Crusoe blade strikes 1Ghz, fries Banias?
Still the low power champ
It's plowing a lonely furrow, but Texas-based blade pioneer RLXcontinues to champion ultra high density boxes that put rivals to shame in the power stakes.
Today RLX announced a blade based on the 1Ghz Crusoe processor from Transmeta. It fits the 24 blades in the original 3U box form factor that RLX launched in May last year.
Hewlett Packard and Sun Microsystems have adopted the "blade" moniker for their carrier-grade servers, but they bear little resemblance to the densities that RLX can boast, and they don't address the same markets.
RLX's system isn't a shared memory SMP, but really a Beowulf cluster in a box.
RLX predated the announcement of the public unveiling Crusoe, and the early lab work was around an Intel mobile Celeron CPU. The Crusoe gave the company a much higher density. RLX officially added a Pentium III-M based blade earlier this year.
However the 1.2Ghz mobile PIII system only uses every other slot in the system: with only half the density, at 12 in a 3U box. This isn't immediately obvious, but the giveaway is in the dimensions: the Crusoe blade is .58" wide, while the Pentium blade 1.16" wide.
And that low power, high density combination was the clincher for Los Alamos labs when it unveiled its Green Density cluster seven long months ago. (That story has a nice picture of the Intel CTO demonstrating why you shouldn't tamper with the fans on an Itanium system).
Now, despite the propaganda war, surely that tells you everything you need to know about the state of Intel's pre-Banias mobile chips.
Speaking of which, the splendid Mooly Eden, who's in charge of Banias development for in Intel seemed quite bullish about the prospect of OEMs looking at Banias in blade servers. Presumably those OEMs will want an as yet-announced chipset, as both Calexico and Odem chipsets have support for 802.11 wireless networking, and we can't quite see why you'd want 24 radios humming away in your 3U server. ®
Sponsored: Benefits from the lessons learned in HPC