>> Beyond 8-way, you need glue logic (hypertransport switches?) and
>> latency seems bound to become an issue.
>Nope. Just extend the ladder. Each cpu talks to three other entities,
>either cpu or I/O. Can be extended arbitrarily until latencies are too
>high.
You seem to be missing one critical piece from the OLS talk. The HT
protocol (or something related) can't handle more than 8 CPU's in a single
configuration. You need to have some kind of bridge to connect
more than 8CPU's together, although systems with more than 8 CPU's have
not been discussed officially anywhere, afaik.
8 CPU's and less belongs to the SUMO category (Sufficiently Uniform Memory
Organization, apparently new AMD terminology) whereas 9 CPU's and more is
likely to be NUMA.
-- Jukka Honkela
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/