IBM is taking the wraps off the Power7 architecture.
Eight cores per chip with four-way simultaneous multithreading on each core. That's 32 simultaneous threads executing on one CPU. Dual DDR3 memory controllers for 100GB/sec of sustained memory bandwidth per chip. 32MB of shared L3 eDRAM cache in the center of each chip is almost as fast as SRAM, but takes a third of the space and consumes a fifth of the standby power, and IBM claims six times better latency than off-chip L3 cache. Imagine having your entire OS kernel in cache. Power saving options include a CPU sleep mode that drops voltage to the minimum required to maintain state, and data-integrity features include 64-byte ECC, the ability to do RAM mirroring, and CPU fail-over.
And they've already ported Linux to it, both natively and in x86 emulation mode, as well as AIX and IBM's i OS.
Gentlemen, start your drooling. There's a new big dog on the block.