IBMs plunders copper mine for new RS/6000s
But it's the I/O that steals the show
IBM has moved the switched SMP architecture it uses in its RS/6000 S80 down into its more affordable mid-range Unix servers. Three new lines – two six-ways and an eight-way – were rolled out today and as expected, all three use copper interconnects in their CPUs.
In case anyone misses the point, IBM mentions the cu-word seven times in as many sentences in its announcement.
But it's the switch that probably has the widest reaching implications.
IBM has been clobbering Sun's E10000 with S80 performance figures – thanks in no small part to the I/O throughput – for some months now. However, starting at around $250,000 a pop, this has had pretty much a symbolic significance. Today's announcements puts similar bandwidth down into the $20,000 range.
Of the three new lines, only the rackmounted two-to-eight way, M80 class uses the new switch, but IBM's Mike Mayes reckons this gives it half as much I/O throughput again as Sun's E10000, somewhere in the region of 18GBps. Compaq's Wildfire has the capacity to 20GBps, and has already clocked-in at 12GBps.
IBM introduced one-to-six way rack-mounted H80, and the one-to-six way F80 line carry what IBM now names "RS64" Power CPUs at 450MHz or 500MHz. All use IBM's own system bus and a hot-swap PCI bus. The latter is enabled for remote I/O switching, so you can drop an extra 14-slot board in other part of the room.
IBM also claims that its "Dynamic CPU Deallocation" routes around a failed processor – although this doesn't mean that the instructions are failed over. Hot-swap CPU upgrades for AIX are still out in the future, somewhere.
Thanks to our friend Terry Shannon, of Shannon Knows Compaq, for consistently pointing out the IBM's sales of its S80 are a tad misleading. With many of the S80s being considerably smaller and cheaper configurations than the E10000s, IBM is, to an extent, cannibalising its own margins. But it's equally true that the latest 80 range benchmarks puts Sun further on the defensive. ®
Sponsored: Beyond the Data Frontier