. And today we are going to examine it.
As usual, NVIDIA specifies that GeForce GTX 275 uses the latest GPU, offers the best performance and functionality (PhysX, CUDA) for its price. No cheating here, but it's actually the 55nm GT200. It differs from GTX 260 and GTX 285 only in some characteristics: the number of active execution units and clock rates.
This product fit between GeForce GTX 260 and GeForce GTX 285 in terms of performance and price. As we have already mentioned, the new card has been designed to oppose RADEON HD 4890, the most powerful single-GPU card from AMD. This corresponds to the $260 price tag -- more expensive than GTX 260, but cheaper than GTX 285.
Note that some price ranges are very crowded now. There are a lot of solutions with similar characteristics and performance. Let's take NVIDIA for example: overhauled GeForce GTX 260 with 216 stream processors is not much slower than the GTX 275 even theoretically, and their results will be even closer in real games.
The GTX 275 should perform on a par with the discontinued GTX 280, and its overclocked modifications can come closer to the GTX 285. They cannot compete with the dual-GPU GTX 295, of course, it's in a different price range. But the $199-299 segment is already overcrowded in our opinion. Game tests will prove whether we are right or wrong.
If you are not familiar with the GeForce GTX 200 (GT200) architecture, you can read about it in our baseline review. This architecture was developed from G8x/G9x with some modifications. The detailed information about unified architectures of NVIDIA G8x/G9x/GT2xx can be found in the following articles:
We proceed from the assumption that you are already familiar with the GT200(b) architecture. Now we shall examine the new card from the GeForce GTX 200 series based on a 55nm GPU. The theoretical part will be very short again, because we have already examined the architecture and overhauled solutions based on GT200b.
GeForce GTX 275
- Codename: GT200b
- Process technology: 55 nm
- 1.4 billion transistors
- Unified architecture with an array of shared processors for streaming processing of vertices and pixels, as well as other data
- Hardware support for DirectX 10, including Shader Model 4.0, geometry generation, and stream output
- 448-bit memory bus, seven independent 64-bit controllers
- Core clock: 633 MHz
- Higher than doubled ALU frequency: 1404 MHz
- 240 scalar floating-point ALUs (integer and floating-point formats, support for FP32 and FP64 according to IEEE 754(R), two MAD+MUL per cycle)
- 80 texture address and filtering units, support for FP16 and FP32 components in textures
- Dynamic branching in pixel and vertex shaders
- 7 wide ROPs (28 pixels) supporting antialiasing with up to 16 samples per pixel, including FP16 or FP32 frame buffer. Each unit consists of an array of flexibly configurable ALUs and is responsible for Z generation and comparison, MSAA, blending. Peak performance of this subsystem is up to 112 MSAA samples (+ 112 Z) per cycle, in Z only mode - 224 samples per cycle
- Multiple render targets (up to 8 buffers)
- Interfaces (2 x RAMDAC, 2 x Dual Link DVI, HDMI, DisplayPort, HDTV) are integrated into a separate chip
GeForce GTX 275 reference specifications
- Core clock: 633 MHz
- Frequency of unified processors: 1404 MHz
- Unified processors: 240
- 80 texture units, 28 blending units
- Effective memory frequency: 2268 MHz (2*1134)
- Memory type: GDDR3
- Memory: 896 MB
- Memory bandwidth: 127 GB/s
- Maximum theoretical fill rate: 17.7 gigapixel per second
- Theoretical texture sampling rate: 50.6 gigatexel per second
- 2 x DVI-I Dual Link, 2560x1600 video output
- Double SLI connector
- PCI Express 2.0
- TV-Out, HDTV-Out, HDCP support, HDMI, DisplayPort
- Power consumption: up to 219 W (two 6-pin connectors)
- Dual-slot design
- Recommended price: $259
As GT200b chips are manufactured by the 55nm process technology, NVIDIA managed to replace the old GeForce GTX 260 and GTX 280 with new faster products. Along with improved performance, this solution consumes much less power than its predecessors.
As usual, we should take a look at memory volume and the name of this product. This amount of video memory is determined by the PCB from the overhauled GTX 260 card with a 448-bit memory bus. Thus we have 896 MB here. It's sufficient for the great majority of applications, and our tests in games will prove it once again.
The title is also simple. The new card from NVIDIA is a tad faster than GTX 260, so it was logically called GTX 275. This change in product number makes sense to users, it shows that the card stands in between GTX 260 and GTX 285. Figure five in the end stands for the 55nm GPU, and thus improved characteristics (power consumption and heat release). By the way, the overhauled GTX 260 with 216 SPs should have been logically called GTX 265.
Architecture and features
We are again forced to tell you that we have nothing interesting to add here, because GT200b is no different from GT200, it's just smaller and consumes less power. The improved GT200 architecture based on the 55nm process technology makes NVIDIA GeForce GTX 275 faster, quieter, and less power hungry than GTX 280, while their relative performance is very close to each other. That could be the end of our theoretical part, but we decided to describe the other peculiarities, GPU part being self-explanatory.
Graphics driver v.185
As it often happens with new cards, NVIDIA timed the rollout to coincide with the release of the new video driver (v.185). It's a major update since v.180 that includes performance improvements for single- and multi-GPU solutions as well as new functions.
Performance changes are provided by the following modifications: optimized texture memory management in Direct3D10 applications, higher efficiency of Z-culling and buffer compression for antialiasing, SLI scaling improvements, especially in games cased on the Half-Life 2 engine. The highest performance gains appear in high resolutions with antialiasing. Difference between Drivers v.182 and v.185 in these conditions amounts to 5-25% (it's even higher in Mirror's Edge with antialiasing), according to NVIDIA.
Along with many bug fixes (and new bugs, of course), there appeared support for CUDA 2.2 that added useful features for developers and accelerated GPU-assisted computations in some tasks, as well as added Ambient Occlusion to some 3D games.
Ambient Occlusion
Ambient Occlusion (AO) is a global lighting (shadowing) model used in 3D graphics, which makes scenes look more realistic by calculating light intensity on object surfaces. Brightness value of each point on the surface depends on mutual positions of other objects in a scene. More simply, Ambient Occlusion calculates how much a given surface is blocked from light sources by other objects, and how much dissipated light it gets.
The traditional light model calculates a surface color solely from its characteristics and light source properties. Objects in the light throw shadows, but they do not affect lighting conditions of the other objects in a scene. AO improves the traditional light model by calculating shadows from objects standing in the light. Here is an example:
These fragments show the effect from forced AO in Half-Life 2. The usual render mode with 'flat' lighting is on the left, Ambient Occlusion is on the right. When AO is enabled, angles between surfaces are shaded, as they get less light because of other objects. As a result, we get a more realistic image, deep and tridimensional.
AO can be enabled in driver options. This feature is supported by application profiles in NVIDIA Control Panel, as each game needs individual adjustments for the algorithm. As for now, AO is supported in 22 popular applications: Assassin's Creed, BioShock, Call of Duty 4, Call of Duty 5, Call Of Juarez, Company of Heroes, Counter-Strike Source, Dead Space, Devil May Cry 4, F.E.A.R. 2, Fallout 3, Far Cry 2, Half Life 2 (games from this series), Left 4 Dead, Lost Planet: Colonies and Extreme Condition, Mirror's Edge, Portal, Team Fortress 2, Unreal Tournament 3, World In Conflict and World of Warcraft.
The effect from AO is best noticeable in indoor scenes in the following games: Counter-Strike Source, Half Life 2, and Mirror's Edge. There is also a noticeable AO effect for grass and other vegetation in the open spaces in Company Of Heroes, World In Conflict, and World of Warcraft. The most technically advanced games, such as Crysis and STALKER: Clear Sky, initially support AO, it in their code. So they don't need to enable this feature in the drivers.
It must be noted that AO does not make a big visual difference in many cases, but it may significantly reduce overall performance, as it's a compute-intensive algorithm. Performance drops from enabled AO may reach 20-50% depending on a given application, its graphics settings, and resolution. So it's up to you to decide whether you need this minor render quality enhancement or not.
Write a comment below. No registration needed!
