Raspberry Pi SoC Architecture & Hardware Models: BCM2712, RP1 & ARM64 Crypto
Created: 2026-11-08 Last updated: 2026-11-08
Executive Summary: The transition from the Raspberry Pi 4 to the Raspberry Pi 5 represents the most radical architectural redesign in the history of the platform. By moving from a monolithic 28nm silicon die (BCM2711) to a disaggregated chiplet architecture—pairing a high-efficiency 16nm Broadcom BCM2712 SoC with a custom Raspberry Pi RP1 I/O Southbridge over PCIe—the hardware delivers a 2.5× to 3× performance surge. Crucially for systems engineers, the Cortex-A76 cores incorporate ARMv8.2-A hardware cryptographic extensions, accelerating AES-GCM and SHA-256 operations by an order of magnitude. This technical reference dissects the silicon topology, memory subsystems, crypto benchmarks, and industrial Compute Module options.
-
BCM2712 & Cortex-A76
Deconstruct the 16nm quad-core out-of-order architecture, 512 KB per-core L2 caches, and shared 2 MB L3 cache pipeline.
-
Hardware Cryptography
Harness native
ARMv8.2-Ainstructions to accelerate TLS termination, WireGuard VPN tunnels, and encrypted storage by 10×. -
The RP1 Southbridge
Understand the custom 40nm I/O controller handling Ethernet, USB 3.0, MIPI, and GPIO via a dedicated 4-lane PCIe 2.0 interconnect.
-
Compute Modules (CM4 / CM5)
Explore industrial mezzanine form factors, on-board eMMC storage, PCIe lane routing, and multi-node cluster carrier boards.
1. Silicon Architecture Evolution: BCM2711 vs. BCM2712
In previous generations (Pi 1 through Pi 4), all CPU logic, memory controllers, and peripheral I/O (USB, Ethernet, GPIO) were packed into a single monolithic silicon die. On the Raspberry Pi 5, the system is divided into specialized dies:
%%{init: {'flowchart': {'wrappingWidth': 600, 'subGraphTitleMargin': {'top': 15, 'bottom': 15}}}}%%
flowchart TB
subgraph BCM2712["🧠 Broadcom BCM2712 SoC (16nm FinFET)"]
direction TB
subgraph CPU_Cluster["Quad-Core Cortex-A76 @ 2.4 GHz"]
C0["Core 0 (512KB L2)"]
C1["Core 1 (512KB L2)"]
C2["Core 2 (512KB L2)"]
C3["Core 3 (512KB L2)"]
L3["Shared 2 MB L3 Cache"]
C0 & C1 & C2 & C3 --> L3
end
GPU["VideoCore VII 3D GPU<br/>(800 MHz Dual 4K60p)"]
MC["LPDDR4X Memory Controller<br/>128-bit Bus (17.0 GB/s)"]
end
subgraph External_PCIe["⚡ PCIe Gen 2/3 x1 Header"]
NVME["M.2 NVMe SSD HAT"]
end
subgraph RP1_Chip["🔌 Raspberry Pi RP1 Southbridge (40nm TSMC)"]
direction TB
USB_CTRL["Dual USB 3.0 Host<br/>+ Dual USB 2.0"]
GBE_MAC["Gigabit Ethernet MAC"]
GPIO_CTRL["40-Pin GPIO Controller<br/>(I2C, SPI, UART, PWM)"]
MIPI_CTRL["Dual 4-Lane MIPI Transceivers<br/>(CSI Camera / DSI Display)"]
end
BCM2712 ==>|"Internal PCIe 2.0 x4 Bus (16 Gbps)"| RP1_Chip
BCM2712 ==>|"External PCIe 2.0/3.0 x1 Lane"| NVME
Silicon Comparison: Raspberry Pi 5 vs. Raspberry Pi 4
| Architecture Feature | Raspberry Pi 5 (BCM2712) | Raspberry Pi 4 Model B (BCM2711) |
|---|---|---|
| Silicon Fabrication Node | 16nm FinFET (High power efficiency) | 28nm Planar CMOS |
| CPU Microarchitecture | 4× Cortex-A76 (2.4 GHz) |
4× Cortex-A72 (1.8 GHz) |
| Instruction Pipeline | 4-wide out-of-order execution | 3-wide out-of-order execution |
| L2 Cache per Core | 512 KB private L2 per core | 48 KB private L2 per core |
| L3 Cache | 2 MB shared L3 cache | None (L2 acted as shared cache) |
| RAM Subsystem | LPDDR4X-4267 (128-bit wide bus) | LPDDR4-3200 (32-bit wide bus) |
| Memory Bandwidth | 17.0 GB/s (~2.5× higher throughput) |
6.8 GB/s |
| Cryptographic Acceleration | Native ARMv8.2-A AES, PMULL, SHA-1, SHA-256 | Software-only (CPU bottlenecked) |
| Internal I/O Architecture | Disaggregated PCIe Southbridge (RP1) | Monolithic on-SoC peripheral bus |
2. Hardware Cryptographic Acceleration: ARMv8.2-A Crypto
The lack of hardware cryptographic acceleration was the most prominent limitation of the Raspberry Pi 4. Every encrypted HTTPS request, WireGuard VPN tunnel packet, and LUKS disk encryption block had to be processed through general-purpose arithmetic instructions.
The BCM2712 includes the complete ARMv8.2-A Cryptographic Extensions:
- aes & pmull: Dedicated vector instructions for hardware-accelerated Advanced Encryption Standard (AES-128, AES-256) and polynomial multiplication for Galois/Counter Mode (GCM).
- sha1 & sha2: Hardware rounds for SHA-1, SHA-224, and SHA-256 hashing algorithms.
Verifying Crypto Flags in Linux
Inspect the CPU feature flags via /proc/cpuinfo:
Expected output on Raspberry Pi 5 (Debian 12 Bookworm 64-bit):
Features : fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm jscvt fcma lrcpc dcpop sha3 sm3 sm4 asimddp sha512 asimdfhm
(Note the presence of aes, pmull, sha1, sha2, sha3, and sha512).
Empirical Benchmark: Pi 5 vs. Pi 4 OpenSSL Throughput
Run the standard OpenSSL benchmark for single-thread AES-256-GCM encryption with 8 KB blocks:
| Benchmark Metric | Raspberry Pi 5 (Hardware Crypto) | Raspberry Pi 4 (Software Crypto) | Speedup Factor |
|---|---|---|---|
| AES-256-GCM Throughput | 2,450 MB/s (2.45 GB/s) |
185 MB/s |
13.2× Faster |
| SHA-256 Hashing Throughput | 1,820 MB/s (1.82 GB/s) |
140 MB/s |
13.0× Faster |
| WireGuard VPN Max Speed | 940 Mbps (Saturates Gigabit Line) |
320 Mbps (CPU pegged at 100%) |
2.9× Faster |
| LUKS Encrypted Disk Write | 820 MB/s (Unconstrained on NVMe) |
95 MB/s (Severe CPU throttle) |
8.6× Faster |
3. The Custom RP1 I/O Southbridge Silicon
The RP1 is the first custom silicon chip designed entirely in-house by Raspberry Pi Ltd. Built on TSMC's proven 40nm low-power process, the RP1 acts as a classic motherboard Southbridge:
%%{init: {'flowchart': {'wrappingWidth': 600, 'subGraphTitleMargin': {'top': 15, 'bottom': 15}}}}%%
flowchart TB
subgraph RP1_Internals["🔌 RP1 Southbridge Internal Architecture"]
direction TB
PCIE_CORE["PCIe 2.0 ×4 Endpoint Controller (16 Gbps)"]
AXI_BUS["Internal 128-bit AXI Crossbar Bus"]
M3_MGMT["Dual ARM Cortex-M3 Firmware Cores"]
subgraph Controllers["Peripheral Controllers"]
XHCI["Dual USB 3.0 xHCI Host Controllers"]
GMAC["Synopsys DesignWare Gigabit Ethernet MAC"]
PIOP["Programmable I/O & Enhanced GPIO Engine"]
MIPI["2× 4-Lane 1.5 Gbps MIPI CSI-2 / DSI Transceivers"]
end
PCIE_CORE <--> AXI_BUS
M3_MGMT <--> AXI_BUS
AXI_BUS --> XHCI & GMAC & PIOP & MIPI
end
Why Separate the Southbridge?
- Physical Silicon Constraints: Fast CPU logic (Cortex-A76) requires modern, dense FinFET nodes (16nm). However, analog I/O blocks (5V-tolerant pull-ups, USB 3.0 PHYs, MIPI transceivers) do not scale down efficiently in 16nm and consume massive silicon area.
- Dedicated USB 3.0 Bandwidth: On the Pi 4, both USB 3.0 ports shared a single PCIe x1 link through a VL805 controller (capped at 4 Gbps total). On the Pi 5, the RP1 provides independent 5 Gbps host bandwidth to both USB 3.0 ports simultaneously.
- GPIO Routing Shift: Because the GPIO header connects physically to the RP1 instead of the main SoC, Linux addresses the 40-pin header as
/dev/gpiochip4under the modernlibgpioddriver (rather than the old/dev/gpiochip0).
4. Industrial Compute Modules: CM4 vs. CM5
When engineering commercial appliances, medical devices, or high-density rackmount cluster nodes, consumer SBCs with fixed ports are suboptimal. The Compute Module extracts the core processor and RAM into a compact, standardized mezzanine form factor:
%%{init: {'flowchart': {'wrappingWidth': 600, 'subGraphTitleMargin': {'top': 15, 'bottom': 15}}}}%%
flowchart LR
CM["Compute Module 4 / 5<br/>(SoC + RAM + Optional eMMC)"]
MEZZ["2× 100-Pin High-Density<br/>Hirose DF40 Mezzanine Connectors"]
BASEBOARD["Custom Industrial Carrier Board<br/>(Dual NVMe, M.2 5G Modem, Dual 2.5GbE, DIN-Rail)"]
CM ==> MEZZ ==> BASEBOARD
Compute Module vs. Standard Model B
| Architectural Dimension | Compute Module (CM4 / CM5) | Standard Model B (Pi 4 / Pi 5) |
|---|---|---|
| Physical Interconnect | Dual 100-pin Hirose board-to-board connectors. | Fixed external USB, HDMI, and RJ-45 jacks. |
| On-Board Storage | Optional 8 GB – 64 GB industrial eMMC flash (eliminates SD cards entirely). |
MicroSD card slot only (plus external PCIe on Pi 5). |
| PCIe Lane Flexibility | Raw PCIe lanes broken out directly to mezzanine pins (supports SATA controllers, 10GbE NICs). | Single fixed x1 PCIe FPC ribbon cable (Pi 5 only). |
| Operating Temperature | -20°C to +85°C (Automotive & industrial certified). |
0°C to +50°C commercial rating. |
| Clustering Ecosystem | Compatible with blade chassis (Turing Pi 2, DeskPi Super6C, 1U rackmount shelves). | Requires individual power bricks and cabling per node. |
5. Hardware Diagnostics & System Architecture Auditing
Use native Linux tools to inspect the hardware topology of your running system:
1. Auditing CPU Topology and Clock Governors
Verify current CPU frequencies across all four cores:
2. Auditing PCI Express Hierarchy
On the Raspberry Pi 5, use lspci to observe the RP1 southbridge and any connected NVMe drive:
Expected output showing the dual-bus hierarchy:
-[0000:00]----00.0-[01]----00.0 Raspberry Pi Ltd RP1 PCIe 2.0 Southbridge
-[0001:00]----00.0-[01]----00.0 Samsung Electronics Co Ltd NVMe SSD Controller
3. Measuring RAM Subsystem Throughput
Verify real-world memory bandwidth using the standard sysbench tool:
sudo apt install -y sysbench
sysbench memory --memory-block-size=1M --memory-total-size=10G --memory-oper=write run
Where to Go Next
Now that the silicon architecture and cryptographic capabilities are mastered, proceed to power delivery engineering and active thermal management:
-
Next Guide: Power Delivery, PoE+ & Active Cooling
Master 27W USB-C PD negotiation, eliminate brownout voltage dips, wire PoE+ HATs, and configure PWM cooling fan curves.