Core temperatures have been a huge problem, and it is only getting worse. Portions of the core of a modern processor currently run at an excess of 125 degrees C (257 degrees F), while the rest, such as cache, is relatively cool in comparison. A team of researchers within Intel, led by Ram Krishnamurthy, has come up with something that addresses this. Their work enables substantial performance and power benefits for Intel's microprocessor execution cores.
"These technologies enable a huge performance improvement while cutting the energy requirement in half," says Steve Pawlowski, Intel Fellow, Corporate Technology Group and director, Microprocessor Technology Lab. "The technologies reduce active power leakage four times over the best conventional arithmetic circuits available."
This team achieved the fastest reported single-cycle 32-bit integer ALU in 130-nanometer (nm) CMOS (10GHz) and single-cycle 64-bit integer ALU in 90nm CMOS (7GHz) performance on silicon. These performance levels are two to three times higher than the rest of industry. Additionally, this design for a low-power, high-speed ALU that can run both 32- and 64-bit code will allow Intel to make chips that run both types of software very efficiently.
The ALU is code-named Nozomi after the high-speed Japanese bullet train. It is manufactured on the 90nm process and is readily usable on IA-32, IA-64, and XScale architectures.
"ALUs are the brains of a CPU—extremely performance critical and power hungry," Krishnamurthy says. "Integer execution cores consume a substantial portion of the microprocessor’s total power today and it’s getting worse as we scale to future technologies," he explains. "ALUs are a major hotspot."
"Our team developed two novel arithmetic circuit technologies: a sparse-tree carry-merge adder technology called the quaternary tree and a single-rail dynamic circuit technology called the dual function generator."
"These technologies enable energy-efficient integration of multiple cores or multiple ALUs within each of our execution cores with substantial performance benefits. And their energy-efficiency allows us to scale to wider bit-widths in the future. For example, the Nozomi single-cycle 64-bit integer ALU running in our lab consumes only 300 milliwatts. Typically such circuits consume more than twice that amount."
-Raystonn
= The views stated herein are my personal views, and not necessarily the views of my employer. =
"These technologies enable a huge performance improvement while cutting the energy requirement in half," says Steve Pawlowski, Intel Fellow, Corporate Technology Group and director, Microprocessor Technology Lab. "The technologies reduce active power leakage four times over the best conventional arithmetic circuits available."
This team achieved the fastest reported single-cycle 32-bit integer ALU in 130-nanometer (nm) CMOS (10GHz) and single-cycle 64-bit integer ALU in 90nm CMOS (7GHz) performance on silicon. These performance levels are two to three times higher than the rest of industry. Additionally, this design for a low-power, high-speed ALU that can run both 32- and 64-bit code will allow Intel to make chips that run both types of software very efficiently.
The ALU is code-named Nozomi after the high-speed Japanese bullet train. It is manufactured on the 90nm process and is readily usable on IA-32, IA-64, and XScale architectures.
"ALUs are the brains of a CPU—extremely performance critical and power hungry," Krishnamurthy says. "Integer execution cores consume a substantial portion of the microprocessor’s total power today and it’s getting worse as we scale to future technologies," he explains. "ALUs are a major hotspot."
"Our team developed two novel arithmetic circuit technologies: a sparse-tree carry-merge adder technology called the quaternary tree and a single-rail dynamic circuit technology called the dual function generator."
"These technologies enable energy-efficient integration of multiple cores or multiple ALUs within each of our execution cores with substantial performance benefits. And their energy-efficiency allows us to scale to wider bit-widths in the future. For example, the Nozomi single-cycle 64-bit integer ALU running in our lab consumes only 300 milliwatts. Typically such circuits consume more than twice that amount."
-Raystonn
= The views stated herein are my personal views, and not necessarily the views of my employer. =
Facebook
Twitter
Instagram