L1 cache hits 0%

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I am using the NVidia Nsight tools to profile my application and I am seeing that the cache hit percentage for the L1 cache is 0% and around 70% for L2.

Could anyone tell me what could possibly cause this?

2

Let me guess: You are using a Kepler GPU (compute capability 3.x).

Kepler uses L1 only to cache local memory accesses. If your kernel does not use local memory, the profiler shows 0% for the L1 hit rate.

3

Yeah, I’m using a GTX 680M to test. I have a GTX 590 in my desktop so I will try on that. Thanks for the reply.

Edit: Just tried on my GTX 590, getting around a 90% hitrate on the L1 cache, thanks for the help tera.