Hi Zhengrong,

On Thu, Jan 12, 2023 at 2:40 PM Zhengrong Wang via gem5-users <
gem5-users@gem5.org> wrote:

> Hi All,
>
> Our group has been using gem5 for architecture research for a long time,
> and we currently want to explore some ideas on modern multi-chiplet
> architectures, e.g. AMD's EPYC. We want to understand the tradeoffs of NoC
> design and coherence protocol when there are multiple chiplets connected
> together, each chiplet with its own NoC and caches and the LLC can be
> private or shared between chiplets.
>
> To start with, we want to understand the current support in gem5 for such
> systems. Specifically:
>
> 1. Are there existing scripts/configurations modelling an AMD EPYC like
> systems? We couldn't find them in gem5 or gem5 resources. Any pointers
> would be appreciated.
>

I made a WIP a couple of months ago here:
https://gem5-review.googlesource.com/c/public/gem5/+/65612. Note that this
*doesn't work*. I tried to set things up so that the LLC was a victim
cache, but those options seem to have broken the coherence. I haven't had
time to debug.

This is still using the SimpleNetwork, but adding Garnet should be
possible... once it's working ;).

Feel free to pick this up and build on top of it!


>
> 2. If not and we have to model the system by ourselves, what is a good
> starting point? The most reasonable one to us is the CHI protocol, which
> seems to be flexible enough to model the coherence protocol.
>

Yeah, this is exactly what I was thinking as well!


>
> Thanks in advance for any help you can provide.
>
> Bests,
> Zhengrong
>
> *王 钲 荣*
>
> Zhengrong Wang
> Computer Science Department
> University of California, Los Angeles
> California, USA
> 90024
>
> Work Email: seanyukig...@gmail.com
> Mobile :+1 310-447-4568
>
>
> _______________________________________________
> gem5-users mailing list -- gem5-users@gem5.org
> To unsubscribe send an email to gem5-users-le...@gem5.org
>
_______________________________________________
gem5-users mailing list -- gem5-users@gem5.org
To unsubscribe send an email to gem5-users-le...@gem5.org

Reply via email to