GPGPU-Sim provides a detailed simulation model of contemporary NVIDIA GPUs running CUDA and/or OpenCL workloads. It includes support for features such as TensorCores and CUDA Dynamic Parallelism as well as a performance visualization tool, AerialVisoin, and an integrated energy model, GPUWattch.
In the GPUWattch paper, they mention their implementation of dynamic voltage-frequency scaling (DVFS) similar to how it is done on the gtx 480 (with 7 pre-determined "p states" corresponding to different frequency levels and voltage scaling done using the 45 nm predictive model).
Is this DVFS algorithm done in the gpuwattch source code automatically? Else, is there a way to do it during runtime of GPGPU-sim (through adding a feedback mechanism in the source code in power_interface.cc & gpgpu_sim_wrapper.cc)?
I would like to dynamically change the clock rate (which seems like it can only be changed in the initial xml config file) & supply voltage (which I'm not sure how to change) as a program is running depending on other gpgpu-sim measurements, such as # cache misses & stall cycles.
In the GPUWattch paper, they mention their implementation of dynamic voltage-frequency scaling (DVFS) similar to how it is done on the gtx 480 (with 7 pre-determined "p states" corresponding to different frequency levels and voltage scaling done using the 45 nm predictive model).
Is this DVFS algorithm done in the gpuwattch source code automatically? Else, is there a way to do it during runtime of GPGPU-sim (through adding a feedback mechanism in the source code in power_interface.cc & gpgpu_sim_wrapper.cc)?
I would like to dynamically change the clock rate (which seems like it can only be changed in the initial xml config file) & supply voltage (which I'm not sure how to change) as a program is running depending on other gpgpu-sim measurements, such as # cache misses & stall cycles.
Thanks
Ref: https://people.ece.ubc.ca/aamodt/papers/gpuwattch.isca2013.pdf, page 10