Although the SLI technology was finally abandoned,some guys like me were still using the platform with SLI(2xRTX3090 with NVLINK for example).
And more and more people start using external graphic dock with thunderbolt(Which means the system has an intermal mobile version graphic card and an external desktop graphic card)
But enabling the SLI option in the control panel is invalid,because most of the applications were not support it at all.
Since the traditional parallel computing acceleration method is not feasible, can we change the acceleration method？
I guess it could:
The traditional way of acceleration is in parallel computing.The graphics card produces the interleaved frame, and then arranges the frame to output,which is difficult to design the engine and control the latency to sync.
Now we have DLSS and other technology to imporve the original graphic quality and lighten the load of classic computing proccess.And this is a serial process can be divided into two parts, generating the original data and upscaling or HDRing or Interpolating prossess.
Card A do the main work,generate the low res data and transmit it to Card B streamly via PCIe or NVLINK,and then Card B process the low res data and output the video stream.
This can improve the performance for some cases using the built-in display connected to the internal graphic card and decrease frame latency.It can also maximum the usage of multi-card system by offload the main card and reload it on the other card,which could have both higher frame rate and upscale quality.(Maybe also better selling of NVIDIA RTX4series graphic card,because consumers will purchase it to upgrade there computer even by using the mid-end model to work as Card B)
PS:I’m just a Chinese undergraduate with little knowledge and poor English.So if you find my idea silly,I ask for your understanding.