One of the functions of GPUs is the ability to encode and decode video. The latter do this very quickly and allow them to play video file formats at a speed fast enough to be viewed without a problem and without a very large processing load. So you may have wondered if it was possible to use one in a complementary way. In other words, use a second graphics card for streaming.
The problem is that generating a video file is much more expensive and if we want to do it during playback and streaming, the technical requirements skyrocket and the work of the hardware video codec is limited. The solution? Juice the rest of the graphics card’s resources, which means cutting it out of the game itself. This brings us to a very simple question.
Can I use a second graphics card for streaming?
Most likely you have wondered, unfortunately this is not possible and there is a reason for this and we are going to explain it to you. Each graphics card can access two different memory pools: its own memory and its system memory. They cannot directly access the RAM of other devices that share the PCI Express interface because they have no access or coherency mechanisms. In other words, the second graphics card has no way of accessing the memory of the first one and if it could somehow, it wouldn’t be aware of the recent changes.
Which brings us to another question: can you combine a dedicated with an integrated? Well, yes, but the work of synchronization between the two graphics cards must be done down to the millimeter and this is something that can only be achieved if both, iGPU and dGPU, are from the same manufacturer and the same architecture . This is something Intel intends to exploit with its ARC graphics cards combined with its Intel Core and AMD in the Ryzen-Radeon duo. The idea is simply that the graphics card integrated into the processor takes care of encoding the video or serves as its support.
Unfortunately, this is something very new and based on the fact that over 85% of PC graphics card users use NVIDIA. It is therefore clear that such a scenario can only be used by a few users. Moreover, with the loss of dual configurations such as SLI and Crossfire, communication between two graphics cards is no longer possible.
The situation might change in the future, or not
The reason for this is the CXL standard which, among other benefits, adds memory coherency to all PCI Express devices, which would allow communication between multiple graphics cards at the same time. That is, you could have one GPU generating the game image, then a bunch of additional cards splitting the high-speed image encoding.
Let’s not forget that video codecs encode the image in blocks, and can therefore divide the work between them. Additionally, it will be something common in more advanced Cloud Gaming systems if it is not already. The objective? This will minimize latency in game streaming without sacrificing image quality. The problem is that everything indicates that the CXL for now will be something only for workstations and servers and not for the home market.