How to use Accelwattch with Ampere architecture
66 views
Skip to first unread message
seunghwan
<rtwm27@gmail.com>Apr 14, 2022, 8:45:15 AM4/14/22
to accel-sim
Hi, i'm trying to estimate power consumption of Ampere architecture GPU (RTX 3090) during trace-driven execution.
But it seems to be applicable only to older architecture like Volta.
Is there any way to apply Accelwattch to Ampere architecture?
Regards,
Seunghwan
Vijay Kandiah
<vijaykandiah2018@u.northwestern.edu>May 6, 2022, 2:02:27 AM5/6/22
to accel-sim
Hi,
AccelWattch has not been tuned or validated yet for the Ampere architecture. The tuning process involves modifying the microbenchmarks to stress the Ampere architecture chip, adding support for any new Ampere specific components that are not modeled in AccelWattch, and going through the quadratic solver process detailed in Section 5 of our paper.
AccelWattch has not been tuned or validated yet for the Ampere architecture. The tuning process involves modifying the microbenchmarks to stress the Ampere architecture chip, adding support for any new Ampere specific components that are not modeled in AccelWattch, and going through the quadratic solver process detailed in Section 5 of our paper.
Regards,
Vijay
Reply all
Reply to author
Forward
0 new messages