Hardware DVXA h264 decoding: number of simultaneous streams?

Question

GPGPU

We are developing a video surveillance system for streaming ip cameras of the h264 codec.
Solutions for hardware video stream decoding on a graphics card look tempting (we use DXVA, because windows). However, in our task, we need to decode not one stream with a large bitrate, but many streams with an average / low bitrate. And here we come across the unpredictability of the solution: it is impossible to determine in advance the number of video streams that will work stably on a given graphics card. For example, nvidia gtx 680 gives 12 streams at 720p and about 8 at 1080p. Further, the streams either simply do not open or cause terrible glitches up to the fall of the graphics driver. Buying another/two or three cards in sli is not a problem, the problem is understanding the system load.
Are there any utilities to estimate the loading of a graphics card? Is there any information / general recommendations / comparison tables on the number of possible simultaneous video decoding processes on graphic cards?