
Keysight Technologies has used its network emulation expertise to enter the artificial intelligence (AI) and machine learning (ML) infrastructure market with a system that can emulate hundreds of accelerators.
The Keysight AI Data Centre Test Platform significantly improves benchmarking of GPU-based AI infrastructures at scale. This allows developers to reproduce realistic network behaviour of AI and ML workloads at scaler with measurable fidelity
The deployment and use of AI is growing rapidly across all industrial segments, and the race to train and deliver new AI models quickly and efficiently is a top priority for corporations. AI / ML workloads process vast amounts of data which requires high networking bandwidth and computing performance to reduce training time. However, the cost to design and validate large-scale “what-if” assessments is prohibitive even for the largest AI operators and varies significantly between scenarios.
The Keysight AI Data Centre Test Platform delivers highly tunable AI workload emulation, pre-packaged benchmarking apps and dataset analysis tools to significantly improve performance of the AI / ML cluster network fabric.
The platform emulates high-scale AI workloads with measurable fidelity via pre-packaged benchmark applications. These execute defined AI / ML behavioural models so that users and customers can reproduce experiments.
There is also a choice of test engines, from AI workload emulation on Keysight hardware load appliances to software endpoints on real AI accelerators to compare benchmarking results.
“800GE is a critical driver for AI / ML growth with the technology poised to have the fastest ramp of any data centre port speed in history. AI / ML bandwidth growth will grow at over 100% per year through the end of the decade, and 800GE will play a key role in that growth. The Keysight solution is at the forefront of testing AI / ML infrastructure with many deployments set to reach production in the next 18 months,” said Alan Weckel, Founder and Technology Analyst at the 650 Group.
Martin Hull, Vice President, Cloud and Platforms at Arista Networks said: “It’s been proven that a high-speed Ethernet network delivers the economical scale and performance needed to reduce job completion time for processing AI / ML workloads while adhering to open standards. Keysight’s AI Test Platform is a valuable tool to validate Arista’s technological leadership in AI network design.”
Ram Periakaruppan, Vice President and General Manager, Network Test & Security Solutions, Keysight, said: “As a leader in testing ultra-high-speed 800G Ethernet networks, Keysight continues to lead the way in innovation, having formed close partnerships with hyperscalers to co-design this groundbreaking new AI test platform that realistically benchmarks and emulates high-scale AI workloads unlike other test tools available today.”
