Lambda connects GB300 racks with NVIDIA Photonics co-packaged optics
Original title: Lambda Boosts AI Factory Resiliency and Efficiency With NVIDIA Photonics
Lambda reports about 4% lower rack-scale latency and up to about 10% lower pod-scale latency versus its tests of conventional optics.
Cloud provider Lambda is using NVIDIA Photonics co-packaged optics switches to link NVIDIA GB300 NVL72 racks with more than 10,000 GPUs. NVIDIA says co-packaged optical engines replace pluggable transceivers and can deliver up to 5x better power efficiency, up to 15x higher reliability, and 4x fewer lasers. Each switch has 144 ports at 800Gb/s, liquid cooling, and 18 field-replaceable external light sources. Lambda reports a median optics temperature of 36°C versus 51°C for conventional pluggable optics, with the same throughput.
Unconfirmed: Power, reliability, and latency figures come from NVIDIA and Lambda’s own tests, not an independent audit.
- Switches have 144 ports at 800Gb/s and 18 removable external light sources.
- Median optics temperature was 36°C versus 51°C for conventional pluggable optics.
- Co-packaging shortens the electrical path and removes retimer chips, NVIDIA says.
- One interrupted training link can stall thousands of GPUs, the post argues.
- Products
- NVIDIA PhotonicsGB300 NVL72
From the original article
2 min read at blogs.nvidia.com
To train an AI model, thousands of GPUs have to work together, requiring a network that’s fast, efficient and resilient. Lambda is tackling the network piece of the puzzle with NVIDIA Photonics co-packaged optics (CPO) switches. The cloud provider is using the switches to connect NVIDIA GB300 NVL72 racks with more than 10,000 GPUs, delivering […]
By Brian Sparks · blogs.nvidia.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.