Taking advantage of the CES, Nvidia decided to unveil Hybrid SLI (which we have already mentioned several times) in response to AMD’s various demonstrations of Hybrid CrossFire.
You may recall, Hybrid SLI enables combining a dedicated graphic card (dGPU) to an integrated graphic core (mGPU) in order to improve performances by reducing power consumption (and related problems) by deactivating the more powerful graphic card(s) when not needed.
Higher performances
The performance mode, called GeForce Boost, simply consists of SLI between the graphic card and the integrated graphic core. Gains can vary from significant, if the dedicated graphic card is low performance, to symbolic, if a more performance model is involved. This is a way for Nvidia to give better results when one of its GPUs is associated to one of its chipsets.
Of course, it doesn’t transform a machine not adapted for games into a high performance configuration. n its website, Nvidia announces that the GeForce 8400 GS, 8500 GT are compatible.
Lower power consumption
The energy savings mode concept, Hybrid Power, is rather simple: deactivate the graphic cards in idle and let the integrated graphic core, which is more economical in this area, handle desktop display or the decoding of HD video. Given that a GeForce 8800 Ultra consumes almost 100 watts in idle (300 watts wasted in a tri-SLI system !), the interest is evident. In practice however its implementation is rather complex.
Moving from the integrated graphic core to the dedicated system is not as simple as it appears especially with Windows Vista and its 3D desktop, protection and various restrictions. In our talk with Nvidia about this technology a few months ago, the manufacturer told us it wasn’t possible given the current situation to dynamically carry out this transition on-the-fly and that it would therefore be manual. This is a significant weakness because we can see most users finding it bothersome to constantly activate/deactivate Hybrid Power. Nvidia specifies in a press release that future drivers should add an automatic control of Hybrid Power. In the meantime, the opposite effect could be achieved: when graphic cards are functioning the graphic core will also run because it is the only one to manage video outputs in these modes.
The only thing is that no current cards on the market are announced as being Hybrid Power compatible. We can therefore hope that Nvidia made this choice so that it will only work with flexible enouh graphic cards to make automatic functioning possible.
Some practical and complex details
To be able to go from one rendering system to another, in theory, it would have been required to move the monitor’s connection from one output to another. Of course, this is unthinkable. The solution consists of connecting the monitor to the integrated graphic core’s output (and therefore on the motherboard) and in the case of rendering via the graphic cards to move their frame buffer to the integrated graphic core which will then be in charge of display.
This raises several questions.
- What effect will this have on performances or rather the increase in the delay between the start of image processing and its display? Nvidia’s response is that this technology doesn’t have an impact on performances. Of course we don’t believe this for a second and would prefer that Nvidia give us the extra delay as little as it may be. Transfer time via the PCI Express bus + 1 frame ?
- What about decoding of protected videos? Is the protection chain respected if the video has to go from the dedicated graphic cards to the integrated graphic core? Or do we always have to go into energy savings mode to take advantage of video acceleration?
- What about the maximum supported resolution? In our last talk with Nvidia on the subject, they told us in the current situation, resolution was limited to 1600x1200 but they hoped to eventually attain higher resolutions. What will this be in the final product?
A new chip for Hybrid SLIThis technology will appear with the MCP78, Nvidia’s new integrated chipset which will support HyperTransport 3.0, PCI Express 2.0 and an integrated graphic core that is DirectX 10 compatible (thus based on GeForce 8 architecture) and has a new video engine. This last component entirely supports the decoding of VC-1 videos, a deficiency of the previous video engine implemented in GeForce 8s. Four versions have been announced.
There is the GeForce 8200 (indeed the chipset’s name), a "simple" MCP78 with a graphic port, the nForce 730a, theoretically similar to the GeForce 8200 but without Hybrid Power support (there is GeForce Boost), and finally, the nForce 750a and 780a which have a respective 2 and 3 graphic ports. Nvidia therefore offers a complete line based on a single and unique chip in which a PCI Express switch was added to the nForce 750a and 780a as the MCP78 does not natively support enough PCI Express lanes.
A revolution ?For Nvidia Hybrid SLI is a very strategic move. The chipset market is not easy or rather is not easy to make money on. AMD and Intel can easily offer CPU + chipset packages which Nvidia cannot do. In addition, more and more significant parts of chipsets will be integrated to CPUs (notably with the Nehalem) only leaving more or less external "commodities". Nvidia thus needs to rapidly find ways to add something that will keep value and interest in its chipsets. SLI is one of them but this is a minuscule market whereas Hybrid SLI involves a much larger one.
The only thing is that Nvidia has only announced the technological concepts. While these are ultimately interesting there is still some ambiguity and it would be unfortunate if it was launched too early with too many limitations.
Finally, a last thought on the subject. Nvidia announces that all of its chipsets will now have an integrated graphic core. CPUs will also start to do the same. Isn't there an opportunity to make the graphic card becoming an accelerator (in the broad sense of this word)? To purge GPUs of all management of video connections and decoding? This will soon be a redundant capability…
