multiple gpus(3 quadro m4000) not able to run 100 transcode ffmpeg(cuda,nvdec,cuvid) services

hardware:
processor:dual xeon e5-2650 v4 (2x12core 24 threads) 24 core 48 threads
graphics processor: 3x nvidia quadro m4000/RTX4000(8gb gddr5)
ram: samsung ecc r dimm 256gb
psu:1600w
storage: 120gb ssd

we are trying to transcode 100 services h264 576x720 25 frame, ffmpeg 4.1.1 and ubuntu 18.04 if we use 1st card we are able to transcode 40 services once we add 2nd card we are able to transcode 75 services and when we add 3rd card we can add only 5 services please help us.

thanks in advance