-
Interconnect and Network Performance Tester
Test interconnects and emulate traffic up to 1.6T per port
-
Functional Interconnect Test Solutions
Error performance validation system for 1.6 Terabit (224G) interconnects
Interconnect and Network Performance Tester
Keysight interconnect and network performance testers deliver ultra-high-speed traffic generation for validating the bandwidth, latency, and stability of copper and optical interconnects in AI clusters and data center networks. Available in 800GE and 1600GE models, these testers emulate realistic traffic profiles and congestion control protocols to characterize the performance of switches, cables, and accelerators in AI / ML environments, with detailed metrics from PHY and link-layer error correction algorithms. With 1.6T support in both benchtop and rack-mount form factors, these systems give network equipment developers and data center operators the insight needed to optimize interconnects for AI workload processing. Need help selecting? Check out the resources below.
Functional Interconnect Test Solutions
Demand for higher data throughput in AI and data center networks is accelerating the industry’s transition to 800GE and 1.6 Terabit Ethernet (1.6T) network interconnects and beyond. Interfaces operating over 212Gb/s and 106Gb/s electrical lanes are critical to meeting higher reliability targets across network infrastructure. Networks must deliver increased bandwidth and lower cost per Gigabit. Industry standards for 800GE and 1.6T pluggable MSAs define a broad range of high speed fiber and copper interconnect types. Keysight’s multiple lane Functional Interconnect Test Solutions (FITS 8CH) provides 8 Tx and 8 Rx channels of functional BER and Forward Error Correction (FEC) error performance validation.
Meet the Interconnect and Network Performance Tester 1600GE
Validate next-generation 1.6T Ethernet interconnects with a high-speed test platform designed for AI data centers. The INPT-1600GE combines advanced traffic generation and analysis with intelligent automation to benchmark silicon, cables, and networking gear across 212 Gb/s electrical lanes. Paired with the Interconnect Test System (ITS) software, it streamlines validation with multi-user scheduling, automated reporting, and a first-of-its-kind interconnect library, accelerating test throughput and reducing time to insight.
Emulate real AI workloads, optimize interconnect performance, and future-proof your infrastructure with a scalable, rackmount or benchtop solution built for the demands of tomorrow’s networks.
Services and support
Innovate at speed with curated support plans and prioritized response and turn-around times.
Get predictable, lease-based subscriptions and full lifecycle management solutions—so you reach your business goals faster.
Experience elevated service as a KeysightCare subscriber to get committed technical response and more.
Ensure your test system performs to specification and meets local and global standards.
Make measurements quickly with in-house, instructor-led training, and eLearning.
Download Keysight software or update your software to the newest version.
Frequently asked questions
AI data centers rely on tightly synchronized, high-bandwidth connections between GPUs, switches, and accelerators. Any weakness in the copper or optical interconnects — such as signal degradation, power instability, or thermal issues — can introduce latency, cause training stalls, or reduce overall job completion efficiency. Interconnect validation ensures that high-speed links perform reliably under the intense, sustained workloads generated by AI training and inference. Keysight Interconnect and Network Performance Test products simulate full-rate traffic up to 1.6 Tbps across 212 Gb/s electrical lanes, allowing engineers to validate interoperability, performance margins, and stress response in transceivers, AOCs, DACs, and mid-board optics before deployment in large-scale GPU clusters or AI fabric networks.
AI workloads produce highly synchronized, non-uniform traffic patterns that differ significantly from conventional enterprise or cloud workloads. These include bursty, high-fanout flows, collective operations, and microsecond-level latency sensitivity across east-west traffic in the data center fabric. Traditional traffic generators often use steady-state, synthetic packet streams that fail to expose real-world performance bottlenecks. INPT systems use AI-specific traffic modeling to accurately emulate communication behaviors between AI accelerators and GPUs, replicating job-level timing, burst size, and synchronization dynamics. This approach enables meaningful performance evaluation of interconnects under conditions that mirror actual AI training scenarios, helping uncover congestion points and optimize fabric-level routing and design.
Testing at 800GE and 1.6T speeds requires traffic generators capable of full line-rate transmission using PAM4 signaling across 112 Gb/s and 224 Gb/s electrical lanes. Keysight’s INPT solutions deliver high-density Ethernet traffic generation up to 1.6 Tbps per port, with configurable support for 1x1600GE, 2x800GE, 4x400GE, or 8x200GE configurations. These platforms emulate worst-case scenarios using real application-layer protocols, enabling engineers to measure throughput, latency, and FEC tolerance under load. The system can stress test transceivers with up to 40W power draw per port, ensuring proper performance of next-generation optical interconnects under realistic thermal and electrical operating conditions.
Interconnect performance must be measured across multiple layers to ensure robust operation in AI infrastructure. Critical metrics include link-level latency, FEC correction rates, packet loss, jitter, congestion response, and queue depth. Keysight INPT platforms provide fine-grained telemetry and real-time analytics that expose bottlenecks during high-intensity data transfers. Engineers can monitor device behavior under conditions such as buffer pressure, retransmission events, or asymmetric traffic loads. This visibility helps optimize silicon-level design, switch firmware, and topology decisions, ultimately improving network efficiency and stability during AI model training and inference at scale.