AN 1003: Multi Memory IP System Resource Planning: for Agilex™ 7 M-Series FPGAs

Download PDF
ID 788295
Date 2/07/2025
Public

Visible to Intel only — GUID: syj1696532696619

Ixiasoft

View Details

Document Table of Contents
Document Table of Contents x
Answers to Top FAQs 1. About This Application Note 2. Component Bandwidth Projections and Limitations 3. Resource Planning for Agilex™ 7 M-Series FPGAs 4. Factors Affecting NoC Performance 5. Debugging the NoC 6. Document Revision History of AN 1003: Multi Memory IP System Resource Planning for Agilex™ 7 M-Series FPGAs
1. About This Application Note x
1.1. Preliminary Guidelines 1.2. Document Prerequisites 1.3. Terminology for Agilex™ 7 M-Series FPGAs
2. Component Bandwidth Projections and Limitations x
2.1. Memory Controller Efficiency 2.2. Hard Memory NoC Bandwidth 2.3. Overall Bandwidth for NoC Systems
2.1. Memory Controller Efficiency x
2.1.1. HBM2E UIB Controller Efficiency 2.1.2. External Memory Interface Controller Efficiency
2.2. Hard Memory NoC Bandwidth x
2.2.1. Horizontal NoC Bandwidth 2.2.2. Initiator NoC Bandwidth 2.2.3. Target NoC Bandwidth 2.2.4. Initiator Target Link Bandwidth
3. Resource Planning for Agilex™ 7 M-Series FPGAs x
3.1. Hard Memory NoC Resource Planning Overview 3.2. I/O Bank Blockage 3.3. Planning Avalon® Streaming Utilization 3.4. Planning for Initiator Blockage Impact from GPIO, LVDS SERDES, and PHY Lite Bypass Mode 3.5. Planning NoC PLL and I/O PLL 3.6. Pin Planning for HPS EMIF 3.7. Planning for an External Memory Interface 3.8. Planning for HBM2E 3.9. Planning for the Fabric NoC 3.10. Planning for AXI4-Lite 3.11. Planning NoC and Memory Solution Clocks
3.2. I/O Bank Blockage x
3.2.1. Top Hard Memory NoC 3.2.2. Bottom Hard Memory NoC
3.10. Planning for AXI4-Lite x
3.10.1. Single AXI4 Lite Initiator for Each Horizontal NoC 3.10.2. AXI4-Lite Initiator for HBM2E 3.10.3. AXI4-Lite Initiator for EMIF
3.11. Planning NoC and Memory Solution Clocks x
3.11.1. Clocking without the Fabric NoC 3.11.2. Clocking with the Fabric NoC
4. Factors Affecting NoC Performance x
4.1. Recommended Performance Tuning Procedure 4.2. NoC Initiator and Target Clock Rate 4.3. Recommended NoC Design Topologies 4.4. Traffic Access Pattern and Memory Controller Efficiency 4.5. Traffic Access Pattern Due To Multiple Traffic Flows 4.6. Transaction Size 4.7. Congestion Interaction 4.8. Bandwidth Sharing At Each Switch 4.9. Exceeding NoC Bandwidth Limits 4.10. Maximum Number of Outstanding Transactions 4.11. QoS Priority 4.12. AxID 4.13. Example: 2x2 HBM Crossbars 4.14. Example: 16x16 Crossbar
4.3. Recommended NoC Design Topologies x
4.3.1. 1:1 Connectivity with EMIF and HBM 4.3.2. HBM Crossbars 4.3.3. Connectivity with HPS
5. Debugging the NoC x
5.1. Debugging NoC AXI Transaction Issues 5.2. Debugging NoC Bit Errors 5.3. Debugging NoC Performance Issues
5.3. Debugging NoC Performance Issues x
5.3.1. Measuring Performance with the Signal Tap Logic Analyzer 5.3.2. Measuring Performance with the Performance Monitor 5.3.3. Performance Debug Steps
Answers to Top FAQs
1. About This Application Note
2. Component Bandwidth Projections and Limitations
3. Resource Planning for Agilex™ 7 M-Series FPGAs
4. Factors Affecting NoC Performance
5. Debugging the NoC
6. Document Revision History of AN 1003: Multi Memory IP System Resource Planning for Agilex™ 7 M-Series FPGAs

4.5. Traffic Access Pattern Due To Multiple Traffic Flows

A consideration for traffic access pattern in Agilex™ 7 M-Series FPGAs is the combined effect of more than one transaction source at a memory controller.

The NoC readily permits multiple initiators to issue transactions to a single target. Although a single initiator may be issuing well-ordered transactions, the combined effect of multiple initiators at the target can be transactions that are not well ordered, and require activating rows more frequently.

To mitigate this problem, limit the number of independent sources that access an individual memory, or synchronize accesses to provide a better structured access pattern.

Figure 27. Sequential Address Access Issued by Separate Initiators—Amounts to Non-Sequential Access at Target


Level Two Title