Though the deal may be a good one for Intel and Cray, some analysts don’t think all the players in the HPC ecosystem will be happy about it.
“This is pretty bad news for AMD,” said David Kanter of Real World Technologies.
Bolding, however, was adamant that the “strong partnership” with Intel would not affect Cray’s relationship with either AMD or Nvidia, both of whose chips it uses in its current and upcoming systems.
Cray is currently building out its Titan supercomputer based on the Cray X6 at Oakridge, which uses both AMD and Nvidia chips, while its Blue Waters system is based on the Cray Xe6 and is also largely dependent on AMD processors.
“It may not impact the relationship, but the systems AMD had were largely fabric driven, so if the fabric is now an Intel fabric, it’s not good news [for AMD],” said Kanter, adding that it was a “big converged interconnect play” by Intel.
An Intel spokesman admitted that the deal gave the firm access not only to interconnect intellectual property (IP) but also to “a world class team of Cray HPC interconnect veterans,” including those responsible for developing the much lauded Gemini interconnect.
Intel has set itself a target of hitting ExaFLOP/s performance by 2018, which would require a scalable system fabric. “The technology and team from Cray provides important intellectual property and resources for this development,” said the spokesman.
“This deal does not affect our current Infiniband product plans and at this moment we don’t disclose future product plans related to acquired assets,” he added.
Meanwhile, Cray is currently working on “Cascade” – its first supercomputer based on next-generation Intel Xeon processors.
Bolding said Cray would also continue to work with Intel on how to differentiate on any future interconnects that might be developed, while it would switch its own focus to other areas, like software.
“If interconnects are being incorporated into processors, we want to look at other areas where we can differentiate,” Bolding said, noting that software now made up 60 percent of the firm’s investment.
“We want to invest in long term differentiation and technology that can help us both in the short term and past the 2016 timeframe, because we want to continue to build the largest, most scalable supercomputers in the world. This doesn’t change that goal,” Bolding said.
In terms of software differentiation, Cray says it has its compiler team working on accelerator based compilers, as well as other programming tools to aid acceleration. “We’re using those with Nvidia today but we anticipate using them with other accelerators in the future,” Bolding added.
In recent times, even in downturn Intel has done some smart purchases. I feel this deal is win-win for both the Intel and Cray. I feel this deal is more in favour of Cray because they still keep the IP rights which means they can build on those IP's.
I am not sure what is Cray's advantages are in their interconnects are... at the base band, for a given signaling frequency (& wire gauge in case of Copper), the latencies are comparable between IB and Cat6A. The hardware-dependent latencies are a small fraction of the overall latency figures.
On the mother boards, it is a different story -perhaps this is where Cray brings in some advantages? It is not clear from the article...
"Bolding, however, was adamant that the “strong partnership” with Intel would not affect Cray’s relationship with either AMD or Nvidia..", this is like telling your girlfriends that marriage will not affect their relationship.
I'm sure that Intel did not enjoy that the X6 was filled with AMD cpus. This could be looked at similar to sponsoring a cycling team (which can sometimes cost in the neighborhood of 50 million). Plus you get technology and IP on top of it! Cray doesn't really lose much, presumably, since they can still use the technology...
In the long term, I can see Intel extending its on chip ring bus and QPI processor bus with hooks into the system clustering interconnect to provide performance advantages and extend its reach into systems (i.e. platform) technologies. Such hooks could make its processors and networking chips preferred solutions in HPC systems. Look out Mellanox!
This is a very great business technique Cray is getting benefited two ways getting more money and simultaneously not reducing the overhead without completely loosing them, its really a great work done by Cray.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.