adi

Did you ever fix this issue?

I agree, we should throw together some quick prototypes that use different methods of varying complexity/method to see which one gets us closest to our goal. @leofang yes, vanilla nvbench...

Here are some initial comparision runs I did based on `time.perf_counter()` for python and ``` auto start = std::chrono::high_resolution_clock::now(); auto end = std::chrono::high_resolution_clock::now(); auto duration = std::chrono::duration_cast(end - start); ```...

> [@adii-n](https://github.com/adii-n) interesting, thanks! Could you post the benchmark code that you used? For example, here is the code python/c++ used for context ops measurement ``` for _ in range(iterations):...