Pages Menu
TwitterRssFacebook
Categories Menu

Posted by on Jun 18, 2013 in Hardware Articles |

Intel Announces Xeon Phi coprocessors second generation “Knights Landing”

A year after the announcement of the coprocessors Phi Xeon , announces to his successors, the new parallel accelerators “Knights Landing” ( 4 or 4).

The family of coprocessors Many Integrated Core “MIC”

As many will recall, initially developed Intel Many Integrated Core architecture (formerly known as Larrabee or MiniChips) to constitute both GPUs dedicated video card, as cGPUs for massively parallel accelerators or coprocessors, which like the Nvidia GPUs and AMD, running applications accelerated by OpenCL GPGPU APIs.

Larrabee (MIC 1) never came out commercially, nor his successor known as Knights Ferry (MIC 2 or Larrabee 2) which was distributed to a few manufacturers of supercomputers. Xeon Phi “Knights Corner” (MIC 3 or Larrabee 3) was the first commercial version based on MIC, and later this year will be replaced by second-generation Xeon Phi “Knights Landing” (MIC 4 or Larrabee 4).

Knights Landing

Similar to other architecture based coprocessors MIC consists x86 multicore accompanied by a 16-Wide vector unit (similar to a SIMD unit 16 shader processors as used in the architecture based GPUs Graphics AMD Graphics Core Next).

Knights Landing each core is capable of executing four processing threads; factor that adding it to be manufactured through the manufacturing process to Tri Gate Intel, has a greater number of nuclei compared with its predecessor, have to easily exceed the 1.2 teraflops of computing power in double-precision operations that shows the current Xeon 7100 Phi “Knights Corner”.

Memory on-die

Knights Landing incorporate a yet undisclosed amount of high performance memory built into the same chip package (very similar to the memory of the SoC Crystalwell Haswell-H for high-performance notebook PCs All-in-one), which will be used by way of a large and fast cache, which Intel said the chip will allow you to have a exceed bandwidth GDDR5 memory present in competing products (cGPUs Nvidia Tesla and AMD FirePro).

In addition to on-die memory, Knights Landing will also DDR3 or DDR4 memory with 512-bit bus and main memory.

The two modes of Knights Corner

First we have that cGPU second generation Xeon Phi format available PCI Express 3.0 card with dual slot cooling, format will be updated any or supercomputer that has PCI Express cards with several second-generation Xeon Phi.

Second-generation Xeon Phi “Knights Landing” also available is the LGA socket (not yet disclosed how many pins you possess), which is focused on the market for micro-servers. The main novelty of this version is that it is self-sufficient and does not require a dedicated microprocessor. Can be used in mini-format cards servers.

Another interesting socket version of Knights Landing, is that you can install ECC DDR3 or DDR4 traditional, which will be used as main memory for the chip, and it can be accelerated by its on-die memory. No doubt an interesting option that will shake the market for micro-servers currently dominated by ARM.

Less use

Thanks to be manufactured with the 14nm manufacturing process Intel Tri Gate, and based on a new modern micro-architecture aimed at improving performance per watt and increase interconnection capacity, Knights Landing allow the construction of super-computers exceeding 10,000 nodes and even above configurations monstrous 100 thousand nodes.

Comes computing

While AMD coined the term “heterogeneous computing” as the industry standard term to refer to solutions that combine the power of CPU and GPU computing for optimal execution of all tasks in the optimal microprocessor.

Intel announces that heterogeneous computing is long gone and announced the early arrival of neo-heterogeneous computing (neo-heterogeneity), which expands the capabilities of heterogeneous computing by providing it with a single simplified programming model.

We could say that was erected as the direct rival to Heterogeneous System Architecture “HSA” AMD (which also promises a unified programming model and simplified for CPU and GPU), which should arrive later this year.

Availability

Probably the new Intel Xeon Phi coprocessors second generation be released in the course of the next year (or early 2015), most likely close to the launch dates of future Haswell-EX/EP/EN Xeon microprocessors, which premiered a controller DDR4-2133 memory.

Conclusions

Second-generation Xeon Phi “Knights Landing” will undoubtedly be a very tough opponent for Nvidia Tesla based cGPUs Maxwell and AMD FirePro based on Volcanic Islands, which must get almost the same date.

Much more interesting is the version socket LGA, which will redefine the market for micro-servers thanks to its high density, although we should mention that for next year expected product massive multicore ARM-based Cortex-A57, as well as solutions Similar based on future Jaguar + micro-architecture.

For now that’s all that Intel has revealed Knights Landing, but we assume that this new chip lurks Intel further secrets will be revealed slowly as you approach the release date.

Links:
Unleashes Next-Generation Intel Xeon Phi “Knights Landing” Co-Processor (xbit Labs)
Intel previews future ‘Knights Landing’ Xeon Phi x86 coprocessor with integrated memory (The Register)
Knights Landing Intel announces 14nm Xeon processor in Phi update (V3)

.

Tags: , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,