PEX8748-CA80BC - 48-Lane Gen 3 PCIe Switch | Broadcom
MPN: PEX8748-CA80BC ✓ 啟用中PEX8748-CA80BC 概覽
A PCI Express switch is a semiconductor device that routes high-speed serial lanes between one host or upstream connection and multiple downstream endpoints. Within the interface hierarchy, the PEX8748 is a PCI Express interface IC, network-interface switch, and multi-host fabric controller. It provides non-blocking, high-bandwidth connectivity while allowing multiple PCI Express domains to share switching resources.
The PEX8748 combines 48 Gen 3 lanes with 12 configurable ports. This lane-to-port arrangement supports dense server and storage topologies that require more fan-out than a conventional endpoint controller. Its FCBGA-676 package integrates a large number of high-speed differential I/O connections into a compact 27 x 27 mm area, making careful ball assignment, reference-plane continuity, and thermal design essential.
The switch is intended for PCIe Gen 3 fabrics spanning multiple hosts and endpoints. Its non-blocking architecture is suited to server consolidation, storage aggregation, communications equipment, and graphics platforms. Configuration flexibility also supports peer-to-peer arrangements and failover topologies, allowing redundant paths or workload distribution across multiple downstream devices.
Typical uses include high-performance computing servers, AI inference platforms, NVMe storage fabrics, GPU acceleration systems, and industrial communication equipment. The 48-lane capacity and 12-port configuration are particularly useful where many PCIe endpoints must converge through a shared switching fabric without relying on separate host-to-endpoint switching paths.
The main design trade-off is thermal and routing complexity. An 8.0 W typical-power device in a 27 x 27 mm FCBGA requires a multilayer PCB with an uninterrupted ground plane, carefully controlled high-speed impedance, and a thermal strategy suited to sustained compute workloads. The package also requires a 676-ball PCB footprint and exact land-pattern compliance; board changes are not practical after layout release.
This product page consolidates the verified electrical identity, package information, multi-host application role, comparable switch families, and practical implementation considerations for engineers evaluating the Broadcom PEX8748-CA80BC for dense PCIe Gen 3 systems.
直接替換替代品,適用於 PEX8748-CA80BC — 相同封裝,腳位對腳位相容。需重新設計 PCB 的不同封裝零件不列入。
此系列的變體
與下列產品可直接替換的同系列型號: PEX8748-CA80BC (相同外形與佔位) — 差異在於 Package Type.
PEX8749-CA80BC-G
✅ 直接替換 ⚠️ 規格未經驗證📋 參考替代品(不在目錄中)
PEX8747-CA80BC-G
✅ 直接替換 ⚠️ 規格未經驗證📋 參考替代品(不在目錄中)
PEX8745-CA80BC-G
✅ 直接替換 ⚠️ 規格未經驗證📋 參考替代品(不在目錄中)
PEX8744-CA80BC-G
✅ 直接替換 ⚠️ 規格未經驗證📋 參考替代品(不在目錄中)
PEX8743-CA80BC-G
✅ 直接替換 ⚠️ 規格未經驗證📋 參考替代品(不在目錄中)
PEX8748-CA80BC 最大額定值與電氣特性
| Product Type | PCI Express switch IC |
| PCI Express Generation | Gen 3 |
| Total Lane Count | 48 lanes |
| Port Count | 12 ports |
| Switching Architecture | Multi-host, non-blocking PCI Express switch |
| Typical Power Consumption | 8.0 W |
| Package Type | FCBGA-676 |
| Package Pin Count | 676 balls |
| Package Dimensions | 27 x 27 mm |
| 安裝類型 | Surface mount |
| Signal Standard | PCI Express |
| Supported Traffic Patterns | Fan-out, aggregation, peer-to-peer, and multi-host interconnection |
| RoHS 狀態 | Compliant |
PEX8748-CA80BC 27 x 27 mm 腳位配置指南
PEX8748-CA80BC(27 x 27 mm 封裝)的腳位配置。 此類比元件具有輸入、輸出與電源腳位。偏移歸零、補償與致能腳位配置請參閱製造商資料表。適用於訊號調理與放大器電路。
無詳細腳位定義資料: PEX8748-CA80BC.
請參閱 資料表 以查看完整腳位配置。
典型應用
PEX8748-CA80BC 適用於 6 應用s: Multi-Host Server PCI Express Fabrics, AI Inference and GPU Acceleration Systems, NVMe Storage Aggregation, High-Performance Computing Interconnects, Communications Infrastructure Equipment, Graphical Processing and Workstation Systems.
Multi-Host Server PCI Express Fabrics
The PEX8748-CA80BC is well suited to servers that connect multiple hosts or endpoint groups through a shared PCI Express fabric. Its 48 Gen 3 lanes and 12 configurable ports provide the fan-out needed for dense accelerator, storage-controller, and network-interface arrangements. The switch architecture supports non-blocking traffic paths, aggregation, and peer-to-peer transfers while keeping the system topology centralized in one 27 x 27 mm FCBGA device. Engineering teams should model traffic classes and failover routes during configuration, then validate throughput and latency under realistic mixed workloads.
推薦
AI Inference and GPU Acceleration Systems
The PEX8748-CA80BC can form the switching backbone of an AI inference platform that connects accelerators, host processors, and memory or storage controllers. Forty-eight Gen 3 lanes distributed across 12 ports allow multiple endpoint groups to operate behind a scalable fabric rather than requiring independent host-to-device paths. The non-blocking switching behavior supports concurrent accelerator traffic and peer-to-peer movement of intermediate data. The design must account for the documented 8.0 W typical switch power, requiring a multilayer PCB, continuous ground planes, and system-level thermal testing near dense accelerator hardware.
推薦
NVMe Storage Aggregation
The PEX8748-CA80BC provides a practical way to aggregate multiple PCI Express storage paths behind a shared switch fabric. Its 12 ports and 48 lanes allow controllers or host domains to be organized into a high-density topology, while non-blocking operation helps maintain concurrent storage traffic. For high-latency-sensitive workloads, validate queue behavior, congestion controls, and failover behavior rather than relying only on theoretical lane bandwidth. Board implementation should use the manufacturer reference design, maintain continuous reference planes, and reserve generous thermal capacity for sustained transfer activity.
推薦
High-Performance Computing Interconnects
In high-performance computing systems, the PEX8748-CA80BC can provide scalable fan-out and aggregation between multiple compute nodes, accelerators, and I/O devices. Its 48-lane Gen 3 fabric and 12-port configuration suit systems that need a shared, non-blocking PCI Express switching layer for tightly coupled accelerator workloads. Peer-to-peer traffic paths can reduce unnecessary routing through host software stacks when the configuration and firmware support them. Integration requires high-speed layout discipline, comprehensive lane testing, and thermal design capable of handling the switch's 8.0 W typical operating reference in sustained workloads.
推薦
Communications Infrastructure Equipment
The PEX8748-CA80BC can support communications equipment that needs redundant or consolidated PCI Express connectivity across line cards, processing modules, and service processors. Twelve configurable ports and 48 lanes provide enough switching density for multi-host arrangements, while aggregation and peer-to-peer modes can organize traffic according to the system's availability requirements. Failover paths should be tested at both the physical-link and configuration layers to ensure that the intended traffic is restored after a link or endpoint failure. The 676-ball package requires high-layer-count PCB construction and careful power-delivery and thermal implementation.
推薦
Graphical Processing and Workstation Systems
The PEX8748-CA80BC is suitable for graphics and workstation platforms that need multiple PCI Express endpoints behind a high-density switching fabric. Its 48 Gen 3 lanes and 12 ports can support endpoints such as graphics accelerators, storage controllers, and high-speed capture devices while preserving non-blocking paths between active devices. Peer-to-peer configurations may improve data movement between compatible endpoints when supported by the complete platform. Because this is a 676-pin FCBGA with an 8.0 W typical-power reference, designers must use the manufacturer's land pattern and reference circuitry and should verify airflow, board warpage, and connector-dependent mechanical constraints.
推薦
推薦產品摘要
工程參考資料,適用於 PEX8748-CA80BC — 比較、設計指引及合規資訊。
選型指南
與替代品比較
| 參數 | 本產品 |
|---|---|
| Product Type | PCI Express Gen 3 switch IC |
| Brand | Broadcom |
| Package | FCBGA-676 (27 x 27 mm) |
| Total Lane Count | 48 lanes |
| Port Count | 12 ports |
| Typical Power | 8.0 W |
| Switching Architecture | Multi-host, non-blocking |
| Supported Traffic Patterns | Fan-out, aggregation, peer-to-peer, and failover |
| RoHS Status | Compliant |
主要差異化特點
- Higher downstream fan-out than the compared PEX8747 configuration (對比PEX8747-CA80BC)
- Multi-host non-blocking switching architecture (對比PEX8747-CA80BC)
- Broader endpoint-fan-out capacity for accelerator fabrics (對比PEX8796-AB80BI)
設計說明
Design the PCB and enclosure for the documented 8.0 W typical-power reference before finalizing placement. Provide uninterrupted ground planes beneath the FCBGA-676 area, minimize thermal resistance through the board stack, and use the Broadcom reference design to identify thermal-via and pad-array requirements. Validate adjacent component temperatures and junction behavior under sustained endpoint traffic rather than relying on a lightly loaded board. Actual power varies with port activity and traffic, so correlation testing should cover the intended maximum-use case.
Use the manufacturer FCBGA-676 land pattern and ball map without substituting a generic BGA footprint. Every differential PCI Express pair must follow the prescribed length-matching, spacing, and impedance rules, while supply and ground balls require the shortest practical connections to low-inductance decoupling. Maintain continuous reference planes beneath high-speed lanes and avoid splits or routing voids that create impedance discontinuities. PCB stackup should be selected before placement because it determines achievable impedance and loss.
Do not assume that two Broadcom switches in FCBGA-676 packages are interchangeable. The PEX8747-CA80BC is listed with 5 ports versus 12 ports for the PEX8748-CA80BC, and package equality does not prove identical ball assignments, power sequencing, firmware behavior, or port configuration. For any replacement project, compare the manufacturer ball map, electrical limits, initialization sequence, reference schematic, and firmware configuration. PCB respin may be required if any assignment or function differs.
Treat all 48 Gen 3 lanes as a simultaneous high-speed signal-integrity problem. Follow the reference design's stackup and via geometry, control lane skew across the complete route, and verify differential impedance with the actual PCB material set. Keep test points and stubs tightly controlled, and maintain proper reference-plane return paths for every high-speed pair. Conduct eye, jitter, compliance, and interoperability testing with realistic endpoint traffic because individual lane testing does not reveal every multi-lane or power-integrity interaction.
合規資訊
RoHS compliance is stated for the PEX8748-CA80BC. AEC-Q100 status is recorded as not applicable because this is a general-purpose PCIe switch rather than an automotive-qualified device. No additional compliance claims are made without explicit manufacturer documentation.