Overview

Choosing between merchant silicon and custom ASIC switches impacts your data center’s throughput, power efficiency, and feature velocity. Merchant silicon (e.g., Broadcom Tomahawk, Jericho) offers faster time-to-market and lower unit costs, while custom ASICs deliver deterministic latency and differentiated control-plane features for hyperscale or carrier-grade environments. This FAQ answers the most common pre-sales and post-sales questions from B2B network engineers.

Frequently Asked Questions

Q1: What is the maximum switching throughput and backplane capacity difference between merchant silicon and custom ASIC switches?

Merchant silicon-based switches typically provide up to 51.2 Tbps half-duplex switching capacity (e.g., 32x400G) with a backplane that matches port-side density. Custom ASICs can achieve similar aggregate throughput but often include larger internal buffers (2-4x per port) and oversubscription ratios tailored to specific topologies. For line-rate 400G, merchant silicon currently leads in raw Gbps/$; for deep-buffered 100G core routing, high-end custom ASICs maintain lower jitter.

Q2: Are third-party optical modules (SFP28, QSFP56, QSFP-DD) compatible with both architectures?

Yes, but merchant silicon switches generally enforce less strict DOM (Digital Optical Monitoring) coding due to open SDKs, making them more tolerant of third-party transceivers. Custom ASIC switches—especially from large OEMs—often require coded EEPROMs or commands like 'service unsupported-transceiver' to bypass vendor lock. Always verify IEEE 802.3 compliance: merchant silicon platforms typically accept generic optics after disabling ‘transceiver-type-check’ in CLI.

Q3: How do I configure basic Layer 3 interfaces on a merchant silicon-based switch via CLI?

Access console via baud rate 115200, enter global config mode, then use 'interface vlan 10' → 'ip address 192.168.10.1/24' → 'no shutdown'. Merchant silicon CLIs commonly resemble industry-standard SONiC or Cumulus Linux syntax. For custom ASICs (e.g., Cisco Silicon One, Juniper Express), the command set changes: 'set interfaces vlan unit 10 family inet address 192.168.10.1/24'. Commit the configuration; verify with 'show ip interface brief'.

Q4: Can I stack merchant silicon switches with custom ASIC switches in a mixed virtual chassis?

Generally no—stacking protocols (VPC, VLT, VC, VSX) are ASIC-family-specific. A merchant silicon switch using Broadcom SDK will not stack with a custom ASIC switch from another vendor. However, both can participate in the same MLAG (multi-chassis link aggregation) pair if they support LACP. For mixed environments, standardize on routed spine-leaf with BGP EVPN rather than proprietary stacking interfaces.

Q5: How do redundant power supplies fail over, and how do I troubleshoot a PSU fault on each architecture?

Both architectures support N+1 or N+N hot-swap PSUs with automatic load sharing. If one PSU fails, the surviving unit(s) take full load within 10-50ms (no traffic loss). To troubleshoot: on merchant silicon, use 'show system power' and 'show environment'; on custom ASIC, 'show chassis power' or 'show hardware psu'. Common fix: reseat PSU, verify input voltage (200-240V AC for high efficiency), or replace fan tray if thermal shutdown occurred. Logs are typically in '/var/log/syslog' (merchant silicon) or 'show log chassis' (custom ASIC).

Q6: Which architecture delivers lower latency for high-frequency trading or RoCEv2 workloads?

Cut-through merchant silicon (e.g., NVIDIA Spectrum, Broadcom Tomahawk) achieves sub-600ns port-to-port latency. Custom ASICs optimized for deterministic forwarding (e.g., Arista, Cisco G100) can go as low as 380ns but at 2-3x higher cost-per-port. For mixed workloads (storage + IP), merchant silicon offers programmable pipeline latency that varies by feature set; custom ASICs lock latency to a fixed value, which is preferable for time-sensitive finance applications.

Q7: What is the maximum packet forwarding rate (Mpps) for 100G ports on merchant silicon vs custom ASICs?

For 64-byte packets, a single 100G port on merchant silicon forwards 148.8 Mpps (line-rate). A custom ASIC achieves the same theoretical maximum but may drop to 130-140 Mpps when ACLs or VXLAN tunnel termination are enabled due to TCAM limitations. Merchant silicon with shared buffer architecture typically maintains 99% line-rate even with L3 routing enabled. Always check the vendor’s ‘wire-speed forwarding with IPv4/IPv6/MPLS’ footnote.

Q8: How do I migrate a production core from custom ASIC switches to merchant silicon without downtime?

Use a maintenance window with graceful insertion: (1) connect new merchant silicon switch as a parallel leaf via 100G uplinks, (2) run BGP with ‘max-metric’ to prevent traffic steering, (3) migrate one VLAN/L3 subnet at a time using static ARP or NSH rewriting, (4) use ‘monitor session’ to compare forwarding behavior. Merchant silicon’s ONIE boot loader allows zero-touch provisioning to mimic your custom ASIC’s VLAN and QoS config. Rollback script: ‘reload’ to prior config.