PCI Express and AMD Technology

Overview

PCI Express® (PCIe®) is a general-purpose serial interconnect suitable for a broad range of applications across communications, data center, enterprise, embedded, test & measurement, military, and other markets. It can be used as a peripheral device interconnect, a chip-to-chip interface, and as a bridge to many other protocol standards.

AMD provides high-performance, low-power integrated blocks for PCI Express in many devices.

AMD also provides DMA and bridge hard and soft IP blocks that utilize the integrated block for PCI Express, boards with PCI Express connectors, connectivity kits, reference designs, drivers, and tools to make it easy to implement PCIe-based designs.

Select a device family from the sidebar navigation tree to review the AMD solution portfolio for PCI Express.

AMD Versal Adaptive SoC Solutions

Systems leveraging implementations of the PCI Express® specification are pervasive in data center, communications, and embedded applications. The multiple connectivity options available in the Versal™ architecture directly support users’ needs to focus on their core competency while leveraging advanced standards-based interfaces. The integrated blocks for PCI Express in the Versal architecture offer premium performance levels with ease of use and efficiency over fully soft IP solutions. The Versal architecture integrates five types of integrated blocks for PCI Express:

• MDB5: integrated block for PCI Express Rev. 5.0 with DMA/bridge
• CPM5: integrated block for PCI Express Rev. 5.0 with DMA/bridge
• PL PCIE5: integrated block for PCI Express Rev. 5.0
• CPM4: integrated block for PCI Express Rev. 4.0 with DMA/bridge
• PL PCIE4: integrated block for PCI Express Rev. 4.0

MDB5, CPM5, PL PCIE5, CPM4, and PL PCIE4, in conjunction with available GTYP and GTY transceivers, enable interface operation at specification-defined data rates. These range from 2.5 GT/s per lane with one lane (Gen1x1) up to their rated maximum link configurations, with some integrated blocks reaching 32 GT/s per lane with 8 lanes (Gen5x8) and 16 GT/s per lane with 16 lanes (Gen4x16).

The table below is a summary of key characteristics of the integrated blocks for PCI Express in the Versal architecture. Refer to the Versal Architecture and Product Data Sheet: Overview (DS950) for additional information on available resources and capabilities based on orderable device, package, and speed/voltage grade combinations.

MDB5, CPM5, PL PCIE5, CPM4, and PL PCIE4 can be used at reduced link configurations to optimize product designs for cost, support novel form factors, alleviate board complexity, and shrink power budgets. Additionally, reduced link configurations can support lower programmable logic resource utilization, depending on the nature of any soft IP solutions used to expand the application-level capabilities of these integrated blocks. CPM5, PL PCIE5, CPM4, and PL PCIE4 can be configured as Endpoint (EP), Root Port (RP), and switch port types. MDB5 can be configured as Endpoint (EP) and Root Port (RP) port types.

Pre-verified high-performance DMA/bridge subsystems for the integrated blocks for PCI Express in the Versal architecture enable users to focus design investment in their areas of greatest value.  Available DMA/bridge subsystems include:

• MDB5 contains two controllers for PCI Express and integrates two instances of a DMA/bridge subsystem. Use of the integrated DMA is optional, and each instance is independently customizable. The DMA subsystems provide a tight coupling to integrated processors and the AXI interconnect. The subsystems also include bridge functionality. Data can be moved with memory-mapped techniques, including the programmable network on chip (NoC).
• CPM5 contains two controllers for PCI Express and also integrates two instances of a QDMA/bridge subsystem. Use of the integrated DMA is optional, and each instance is independently customizable. The QDMA subsystems provide scalable queue-based DMA for moving enormous volumes of data with low latency, plus support for multiple physical and virtual functions commonly required by enterprise class products. Data can be moved with memory-mapped techniques, including the programmable network on chip (NoC), or with streaming techniques, into the Versal adaptive SoC’s programmable logic. The subsystems also include bridge functionality to the AXI interconnect.
• CPM4 contains two controllers for PCI Express and also integrates one instance of a QDMA/XDMA/bridge subsystem. Use of the integrated DMA is optional, and when used, can be configured as a QDMA subsystem like that in the CPM5 or as an XDMA subsystem. Data can be moved with memory-mapped techniques, including the programmable NoC, or with streaming techniques, into the Versal adaptive SoC’s programmable logic. The subsystem also includes bridge functionality to the AXI interconnect.
• PL PCIE5 and PL PCIE4 are individual controllers for PCI Express and are supported by soft IP implementations of DMA/bridge subsystems available from AMD through the Vivado™ Design Suite IP catalog, and additional solutions are available from AMD Adaptive Computing Partners.

For most users, available DMA/bridge subsystems can provide time-saving infrastructure, enabling high-performance turnkey data movement. See the Versal Adaptive SoC CPM DMA and Bridge Mode for PCI Express Product Guide (PG347), the Versal Adaptive SoC DMA and Bridge Subsystem for PCI Express Product Guide (PG344), and the AXI Bridge with DMA for PCIe solution from AMD Adaptive Computing Partner Smartlogic GmbH.

For users seeking to attach their own DMA/bridge subsystem—for preserving their driver and application software investment or to customize or optimize functionality using intimate knowledge of the end application—options are available to design directly to the controllers for PCI Express in the CPM5, PL PCIE5, CPM4, and PL PCIE4 blocks. See the Versal Adaptive SoC CPM Mode for PCI Express Product Guide (PG346) and the Versal Adaptive SoC Integrated Block for PCI Express LogiCORE IP Product Guide (PG343). For ultimate freedom to implement fully custom solutions, AMD offers a soft IP core PHY for PCI Express, enabling designers to attach their own controllers for PCI Express to available GTYP and GTY transceivers. AMD Adaptive Computing Partners, including Fidus Systems, Inc., offer development and consulting services with unique value propositions to differentiate and accelerate design with Versal adaptive SoCs.

UltraScale+ Device Solutions

AMD 16 nm UltraScale+™ devices integrate many essential PCI Express® features required for today’s data center, communications, and embedded applications. UltraScale+ devices use three types of integrated blocks: PCIE4, PCIE4C, and PCIE4CE.

The PCIE4 blocks are compliant to PCI Express Base Specification v3.1 and support up to Gen3x16 and can also be configured for lower link width and speeds. The PCIE4 block does not support Gen4 operation.

PCIE4C blocks are compliant to the PCI Express Base Specification v3.1, supporting up to 8.0 GT/s (Gen3) and compatible with PCI Express Base Specification v4.0, supporting up to 16.0 GT/s (Gen4). PCIE4C blocks are also compliant with CCIX Base Specification v1.0 Version 0.9, supporting speeds up to 16.0 GT/s. PCIE4C blocks support up to 16 lanes at Gen3 or up to 8 lanes at Gen4 and can be configured for lower link widths and speeds to conserve resources and power.

PCIE4CE blocks are compliant to the PCI Express Base Specification v4.0 supporting up to 16.0 GT/s (Gen4). PCIE4CE blocks support up to 16 lanes at Gen3 or up to 8 lanes at Gen4 and can be configured for lower link widths and speeds to conserve resources and power. Some devices contain only one revision of integrated block, whereas other devices provide a mix of revisions of integrated blocks.

All integrated blocks for PCIe® in the UltraScale™ architecture can be configured as Endpoint or Root Port. The Root Port can be used to build the basis for a Root Complex to allow custom chip-to-chip communication via the PCI Express protocol and to attach ASSP Endpoint devices, such as Ethernet controllers, Fibre Channel HBAs, or NVMe SSDs, to the FPGA, MPSoC, or RFSoC.

The Integrated Block for PCI Express supports:

• Capability up to Gen3x16 in all revisions of integrated blocks, plus compatibility with Gen4x8 in devices offering PCIE4C blocks, plus compliance with Gen4x8 in devices offering PCIE4CE blocks. For specific link widths and rates that are supported, consult the IP product guide for the desired IP (UltraScale+ Devices Integrated Block for PCI Express Product Guide (PG213), DMA/Bridge Subsystem for PCI Express Product Guide (PG195), QDMA Subsystem for PCI Express Product Guide (PG302)). For details about PCIE4C block compatibility with Gen4x8, see UltraScale+ Devices Integrated Block for PCI Express Product Guide (PG213). Available link widths can depend on available transceiver counts and package pinout.
• Four physical functions and 252 virtual functions in support of single root I/O virtualization (SR-IOV) to share I/O resources.
• Expanded number of tags (256) to support more requests, enabling improvements in overall system performance.
• Integrated MSI-X tables.

Available soft IP implementations of DMA/bridge include:

• The AMD QDMA IP subsystem is our leading DMA solution for PCIe. The QDMA provides support for SR-IOV with multiple physical and virtual functions with scalable queues. The QDMA also provides AXI bridge functionality. See QDMA Subsystem for PCI Express Product Guide (PG302) for more details.
• The AMD XDMA IP subsystem is our legacy DMA solution for PCIe, widely used by customers. The XDMA also provides AXI bridge functionality. See DMA/Bridge Subsystem for PCI Express Product Guide (PG195) for further details.
• Solutions from AMD Adaptive Computing Partners such as AXI Bridge with DMA for PCIe, High Channel Count DMA for PCIe, and Multichannel DMA for PCIe by Smartlogic GmbH.

AMD Zynq™ UltraScale+™ MPSoCs and RFSoCs have an additional available integrated block type for PCIe including a DMA/bridge, tightly coupled with the integrated processing system, supported by a root port driver for embedded Linux®. Please refer to Zynq UltraScale+ Device Technical Reference Manual (UG1085) for further details. For ultimate freedom to implement fully custom solutions, AMD also provides a soft IP core PHY for PCI Express, enabling designers to attach their own controllers for PCI Express to available transceivers. AMD Adaptive Computing Partners, including Fidus Systems, Inc., offer development and consulting services with unique value propositions to differentiate and accelerate design with UltraScale+ devices.

Supporting Products and Technologies

UltraScale Solutions

AMD 20 nm UltraScale™ devices integrate many essential PCI Express® features required for today’s data center, communications, and embedded applications.

The Integrated Block for PCI Express supports:

• Native Gen3x8* Integrated PCIe® block
  
• 64 tags for PCIe requests
• Multi-vector MSI for up to 32 vectors and MSI-X
• See UltraScale Devices Gen3 Integrated Block for PCI Express LogiCORE IP Product Guide (PG156) for further details

Available soft IP implementations of DMA/bridge include:

• The AMD XDMA IP subsystem is our production PCIe DMA solution, widely used by customers. See DMA/Bridge Subsystem for PCI Express Product Guide (PG195) for more information.
• AMD provides a production AXI Memory Mapped to PCI Express Bridge IP. See AXI Bridge for PCI Express Gen3 Subsystem Product Guide (PG194) for further details.
• Solutions from AMD Adaptive Computing Partners such as AXI Bridge with DMA for PCIe, High Channel Count DMA for PCIe, and Multichannel DMA for PCIe by Smartlogic GmbH.

*For specific link widths and speeds that are supported, see the appropriate product guide for the desired IP.

AMD Adaptive Computing Partners including Fidus Systems, Inc. offer development and consulting services with unique value propositions to differentiate and accelerate design with UltraScale devices.

Supporting Products and Technologies

7 Series Solutions

AMD 28 nm 7 Series devices integrate many essential PCI Express® features required for today’s data center, communications, and embedded applications.

The integrated Block for PCI Express supports:

• Native Gen3x8* integrated PCIe® block
• 64-bit and 128-bit data widths
• See Virtex-7 FPGA Gen3 Integrated Block for PCI Express v4.3 Product Guide (PG023) and 7 Series FPGAs Integrated Block for PCI Express LogiCORE IP Product Guide (PG054) for further details

Available soft IP implementations of DMA/bridge include:

• The AMD XDMA IP subsystem is our production PCIe DMA solution, widely used by customers. See DMA/Bridge Subsystem for PCI Express Product Guide (PG195) for more information.
• AMD provides a production AXI Memory Mapped to PCI Express Gen2 IP. See AXI Memory Mapped to PCI Express (PCIe) Gen2 v2.9 LogiCORE IP Product Guide (PG055) for further details.
• Solutions from AMD Adaptive Computing Partners such as AXI Bridge with DMA for PCIe, High Channel Count DMA for PCIe, and Multichannel DMA for PCIe by Smartlogic GmbH.

*For specific link widths and speeds that are supported, see the appropriate product guide for the desired IP.

AMD Adaptive Computing Partners, including Fidus Systems, Inc., offer development and consulting services with unique value propositions to differentiate and accelerate design with 7 Series devices.

Supporting Products and Technologies