With two GPU, as four, the system must select a mode of returned to determine which graphics processor does what at which time and this for a maximum effectiveness and performances. It is to the ForceWare pilot that this task falls, and as for the SLI, Quad-SLI takes again the modes of returned SFR ( Split Frame Rendering ) and AFR ( Alternate Frame Rendering ). Let us recall for the absent ones, that mode SFR consists in distributing the load between the two graphics processors; it is to some extent a question of dividing the scene according to its heaviness. Thus one can imagine in a very heavy scene that the second GPU calculates the bottom of the scene whereas the first GPU deals with calculations top of the scene.
In AFR, returned is alternate and each graphics processor calculates an image on two. These modes of returned are logically adapted to the configuration four GPU. One is entitled thus to the "4-Way SFR" which consists in no longer distributing the load of calculation of a scene on two but out of four processors. The pilot can for example entrust the shader water to the first GPU, the shader of landscape (trees, sheets, etc) with the second GPU, the third fascinating GPU charges the effects of explosion of them whereas the last GPU deals with the remainder of the scene while rebuilding the image since the three other processors, which will reduce appreciably its effectiveness besides. It is besides about one of the differences of the implementation multi-VPU of NVIDIA vis-a-vis CrossFire of ATI, the Canadian firm using a chip dedicated to rebuild the image.
Another mode of returned, the "4-Way AFR" which consists in making calculate by each GPU an image on four. This last mode is announced like most largely compatible with the plays. But with a total of four graphics processors, the induced latency times can pose problem with 4-way AFR in particular if the profit brought by the SLI is lower than 30%.
With Quad-SLI, NVIDIA in addition inaugurates a new fashion of returned, with the barbarian name one nothing: the AFR of SFR. It is a question here of making calculate part of the image by the GPU 2 and 3 with a distribution of load (where SFR) and that an image on two (AFR thus). Side compatibility with the plays, NVIDIA indicates that the applications supporting the AFR are compatible with the AFR of SFR. As it is the case for the SLI, the profiles play a paramount role in the performances of a Quad-SLI system: if the play or the application which you use does not have of profile in pilots NVIDIA, it is likely not to benefit at all from the graphic power available. The more so as the firm proposes a mode known as of compatibility which decontaminates three of the four GPU! For the hour on the whole of the profiles available for the SLI, only a very restricted number was modified to benefit from the Quad-SLI configurations. Who more is and knowing that relatively few plays are optimized for the SLI, the plays actually optimized to function with two GPU are likely not to benefit at all from the four graphics processors... The more so as the added power will not be, and it rather logical, is fully exploited that in very high resolution by very greedy plays in power of shading like F.E.A.R. or Oblivion. For the other plays Quad SLI will probably make it possible to benefit from a maximum quality of filtering without too significant loss of performances. During our tests we could just like note that Call Of Duty 2 does not draw any advantage from Quad-SLI, besides Splinter Cell: Theory Chaos...
Who more is Quad-SLI is able to carry out an extreme antialiasing, NVIDIA inaugurating for the occasion the SLI 32xAA. To reach the 32x, each processor carries out one returned of the image according to the algorithm 8xS (multisampling 4x with supersampling 2x), with a light shift, then sends the result to the following graphics processor. These are thus not only one GPU which carries out the final mixture to obtain 32x, but each of the four GPU, carry out the mixture coming from the preceding GPU, which is simpler to manage. Rather than to use the bus PCI-Express train for the transfer of data between the GPU, SLI AA uses the bond SLI, bond which is likely besides under-to be slightly dimensioned considering the volume of data to treat. Moreover with SLI AA 32x the grid of the samples is fixed what can be one nothing tedious, the various samples being too close to offer a true difference in quality which is perceptible with the SLI AA 16x. Which more is with a screen 30 inches, the need for such a monumental antialiasing is far from being proven. In the facts the difference in visual quality between SLI AA 32x, and SLI AA 16x is far from being obvious whereas SLI AA 32x puts at knees QuadSLI in term of performances.
The other modes of antialiasing SLI like the 8x and the 16x remain extremely fortunately available. However QuadSLI AA 16x combines four images returned in 4x and not two images in 8x as with SLI AA 16x. Let us recall on this subject that it is still not possible to combine the SLI AA with Transparency Antialiasing which makes it possible for memory to filter textures in standard transparency fits latticework on or foliage in order to avoid the effects of staircase which they can sometimes exhiber. Also let us recall, that contrary to chart ATI Radeon X1xxx, chips NVIDIA of the family GeForce 7 do not make it possible to have simultaneously effects HDR and antialiasing. Let us recall however that certain plays like Half-Life 2: Lost Coast authorize via their implementation different from the HDR the activation of the FSAA on the chips GeForce 6/7.
