Fermi Arrives - NVIDIA GeForce GTX 480

Introducing the GF100 (part 2)

Other Features of the GF100 GPU

Apart from adding extra CUDA cores, NVIDIA has also made improvements to other aspects of the chip and added 3D Vision Surround functionality. Here are the highlights:

GPC Architecture

One of the main features that the DirectX 11 API brings about is tessellation, which increases the number of polygons visible in real time dynamically by means of complex algorithms. This results in greater detail and translates to a higher level of realism.

However, the use of tessellation changes the GPU's workload balance and to counter it, NVIDIA has to reorganize the GF100's internal architecture a little and also introduced a new geometry engine.

Firstly, graphics Processing Clusters (GPCs) are the GF100's dominant high-level hardware block. Each GPC is made up of four SMs sharing a single raster engine, and handles all vertex, geometry, raster, texture and pixel processing operations.

A new geometry engine called the Polymorph Engine was also introduced. Each SM has its own Polymorph engine, which in turn has its own dedicated vertex fetch unit and tessellator. Hence, the GF100 has 15 Polymorph engines alogether, which NVIDIA claims will enable breakthrough tessellation performance.

Texture Units

One way the GF100 is faster is that each SM has now got four dedicated texture units and a dedicated texture cache. Previously in the GT200, three SMs shared a single texture engine, which in turn has eight texture filtering units. This makes the GF100 is more efficient, and on top of that, NVIDIA also says that the internal architecture of the texture units has been enhanced.

And as we've have mentioned earlier, the dedicated L1 texture cache is now more flexible and the large 768KB L2 cache means that the maximum cache size for textures is now three times greater than the GT200. This will benefit games which are running texture-heavy shaders.

Raster Operating Units

We know that enabling anti-aliasing improves graphics quality, but that comes at the price of lower performance. With the new GF100, NVIDIA aims to improve anti-aliasing performance by reducing the performance hit that affect systems when gamers turn on anti-aliasing in their games.

To achieve that, NVIDIA has redesigned the GF100's ROP (raster operating partitions) subsystem for greater throughput. The GF100's ROP partition contains eight individual ROP units, which is twice as much as previous architectures. These units can output a 32-bit integer pixel per clock; an FP16 pixel over two clocks or an FP32 pixel over four clocks. NVIDIA states that the GF100 chip is geared specifically to tackle 8x MSAA, which means enabling anti-aliasing on your games will result in a lesser hit in performance.

NVIDIA 3D Vision Surround

Enhanced texture and raster operating units aside, perhaps the most exciting feature that the GF100 brings is NVIDIA 3D Vision Surround. To be sure, this feature is not exclusive to the GF100 since NVIDIA says support will be extended also to the GT200 series of cards, but nevertheless, what it does is offer stereoscopic 3D gaming on three monitors.

The end result should be fantastic, going by how well the original GeForce 3D Vision kit performed. The only inhibiting factor we can think of is price. Considering that the new NVIDIA cards are incapable of driving three monitors simultaneously, 3D Vision Surround demands a SLI setup. Considering that the 3D workload has also increased tri-fold, that also necessitates the reason why an SLI setup is needed for 3D Vision Surround. Also, you'll need monitors with a refresh rate of at least 120Hz. All in all, the costs involved in getting a 3D Vision Surround system setup is going to be considerable.

Evidently, NVIDIA has done much to Fermi and the GF100 to beef it up for the next-generation of games. But talk is cheap as they say, so how does it actually perform in real-world applications? We'll find out soon enough.

The Good
Untouchable single-GPU performance
3D Gaming
The Bad
Costly to own and run
Runs incredibly hot