0
\$\begingroup\$

Is there any practical method to only measure the power consumption of combined cache memory. I am intereted in measuring the power conumption overhead when replacing blocks, and or filling the cache after start-up. Also of interest is extending this to the power consumption of the whole cache/bus/memory system.

If there is no practicl way, are there any simulator or power calculators which does so. Thanks

\$\endgroup\$
  • \$\begingroup\$ How about measuring the system consumption with caches enabled and vice versa? \$\endgroup\$ – Dzarda Jul 4 '14 at 17:18
  • \$\begingroup\$ That does not capture the block replacment power overhead. Plus, performance and behaviour with no caches is not similar with caches on. Subtracting both does not give the power consumed by caches \$\endgroup\$ – The Byzantine Jul 4 '14 at 23:04
2
\$\begingroup\$

There are device-level modeling tools, e.g. CACTI, NVSim, DESTINY, etc., which have been validated against real-world prototypes. These tools have been used in models such as McPAT, which have been integrated in several simulators such as Gem5, Sniper etc. For reference on these, see my paper and tool-review.

Note that measuring the energy consumption of block replacement or fill precisely is very difficult. For this, circuit-level modeling is required. The code of these tools is publicly available, so you can do desired modeling.

\$\endgroup\$
  • \$\begingroup\$ Man, You saved my PhD :) \$\endgroup\$ – The Byzantine Jul 20 '15 at 6:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.