LT Panel
RT Panel
Just Visiting
Wednesday | September 26, 2018
Popular Review Links:
Nvidia GF100 Details

Nvidia GF100 Details

Last week at the end of CES Hardware Heaven was invited by Nvidia to an event where they covered the architecture of their new GF100 GPU in great detail. Today we are allowed to discuss this information with our readers however as with previous launches we will stick to the basics for now, preferring to cover the product in detail when we are allowed to discuss our experience of using the hardware in real world situations.

Therefore, below we have an overview of the GF100, what Nvidia aimed to achieve and some of the more interesting images from the product whitepaper.

So what is the GF100..?

The GF100 is the first GPU to be based on Nvidia’s Fermi architecture. It is fully compatible with DirectX11 and as a result DirectCompute and tessellation. It also builds on the great foundations of the GeForce 200 series with improved PhysX performance.

Nvidia have also stated that the GF100 offers a significantly improved gaming experience. Part of this is due to improved framerates with image quality receiving a boost also.

The key phrase from Nvidia’s presentation was “geometric realism”. Before we look at geometric realism and what it means for end users it is worth summarising what Nvidia’s goals were when designing GF100:

1) Exceptional Gaming Performance
2) First Rate Image Quality
3) Film-like Geometric Realism
4) A Revolutionary Compute Architecture for Gaming

Exceptional Gaming Performance:
To provide improved gaming performance the GF100  uses a third generation Streaming Multiprocessor architecture, doubling the number of CUDA cores over the previous GeForce products. ROP count is doubled and 8x anti-aliasing performance is increased through enhanced compression. Nvidia also say that the additional ROPs better balance overall GPU throughput for sections of any scene which cannot be compressed.

First Rate Image Quality:
In addition to improved 8x anti-aliasing performance Nvidia have also  implemented a new 32x CSAA mode which is based on eight multisamples and 24 coverage samples. CSAA has also been improved through smoother rendering of foliage and transparent textures with minimal performance penalty. Shadow mapping performance is also increased through the use of hardware accelerated DirectX 11 four-offset Gather 4.

Film-like Geometric Realism
Earlier we mentioned that Nvidia were pushing the phrase geometric realism. Improvements in this area were important for Nvidia as they felt it was an area where real end user benefits could be achieved. In the product papers Nvidia stated that today’s games use between one and two million polygons per frame, a computer generated scene from a film uses hundreds of millions. This is essentially because previous generations have improved the pixel shader architecture but neglected the triangle setup engine. As an example the GTX 285 has 150x the shading power of the GeForce FX but only 3x the geometry processing rate.

To improve geometric realism Nvidia looked to movies for inspiration and saw that two features were important, tessellation and displacement mapping. Improved tessellation is a key feature of the latest generation of GPUs and allows them to split large triangles into collections of smaller triangles. Displacement mapping can then change their relative position. Through using these two features together complex models can be reproduced using relatively simple instructions.

To further enhance performance in this area the GF100 replaces the traditional front end geometry processing architecture with a distributed geometry processing architecture which uses “Polymorph Engines”. Each Polymorph Engine contains a tessellation unit, attribute setup unit and other geometry processing units. Also, each Streaming Multiprocessor contains its own Polymorph Engine. Newly generated primitives are converted into pixels by four Raster Engines which operate in parallel (3 more than the previous generation). On-chip L1 and L2 caches then enable high bandwidth transfers between the Streaming Multiprocessors and tessellation units.

Nvidia state that the ability to perform parallel geometry processing is the single most important architectural improvement in the GF100 as it offers the ability to setup rates exceeding one primitive per clock while maintaining correct rendering order.

Revolutionary Compute Architecture for Gaming:
When gaming on GF100 threads operate independently with a predetermined pipeline and good memory access. For Compute operations the threads often communicate, work in no predetermined fashion and regularly read and write to different parts of memory. In this new architecture we have faster context switching between graphics and PhysX as well as concurrent compute kernel execution and enhanced caching which is ideal for irregular algorithms such as ray tracing and AI.

As an example of fast context switching a game might use DirectX 11 to render a scene, switch to CUDA for ray tracing, call a DirectCompute kernel for post processing and perform fluid simulations on PhysX. The GigaThread engine in GF100 reduces context switch time to 20 microseconds.

So that is a little of what we can expect to see from Nvidia’s products based on GF100 and Fermi when it reaches the market. As always Hardware Heaven will cover the launch with an article which includes real world comparison in gaming, gpu computing and media playback. For now, here are a few of the performance related images from the product whitepaper to give a taste of whats to come…

Product image


Product image


Product image


Product image


Product image


Product image

About Author

Stuart Davidson

It appears you have AdBlocking activated

Unfortunately AdBlockers interfere with the shopping cart process

To continue with the payment process can we ask you to

deactivate your AdBlocking plugin

or to whitelist this site. Then refresh the page

We thank you for your understanding

Hardwareheaven respect you right to employ plugins such as AdBlocker.
We would however ask you to consider whitelisting this site
We do not allow intrusive advertising and all our sponsors supply items
relevant to the content on the site.

Hardwareheaven Webmaster