Intel, NEC to Develop Supercomputer Technologies of Tomorrow

[btarunr]

Intel Corporation and NEC Corporation today have agreed to jointly develop High Performance Computing (HPC) system technologies that will push the boundaries of supercomputing performance. NEC will bring these technologies to market in future supercomputers based on the Intel Xeon processor. NEC’s expertise in this field coupled with the Intel Xeon processor’s outstanding performance and its accelerating vector capabilities such as AVX will allow for higher performance supercomputers, satisfying customer demand for Intel architecture based products.

NEC will also continue to sell their existing SX vector processor-based products. A vector processor can perform a mathematical operation on several numbers simultaneously.

“Intel’s substantial investment in the Intel architecture, including the development of processors, chipsets, software compilers and other related products has expanded the usages of Intel Xeon processors in both the volume and high-end HPC market segments,” said Richard Dracott, general manager of Intel’s High Performance Computing Group. “Now with NEC further innovating on Intel Xeon processor-based systems, Intel is poised to bring Intel Xeon processor performance to an even wider supercomputing audience.”

“NEC’s substantial experience in the development of vector processing systems, including vector pipeline management, memory sub-systems, and high speed interconnect technology is a natural fit for taking Intel architecture further into new markets,” said Fumihiko Hisamitsu, General manager of HPC Division at NEC. “NEC will enrich its HPC product portfolio through this collaboration as well as continuous enhancement of its vector supercomputer.”

The initial focus of the companies’ collaboration will be the development of hardware and software solutions to enhance the memory bandwidth and scalability of Intel Xeon processor-based platforms. Such enhancements are intended to benefit systems targeting not only the very high end of the scientific computing market segment, but also to benefit smaller HPC installations.

HPC or supercomputers are used by scientists, doctors, non-profit organizations and businesses such as oil and gas exploration to speed research and testing. Currently, about four out of five among the top 500 supercomputers have Intel processors inside.

[aj28]

Interesting that this should come out the same day as this article from nVidia...
http://www.techpowerup.com/108389/Ne...tor_Of_10.html

[FordGT90Concept]

Intel moving away from Itanium with NEC's help? This seems like an odd arrangement, to say the least. I thought AVX was already complete or close to complete.

[Yukikaze]

Originally Posted by FordGT90Concept

Intel moving away from Itanium with NEC's help? This seems like an odd arrangement, to say the least. I thought AVX was already complete or close to complete.

AFAIK, Nehalem-EX is pretty much the death-knell for Itanium.

[FordGT90Concept]

Nehalem-EX is only up to 8-way. Itanium really doesn't have a limit (processors numbering in the thousands not unheard of). Intel might be partnering up with NEC to move Xeon to the next level seeing as IBM and other super computer manufacturers wouldn't want anything to do with that.

Itanium was, after all, pretty much a failure (it sold way under Intel's expectations). I wouldn't be surprised if Intel is killing Itanium, exiting from the super computer market, and will only offer Nehalem-EX. It is still mysterious to me why NEC is involved though.

[Jizzler]

I semi-coherently remember those days. Intel had this huge 5+ year plan to introduce IA-64 from the top down. I, being on the bottom (desktop), wasn't too happy. Started to question my faith... why have they forsaken me? But then! A light! AMD came riding in, darkness clearing in their wake, eyes dead set on Intel. The giant asks, what could you possible do to us?! You are small, weak, and hold a laughable market share! With a sly grin AMD responds, we bring 64bit to the masses! And to deal a second blow... "it also runs your current software". With that they knew they were beat. Logic and planning were no match for heart and spirit... along with the sound of untold numbers of wallets opening.

[FordGT90Concept]

x86-64 is vastly inferior to IA-64. IA-64 could handle 6 threads per core simultaneously in addition to processing two 32-bit instructions at once instead of a single 64-bit instruction. By releasing x86-64, AMD cemented the old, slow, tired architecture for another decade. AMD did us no favors and that hurdle will have to be still have to be crossed some day because of the severe limitations of the x86 instruction set.

Intel implemented their own version of x86-64 once AMD closed the door for a future instruction set in the near future. We're stuck with it until 128-bit quads, most likely, and I hope AMD doesn't extend x86 when that time comes or we'll all continue to be stuck with inefficient processors for modern workloads.

[Yukikaze]

Originally Posted by FordGT90Concept

x86-64 is vastly inferior to IA-64. IA-64 could handle 6 threads per core simultaneously in addition to processing two 32-bit instructions at once instead of a single 64-bit instruction. By releasing x86-64, AMD cemented the old, slow, tired architecture for another decade. AMD did us no favors and that hurdle will have to be still have to be crossed some day because of the severe limitations of the x86 instruction set.

Intel implemented their own version of x86-64 once AMD closed the door for a future instruction set in the near future. We're stuck with it until 128-bit quads, most likely, and I hope AMD doesn't extend x86 when that time comes or we'll all continue to be stuck with inefficient processors for modern workloads.

IA-64 was and is notoriously difficult (Along with other VLIW architectures) to write good compilers (That is, those that will be able to actually take advantage of the VLIW execution flow to speed up calculations) for. It isn't the end solution, either.

[Jizzler]

Oh, no dispute in that regard. It was certainly a screw you move, and well, we know how much the masses love a screw you move.

[FordGT90Concept]

Originally Posted by Yukikaze

It isn't the end solution, either.

Indeed, but at least it would have broke the stranglehold x86 has on consumer computing. That will have to happen eventually.