Will NVENC benefit from having dual GTX1070 in SLI bridge instead of a single GTX1070?

Searched high and low, but could not find anything regarding SLI support in the NVidia Video SDK…
Thanks!

We had issues running the Capture SDK on a single 1080TI. So are you sure it works on a single GTX1070?

Have you tested this under Linux? We ran into license violations, and were told to be needing a P5000.

Have not tested anything, just wondering if my encoding fps performance would improve by using 2 cards as opposed to a single one…thanks!

Assuming you are on Windows platform, two GeForce’s in SLI configuration will not double NVENC performance automatically, because Windows will treat them as a single D3D adapter.

If you keep the GPUs separate (without SLI), Windows treats them as two D3D adapters and then the application can distribute encoding sessions between the two GPUs by selecting the appropriate CUDA device or D3D adapter ordinal (depending upon whether the application uses CUDA or D3D for NVENC interaction). In such case, your aggregate encoding performance will double, because you can direct your encoding jobs to NVENC on each GPU independently.

1 Like

I waited for official answer but I suppose that is not exact.

I suppose that one (or more) encoder stream can be encoded only on one HW encoder engine (eg. many-to-one relationship therefore you need at least two streams to utilize two HW encoders). GTX1070 is GP104-200 and it can have two HW encoder engines or only one HW encoder engine because it is partially damaged/disabled chip (You can compare all GP104 chip variants GTX1070/GTX1070Ti/GTX1080/P4/P4000/P5000 https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units. Nvidia does not disclosure number of HW encoder engines for GTX cards.). There is also API limit that enable only 2 simultaneous encoder streams per system (not per card) for customer “GTX” grade and lowend “Quadro” (<P2000).

If your card have two HW encoder engines it useless to add any other GTX card (with or without SLI) to your system for stream encoding due to API limit.