Intel has announced general availability of the latest generation of Xeon Phi processors aimed at high-performance computing (HPC) applications, and unveiled HPC Orchestrator, a pre-integrated software stack that customers can use to power HPC clusters and a key part of its Scalable System Framework (SSF) architecture.
Announced at the International Supercomputing Conference in Frankfurt, the Knights Landing family is the first of Intel's Xeon Phi many integrated core chips that can function as a standalone processor in its own right, rather than a co-processor for another chip, typically an Intel Xeon.
Knights Landing has been on Intel's HPC roadmap since 2013 and was due to ship last year, but is only now generally available, although Intel said that all the early silicon is already allocated to buyers and volume shipments are not due for several months.
The chips have up to 72 processor cores, each with an architecture similar to Intel's Atom processors, and hence are known as the Xeon Phi 7200 series. The line introduces for the first time 16GB of high-speed memory integrated inside the chip package for low latency access, while each chip can access up to 384GB of DDR4 memory in DIMMs.
Customers will also be able to purchase Xeon Phi chips with integrated Omni-Path, Intel's high-speed interconnect fabric launched last year. Versions with this are not set to be available until October, however.
Intel clearly sees systems using GPU accelerators as a competitor in the HPC space, and the firm claims that the platform offers up to five times the performance of a GPU, and that an Xeon Phi system has already achieved a record result for a single-socket system with the SPECfp_rate2006 benchmark.
There are four models in the Xeon Phi 7200 series at launch. The 7290 is the top-end "Formula One" version with 72 cores, and the most costly, but is unlikely to be available in great quantities, according to Charles Wuischpard, vice president of the Data Centre Group and general manager of the HPC Platform Group at Intel.
"Most early customers are focused on the 7230 and 7250, but we think the 7210 will be the most popular part because it delivers 85 to 90 per cent of the performance at less than half the price of the top-end part," he said.
Intel also officially announced HPC Orchestrator, the firm's own version of the OpenHPC platform, an open source framework for operating HPC clusters developed by the Linux Foundation. This will be available from the fourth quarter of 2016 in three editions targeting progressively larger HPC infrastructure deployments.
The first covers the vast majority of the market, comprising relatively small deployments of just a few racks, while the mid-level edition is geared towards deployments on the scale of the TOP500 list of supercomputers, and the third is a flagship product created for the very largest HPC systems that may have 50,000 to 100,000 nodes.
"An HPC software stack is a number of crucial elements of software that need to be integrated and interoperable. Many of them are open source, but not all of them because there are choices we made in some areas, but everything is evolving and versioning, and one of the challenges is having a stable, working cluster stack that gives you an easy out-of-the-box experience," Wuischpard said.
HPC Orchestrator is a paid-for version of the software that comes with Intel support, in the same way that Intel announced its own paid-for version of the Lustre distributed file system last year offering vital technical support for HPC customers, he added.
Xeon Phi and HPC Orchestrator are parts of Intel's SSF architecture for HPC, along with the Omni-Path high-speed fabric launched last year.
SSF is intended to do for HPC systems what Intel has done in the past for the PC and server platforms in delivering a standard on which others can build. Intel also aims to deliver a solution that can scale better than existing solutions, and serve as a general-purpose platform that can be used for a range of HPC workloads.
"The roadmap is to build this interoperability and greater and greater levels of integration over time, but you add to that reference designs and system compatibility designs, and we're in the process of publishing architectural specifications that will gradually get deeper and richer. But the idea is to provide that curated starting place, a base configuration that should work out of the box based on Intel ingredients," Wuischpard explained.
In the future, this will include new ingredients, such as Intel's 3D XPoint non-volatile memory technology and Silicon Photonics optical interconnect, which has been delayed but Intel expects to see production systems using it this year, according to Wuischpard.
Intel is also now shipping the ‘ninja' developer workstation announced last year. This is basically a tower server with a single Xeon Phi processor that developers can use to build and hone code without taking up valuable run-time on the HPC cluster.
Cotton seedling freezes to death as Chang'e-4 shuts down for the Moon's 14-day lunar night
Fortnite easily out-earns PUBG, Assassin's Creed Odyssey and Red Dead Redemption 2 in 2018
Meteor showers as a service will be visible for about 100 kilometres in all directions
Saturn's rings only formed in the past 100 million years, suggests analysis of Cassini space probe data
New findings contradict conventional belief that Saturn's rings were formed along with the planet about 4.5 billion years ago