Re: [Gems-users] simulating a many-core system

[Dan Gibson <degibson@xxxxxxxx>]

64-way is possible with GEMS+Simics, but very much beyond 64 and you will find resistance from Simics. Speicifcally, depending on your Simics version, it will be difficult to even specify a valid Simics configuration with that many processors, and getting an operating system to boot will also be difficult.

With large numbers of processors (each "core" will be one Simics processor), simulation time will also skyrocket (~ to the number of processors). Just FYI.

As far as modifications to GEMS, Ruby will support that many cores directly, though performance will not scale for all coherence protocols. Perhaps my esteemed colleagues will comment on how to create L3+ caches -- I know it has been done before.

Regards,
Dan

Brad Moore wrote:

Hey all,

I would like to simulate a many-core (64+ core) system using GEMS.  This raises the following questions:

a) In general, would the framework support it?
b) If so, would I basically be modifying some parameters and extending an existing coherence protocol to support L3 (possibly L4) caches?

Any quick pointers would be appreciated.  I'm a fairly new user and my end goal is to use such a system to evaluate the performance
of competing benchmarks.  So far, my impression is that GEMS is the simulator most easily extended to these type of systems.

Thanks,
-Brad

_______________________________________________
Gems-users mailing list
Gems-users@xxxxxxxxxxx
https://lists.cs.wisc.edu/mailman/listinfo/gems-users
Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.