Intel shows off 48 cores
Intel demoed its experimental 48 core chip at Cebit this morning as it (almost) nailed down the dates for its latest real world server chips.
The vendor has been touting the 48 core part around as part of its vision of massively parallel systems. It describes it as a single chip cloud.
An Intel researcher said the part used a mesh interconnect, meaning each core could run its own operating sytem, or operate as part of a single computer.
The demo today showed software monitoring energy usage across the cores - a feat in itself. We'd have shown you a picture of the part, but really, all you'd have seen was a lot of fan.
The vendor also showcased its latest real world server platform, the 8 core, 16 thread Nehalem EX, with a system supporting a 32GB slice and dice scan through a human corpse.
The press materials said the part should be available this quarter, EMEA chief Christian Morales said this was not confirmed, but the part should ship in a "matter of weeks".
Intel also showed the Westmere-EP, the 32nm 6 core server chip, also due this quarter.
Intel also announced three new Atoms for the embedded market, including a dual core version. The new platforms feature integrated graphics and memory controllers. Intel flagged digital signage and the automotive sector as major markets for the parts. ®
I do hope so.
I quite fancy a future in which disposing of a Terminator is merely a matter of stuffing some handy object into its cooling fan and waiting for the inevitable meltdown.
Like a One Man Band, only worse!
"Single Chip Cloud" isn't that like a "one man band?" Plays lots of instruments, none of them for very long, none of them particularly well, and people rarely go for second hearing.
Sort of pointless really.
How many pins do you need to get reasonable memory bandwidth off this beast, or are most of the cores waiting for memory most of the time? What size heatsink does it need, or is it water cooled? And other little real-world trivia.
You can already get 48 cores in a Proliant of the right flavour, but there aren't that many people interested, because there aren't that many applications that suit the massive-SMP model.
The idea that this is "a cloud on a chip" is laughable - with all those resources in one chip (and presumably one single OS image?) where's the scalability and diversity and resilience which a properly-implemented disaster-tolerant multi-site cluster (that's what they used to be called but most MBAs and Microsoft-certified folk can't string that many words together so "cloud" came in) can provide?
Beer. Liquid cooling. It's the way of the future.