GeForce G310 Fermi - Nvidia news! 2010

Final Considerations

These series of articles were an overview show all the power behind the Optimus technology. The system seems to work correctly and was implemented well as far as we could test. The possibility of having a good performance in the graphic field hasn’t taken a toll on the battery. And everything without manual switches, restarts or closing applications.

It’s definitely a step forward for those who need a graphic-performing notebook and a good battery life as well.

Final Considerations - Part 3

nVIDIA Optimus will be able to be used in Intel Core 2 Duo “Penryn” platforms, Atom X4xx “Pine Trail” and Core i3, i5 and i7 “Arrandale”. The Arrandale CPUS have the advantage of featuring the GPU integrated on the processor package, but it doesn’t prevent the usage of nVIDIA’s technology.

Together with the platforms, there will be GeForce 300M and 200M cards (depending on the models), and the future ION and GeForce M platforms that will come with Fermi.

Final Considerations - Part 2

The now more simple tasks between hardware and software that the Optimus technology brings is very interesting: the reduction of electric contacts, and the possibility of using either one driver video or the other, will probably result in cheaper computers, that will be able to give satisfying performances without a big hit in the battery life.

nVIDIA is planning on having around 50 models with the Optimus logo by this summer, manufactured by a good number of different companies. No word on the Apple systems, however, but it’s to be expected, as the Mac OS X systems would need different drivers.

Final Considerations - Part 1

The task that is performed by the nVIDIA Optimus technology is simple to describe: it makes it possible for notebooks to have good graphic performances without having a hit on the battery life and with an automatic system that makes the system use the discrete GPU as soon as there’s a system requirement, and turning it off automatically when it’s not needed. No manual intervention is needed in this case; it’s a very invisible, automatic and fast system.

It’s not easy, however, to talk about a technology that has “invisibility” as one of its main points. nVIDIA Optimus changes from a discrete GPU to the integrated one without leaving any signs, in order to enhance the user experience for notebooks where the 3D acceleration is needed. Videogames, high definition movies, CUDA applications, it doesn’t matter: the user profiles (or the ones manually chosen by the user) will take care of the switch.

The nature of the Optimus technology requires the presence of two video drivers simultaneously, where the integrated GPU driver results always active. It’s not that simple to have, and it was only made possible by Microsoft with Windows 7, which lists the two different graphic drivers, including the one for the discrete GPU, even when it’s turned off. This feature has cancelled the need of a driver interposer.

However, those with a notebook featuring other OS and being updated to Windows 7 will still need a driver interposer. In order to take advantage of the Optimus technology, other than having Windows 7, a special architecture is also needed, developed specifically for it.

In normal situations, without the Copy Engine, it’s the 3D Engine to perform the DMA transfer from memory to memory. This means that, many times per second, the engine renders the scene, then stops, starts transfering the scenes from its own dedicated memory to the one used by the integrated GPU, then goes back to rendering scenes. The interruptions are needed in order to avoid mistakes, but they also have an impact on performances, as the 3D Engine doesn’t work in a continuous way.

The Copy Engine, in that sense, works to copy and redistribute the rendered scenes from the 3D Engine, using the bidirectionality of the PCI-Express bus. Inside the discrete GPU there will be a 3D Engine that will work continuously, together with the Copy Engine that will prevent interruptions, copying and redistributing the rendered information. From there on, the work to make it arrive at the Display or video outputs is simple.

We’ve mentioned that the Optimus technology must be seen as a new way to approach a known problem, while there’s little to do with the silicon parts on the system. Truth is, something does change when it comes to the logic management and architecture, as a very important piece on the GPU, together with the Optimus technology, will become what nVIDIA calls Copy Engine.

Optimus does not use muxes and prevents problems that would come with passing from a driver video to the other. In order to do that, when there’s the presence of an application that requires the usage of the discrete GPU, this one transfers the scenes from its frame buffer to the one used by the integrated memory, going through the PCI-Express bus. In order to do so without a hit on performances, nVIDIA created the Copy Engine.

The second approach is one of the biggest new features on this technology, and it comes to the aid of those users that have no idea when to change GPUs, or to those who would rather allow the system to do it on its own. This approach bases itself on usage profiles that were analyzed one by one by nVIDIA. In that case, applications will be analyzed one by one, and encrypted profiles are sent to a nVIDIA dedicated web server.

Once the notebook is connected to the internet, nVIDIA will send the updated profiles to the computer, that will be automatically installed and without the need of an user intervention with the Optimus technology. Somewhat like the Microsoft updates for Windows, for example. These updates make sure that all users will have an automatic profile creation that will tell the OS which applications need to use the discrete GPU and which don’t.

There’s a problem that comes with all this, however: what should be rendered by the integrated GPU and what should be rendered by the discrete one? nVIDIA has adopted a double approach. There are three scenarios that the Routing Layer will take towards the discrete GPU:

- DirectX requests: Any 3D game requests or DirectX applicatoins
- DXVA requests: any video streaming that requires the usage of DirectX APIs (DXVA = DirectX Video Acceleration)
- CUDA requests

The driver panel presents a list of installed applications and it’s also possible to choose which card to use manually. The user can, in that way, decide whether to use the integrated GPU or the discrete one, or let the nVIDIA Optimus technology choose it instead, according to the necessities. This is a first approach, and we mentioned two. We’ll be explaining more in the next few articles.

How does it work?

nVIDIA Optimus works in a different way from any other discreet graphics implementation done previously by nVIDIA. At the start of each application or video streaming, the OS takes information from the nVIDIA driver panel and other profiles we’ll see shortly and is capable of choosing which GPU to use in order to process the video influx. Should the application or video not require extensive graphic elaboration, the choice will be the integrated GPU, which will render the scenes and then send it to the display or video output, depending on the case. In that situation, the discrete GPU remains completely turned off, as if it didn’t exist.

In the case of elaborations that require the intervention of a discrete GPU, Optimus’ “Routing Layer” will monitor the system request from the start of the workflow, sending the requests to the discreet GPU who will execute the necessary procedures, then send the rendered frames directly to the system memory. The OS will then send the data found in the memory to the display. An important thing to notice is that in both cases the rendered scenes will be sent to the same system memory, as there’s a single path to the two GPUs that leads the the visual output. Both scenes generated from either the integrated or discreet GPU will end on the same memory. The route is the same, without the need for mux and extremely complicated software in order to view the scene.