We are experiencing a lot of lag when we start the profiler. Clicking on the different tabs (bottleneck, Utilisation, adding a new draw calls perf marker, etc.) cause huge lags as well. Previously, I was using a 780 ti with 3gig of vram and at that time, just doing frame capture was extremely long. Now that I’m using a Titan Black with 6gig of vram(thanks to NVIDIA for the gift when we pass at BeHaviour Montreal), capturing frame is quite fast, but still extremely long to play with the profiler. Same kind of lags while debugging shaders. The scene used is pretty heavy with an average total of primitive of 20 million (CSM with 4 cascades, etc.). As a side note, my system is using 32gig of system ram and it never reach more than 16gig, CPU is a i7 3930K (Sandy Bridge-E). I’m using nsight 4.1 RC2
I have the feeling that when the complexity go over a certain threshold, performance gets exponentially affected.
So, I was wondering if this kind of issue has been already track down by NVIDIA and if some speed improvement is on the go.
Last question: are you planning to provide a way to save a full replay so we can restart the profiler after a VS crash of application crash? It is very frustrating to spend more then a hour to get the profiler running and then restart everything after a crash.
By the way, congratulation to the dev team, we really enjoy the mindset under NSIGHT and hope that new version will make the performance of the tool manageable in production.