Intel At SC15: Scalable Systems Framework, 3D XPoint Memory, Intel Omni-Path Architecture and More


Intel’s primary focus during SC'15 was the Scalable Systems Framework and the Omni-Path Architecture. The point of these initiatives is to offer a single vendor solution in making giant computing clusters. The downside of being exclusive to Intel is of course, arguably worth it, since this means that Intel will take out most of the headache that comes with establishing a supercomputing cluster. To do this, the company is offering a wide mix of software and hardware that makes up its Scalable Systems Framework ideology.

Intel talks about its Scalable Systems Framework at SC15

The hardware side naturally includes co-processors such as the Xeon Phi platform and processors from the Xeon Side. This is complimented by the Omni-Path architecture which is claimed to be a “disruptive innovation”. It will result in 17% lowered latency and 16% higher messaging rate (than its widely used competitor InfiniBand). It offers upto 26% more compute nodes and allegedly provides better price-performance than Infiniband. Not only that but it uses 60% less power than Infiniband – attributed to the fact that it uses a 48 port chip architecture. Finally talking about error correction, Intel mentioned that there is no latency penalty for error detection with its Packet Integrity Proteciton technology – allowing cluster users to have complete peace of mind without compromising on reactivity.

Intel has previously demoed its 3D Xpoint memory and Optane based SSDs – both of which aim to be part of its framework in the future. The 3D Xpoint memory technology has 1000x the endurance of NAND flash and is 10 times denser and in some cases upto 1000x faster as well. The 3D XPoint memory will be available in market during next year and will revolutionize the tech industry with the latest 3D XPoint based Optane SSDs and DIMMs.

  • The Optane SSD was able to achieve 7.2x times more IOPS at low queue depth and upto 5.21 times the IOPs of conventional SSDs at high queue depths.
  • An Optane Technology based SSD has 10x times the density of conventional SSD drives.
  • The marketing material also claims it is 1000x faster than the competition available on the market but it isn’t clear to what exactly they are referring to – a good guess would be latency, as opposed to bandwidth.
  • Optane SSDs will have 1000x the endurance – which, if true, should mean the device has virtually unlimited life span for practical purposes.

All Optane based devices will feature a cross point array structure, which consists of perpendicular connectors connecting around 128 Billion memory cells (16 Gigabytes per chip).  This “3D” method is the reason why Optane based devices have 10x times the density of conventional solutions. Like DRAM, Optane memory is stackable in nature. One of the biggest changes in this technology however is that it eliminates the need for transistors – accessing the memory cells by varying the voltage sent to the particular sector. Basically, using the bulk of the material itself.

Intel and Altera have also in collaboration recently released the Stratix 10 solutions which are also designed for the HPC market. The technology will use Altera’s Stratix chips (SoCs and FPGAs) and combine them with Intel’s EMIB fabric. The result is a device that can handle some specific work loads exceptionally well. Whether this will have any impact in the graphics segment of the industry – is anyone’s guess. Most of the remaining presentation consisted of partnership disclosures as well as usual marketing material. The press deck provided at the event is given below: