[MITgcm-devel] Hi-res cube stats

[chris hill]

I think 3mb cache would be fine.

I had one go at spanning multiple clusters with Infiniband and the 
results aren't great (don't tell anyone else on the development list but 
the solver slows down at about 64x64 tile size!). I am working on a fix 
for that at the moment. Otherwise the multi-cluster, inifiniband thing 
works.

Chris
Alistair Adcroft wrote:
> Chris,
> 
> Columbia has 6mb cache. Are we cache bound? Would we take a huge hit 
> with a 3mb cache?
> 
> Have you tried spanning multiple clusters (>512 procs) using InfiniBand?
> 
> A.