IBM looking to release 5GHz Power6 server chip

We concur, 5GHz isn't nearly as impressive as IBM's 500GHz microprocessor, and can't hold a candle to the Blue Gene / L supercomputer, but the company is hoping to at least give the likes of AMD, Sun Microsystems, and Intel a run for their money in the oh-so-competitive server market. While it has already been confirmed that Intel's got quad-core on its mind for next year, and AMD's possibilities are wide open now that its chips are comfortably resting within Michael Dell's boxes, IBM has some stiff opposition to topple. While its current Power5+ server chip tops out before hitting the 3GHz mark, Big Blue plans on pushing the Power6 processor to 5GHz thanks to the "65-nanometer process geometry," 75Gbps buses, and voltage thresholds "as low as 0.8 volts." For those keeping track, IBM is doubling the frequency in its new chip and cutting energy requirements by making the CPU "more efficient with improvements such as computing floating point decimals in hardware instead of software." While there's no telling how much these mighty Power6 chips will dent that IT budget, the chips are currently on track for a "mid-2007" release -- but we all know how reliable these corporate giants and their claims are aren't.

[Igor @ Oct 12th 2006 5:27AM]

whule intel and amd double cores they double the speed, nice
i wonder hoe much will power consumption be

[Zorque @ Oct 12th 2006 5:28AM]

It's too bad it's probably geared towards server instruction sets and therefore lacking on the game front. Needless to say, 5GHz would be rather helpful in that respect.

[joram Oudenaarde @ Oct 12th 2006 6:39AM]

GHz Gigasmug... anyone ever heard about the Mhz-myth?
I'm sure that their new processor will be fast, but 2 cores are better then one. I do believe that making 10 cores in the future will be overkill, and other solutions have to be found. But refusing to go forward and standing still shouting that 5Ghz is better then 3Ghz dualcore is just plain stupid. Give us proof instead of hot air please

[James Boswell @ Oct 12th 2006 7:35AM]

a 2.3Ghz POWER5+, which this should scale up from pretty well from, turns in

SPECInt2000 = 1900
SpecFP2000 = 4051 (okay, that's fudged, because I couldn't find the 2.3Ghz number on SPECmine, the FP number is actually from a 2.1Ghz Power5+, so consider it lowballed)

a 2.93Ghz Core 2 Extreme turns in
SPECInt2000 = 3119 (It's a complete and utter beast at this)
and
SPECfp2000 = 3050


Oh, and the POWER series has been dualcore since the POWER4 in 2001.

[Kevin @ Oct 12th 2006 9:53AM]

The Power6 will be Dual Core. It will incororate for the 1st time in a Power design Altivec aka Velocity engine aka Vector units, seen 1st in Motorola PPC G4 chips and in IBM G5 chips.

[celestrion @ Oct 25th 2006 5:43PM]

You do realize that a POWER6 MCM has -eight- cores on it, right? IBM was doing dual cores with POWER5 and POWER4+ before Intel decided to join the rest of the world in the 21st century. As for proof, if it's as least as fast as POWER5+ was clock-for-clock, it's roughly as fast clock-for-clock as Core 2 for integer ops, and nearly twice as fast for floating point ops. Chances are, though, that POWER6 will be faster per clock tick than POWER5+.

The big question is whether or not IBM will be able to keep their promise of POWER6 consuming roughly the same amount of electricity and releasing roughly the same amount of heat as POWER5+.

[Ernie @ Oct 12th 2006 4:49PM]

Would this beat the xeon quad cores in the Xserves right now??