USA: The US-based company Broadcom has unveiled a new supercomputer networking chip. It will be used to connect the systems for artificial intelligence (AI) work and is known as Jericho3-AI.
Due to its best-in-class features like zero-impact failover, perfect load balancing, ultra-high radix, and congestion-free operation, the company claims that Jericho3-AI will revolutionise AI networking.
Supercomputers must be trained on a large amount of data in order for AI applications like Google Bard and ChatGPT to work. Numerous thousands of graphics processing units (GPUs), which can operate continuously for weeks or even months, are needed for this work.
Also Read: PM Modi and Apple CEO discuss India's technological advancements
The training is completed more quickly the faster the connection speed between chips. Jericho3-AI says it will quicken the process.
Up to 32,000 GPUs can be connected simultaneously by the Jericho3-AI. It is thus directly competitive with NVIDIA's InfiniBand hardware.
The chip from Broadcom states that it will guarantee optimal network utilisation even when the network load is "highest."
Additionally, it guarantees end-to-end traffic scheduling, which will prevent network jitter during operations, such as when an AI computation cycle is running.
Also Read: Indonesian hackers promise to keep attacking Israel online
Concerned about completing the job on time? I'm done. Using Jericho's zero-impact failover facility, auto path convergence is guaranteed to occur in under 10ns.
Additionally, as was already mentioned, Jericho3-AI can connect with up to 32,000 GPUs simultaneously at a speed of up to 800Gbps in a single cluster.
Jericho claims to complete tasks 10% faster than competing networking solutions.
Four times as fast as Jericho2, Jericho3-AI has an ethernet bandwidth of 26 petabits per second. Additionally, each gigabit uses 40% less power. Jericho promises low latency as well as high bandwidth.
Ram Velaga, senior vice president and general manager at Broadcom, explained the rationale behind the launch of Jericho3-AI by saying that "the benchmark for AI networking is reducing the time and effort it takes to complete the training and inference of large-scale AI models."
"Jericho3-AI delivers a significant reduction in job completion time compared to any other alternative in the market," he continued.