Specification Now Available to Member Companies

Cache Coherent Interconnect for Accelerators (CCIX)

What is CCIX?

CCIX stands for Cache Coherent Interconnect for Accelerators. This is a new chip-to-chip interconnect that is being defined by the CCIX consortium that will allow two or more devices to share data in a cache coherent manner.

What companies are part of this group?

  • AMD
  • Amphenol Corporation
  • ARM
  • Arteris Inc.
  • Avery Design Systems
  • Broadcom Limited
  • Atos Technologies
  • Cadence Design Systems
  • Cavium Inc.
  • Huawei
  • Integrated Device Technology, Inc.
  • IBM
  • Keysight Technologies
  • Mellanox Technologies
  • Micron Technology Inc.
  • Netspeed Systems
  • Qualcomm Technologies Inc.
  • Red Hat Inc.
  • Synopsys, Inc.
  • Teledyne LeCroy
  • Texas Instruments
  • TSMC
  • Xilinx, Inc. 

Why is this communication standard needed and what problem does it solve?

CCIX is a single standard supported by accelerator, processor, and interconnect companies that can deliver scalable performance for very high bandwidth, low latency, and coherent data movement. The availability of CCIX technology from multiple vendors will give system designers the flexibility in choosing the right combination of heterogeneous components to address the specific needs of their systems.

CCIX allows the accelerator and interconnects to support the advanced coherency capabilities simplifying the use model for the accelerator. The single standard helps improve I/O bandwidth, reduce latency, and improve performance.CCIX will allow components to access and process data irrespective of where it resides, without the need for complex programming environments. This ability to access data coherently dramatically improves performance and usability by enabling a driver-less and interrupt-less usage model.

What are the target markets and applications for acceleration for the new CCIX standard?

While CCIX’s primary focus is in acceleration in Data Center, Cloud, Server and Embedded Computing applications, it can also be used in other markets and applications. 

What is unique about CCIX?

CCIX brings together multiple instruction-set-architectures (ISAs) to a single cache-coherent interconnect standard. It supports scalable performance to very high bandwidths while reducing latency. CCIX also allows the accelerator and interconnects to support the advanced coherency capabilities such as hosting memory in addition to caching the processor’s host memory.

Does CCIX replace existing server interconnects?

CCIX can leverage existing server communication technology. What this means is that it can enhance the capabilities of existing server interconnects to enable coherency, reduce latency and increase bandwidth. But does it replace anything?

What do you mean by “device”, what is a device?

Examples of devices are processors, FPGAs, network adapters, intelligent interconnects, memory or any other system component.

What does cache coherent mean?

Cache coherent means that both devices that share a CCIX link can cache (keep a local copy) of the other devices memory. In addition, any changes that one device makes to that data made visible by hardware to the other device without requiring explicit buffer copies or software intervention.

What is the benefit of cache coherency?

Cache coherency has built-in hardware capability that allows devices to mutually work on data without software intervention. This avoids the overhead of software having to copy data back and forth between the devices. This leads to the real benefit which is simplicity in the software model for data-sharing. With CCIX, software does not need to manage data movement. This results is significant performance improvement as the data-movement happens automatically and seamlessly without software intervention. This new software model does not require runtime drivers, DMAs, or interrupts.

Why is CCIX a low latency interface?

The CCIX link layer is optimized for latency. This is done by avoiding store-forward and ordering bottlenecks. At the application level the latency of data access is reduced by allowing the hardware to move data seamlessly without software intervention.

Why is CCIX a high-bandwidth interface?

CCIX increase the bandwidth beyond today’s interconnects. CCIX supports much higher bandwidth than the existing server chip-chip interconnects as an example. CCIX also optimizes the data transfer across the link to increase the utilization of the link resulting in improved bandwidth.

How does a company join CCIX?

The CCIX agreement has provisions to include additional observing members. The inclusion of these members has to be voted upon and agreed to by the existing members. The new member will have to agree to the clauses in the agreement and sign it. If companies express interest in signing up, refer them to veroniqu@xilinx.com

When will a draft of the specification be available to the outside world?

The CCIX specification is available immediately to members.

Can CCIX support switches?

Yes CCIX devices can be connected over a switch topology.

Can CCIX be used to connect to memory?

CCIX could be used to connect any two devices in the server systems. It may be applicable to connect to specialized memory as well.

Can CCIX connect multiple accelerators in a glue-less fashion?

Yes CCIX can connect multiple accelerators.

What is meant by existing server ecosystem?

Existing server ecosystems include connectors, cables, discovery mechanisms, firmware, and operating systems.

Does CCIX require system software changes?

CCIX will not need modification to the OS. CCIX changes are limited to the platform firmware.

When will silicon partners have CCIX enabled in their systems?

Silicon partners will have initial products starting in 2017.

What bandwidths will CCIX support?

CCIX supports speeds up to 25Gbps today.

How, specifically, will the tech industry benefit from CCIX?

CCIX is optimized to allow processors based on different instruction set architectures to extend their cache coherency to accelerators. These highly capable accelerators become “first class citizens” in the processor system, in that they become, from a software programming view, similar to a core or accelerator within the on-die coherency domain of a multi-core processor. The availability of CCIX technology from multiple vendors will give system designers the flexibility in choosing the right combination of heterogeneous components to address the specific needs of their systems. CCIX has chosen to use PCI Express® as its first transport layer with additional speeds of 25Gbps and higher. CCIX anticipates additional transport layers to be added in the future. Using PCI Express® to transport the CCIX coherency protocol eases the implementation of CCIX in processors and accelerators as well as the deployment of CCIX technology in existing servers by leveraging today’s existing hardware and software infrastructure.

How does CCIX relate to GenZ?

GenZ is a new data access technology that enables memory operations to direct attach and disaggregated memory and storage. CCIX extends the processor’s coherency domain to heterogeneous components. These heterogeneous “nodes” would then get access to the large and disaggregated storage and memory through the GenZ fabric.