In the drive to bring ever-more capacity to the world’s largest networks and data centers, the industry has been building towards 800G Ethernet transport for years. If you attended the Optical Fiber Communication Conference (OFC) this March, however, you couldn’t help noticing that we seem to have reached a tipping point. Based on the equipment and demonstrations on display, 800G technology is officially no longer theoretical. It’s here, and it’s poised to rewrite the rules for the world’s most in-demand networks - potentially even sooner than anticipated.
Among many highlights, OFC showcased:
- Real, market-ready 800G switching and interconnect products, not just demos
- Ongoing interoperability demonstrations from a growing ecosystem of vendors including network equipment manufacturers (NEMs), optical vendors, cable manufacturers, and others
- Rapid growth in the next generation of Ethernet transport, including new 800G ZR/ZR+ optics and even early strides towards 1.6T
At the same time, some questions remained about the optimal path forward, including uncertainty around pluggable form factors, 800G standards, and ongoing technical issues. Fast-changing technologies always present such challenges - especially for vendors trying to make sure market-ready solutions are available when customers need them. But just a few months since OFC, we’re already getting clarity on some of the biggest open questions. Let’s take a closer look at the latest developments in this space and the key trends to watch moving forward.
We already have a “killer app” for 800G: AI
The industry has long expected AI workloads to place new demands on digital providers, data center networks in particular. But with the rapid growth of ChatGPT, the AI revolution looks to be kicking off sooner than expected.
Every prompt given to large language models like ChatGPT generates a massive flood of data center traffic. It’s a different, more demanding type of workload than cloud providers typically contend with, and with explosive growth in AI services, many are already struggling to keep up. More than any other factor, this trend is likely to accelerate 800G implementations, as it’s the quickest, most cost-effective way to contend with this massive new demand that will only keep growing.
State of the standards
One of the biggest questions with 800G has been how the industry would resolve potential differences between the early Ethernet Technology Consortium (ETC) standard (800GBASE-R) and the eventual IEEE 802.3df standard, set to be finalized next year. Early products and prototypes have used the IEEE’s 802.3ck standard for the 112G electrical lanes used in 800G, but the full standard was still under development. That left unanswered questions about interoperability for even core electrical transmission capabilities like auto-negotiation and link training. There was even some concern that this might turn into a “VHS vs. Betamax” kind of battle, with long-term implications across the industry.
Fortunately, it’s looking like the formal IEEE standard will align closely with ETC’s. This shouldn’t be surprising, as ETC has achieved a significant installed base from vendors building early solutions, and many stakeholders are the same across ETC and IEEE.
Today, the industry is talking less about 802.3ck and converging on 802.3df for 800G and beyond. This is good news on two fronts. First, it’s always better when we can reach industry consensus earlier. Additionally, by settling open questions now, the IEEE can get a head start making progress on the 802.3dj standards work needed for next-generation 1.6T Ethernet.
Technical work continues
In addition to auto-negotiation and link training, vendors and testing providers continue to work through other outstanding technical issues with 800G, including:
Debugging: Debugging demands continue to be significant as NEMs adopt new chipsets and architectures. This work is critical and will require significant effort for the foreseeable future. As vendors look to gain any competitive edge they can, testing is essential to not just demonstrate standards compliance, but to validate areas where they seek to differentiate.
Interoperability: Vendors also need to continue performing rich protocol testing to ensure smooth performance in real-world 800G environments. NEMs, hyperscalers, and other early adopters also need to conduct exhaustive testing on physical components like cables, transceivers, and fiber optics. At this point, different vendors may still have different interpretations of the standards. With billions of dollars on the line, customers can’t afford to discover interoperability issues via outages or degradations.
Is a winner emerging in pluggable form factors?
Another open question for 800G Ethernet transport has been which optical form factor would ultimately prevail, OSFP or QSFP-DD. Both offer advantages, and both have advocates. Now, it looks like the industry is leaning towards an answer: OSFP. The reason isn’t necessarily due to 800G; it’s about what comes next. OSFP offers several advantages that will allow for an easier transition when the industry begins moving to 1.6T.
For NEMs and interconnect vendors, OSFP offers a simpler technical path to achieve 224 Gbps per lane than QSFP-DD, as it’s less challenging to achieve the required signal integrity of up to 70 GHz per channel. OSFP’s larger form factor also offers more real estate to work with, especially with new layout designs that minimize concerns about port density. As a result, manufacturers can more easily address the thermal/cooling requirements of 800G transceivers that consume more power and generate more heat.
These electrical and thermal characteristics make OSFP a good choice for 800G, and will be even more valuable as the industry tools up for 1.6Tb technology. As a result, we’re starting to see significant traction towards OSFP.
Looking ahead
It’s encouraging to see how much progress the industry has made, and how quickly, in advancing 800G. There is extreme demand for this technology, and it’s impressive to see the industry respond so rapidly. As this work is still ongoing, however, and standards are still not fully finalized, it’s more important than ever to perform thorough, exhaustive testing.
If you’re among those developing new Ethernet products, be sure to keep the customer experience front and center by testing with as close to real-world scenarios as possible. Additionally, for fast-evolving standards like 800G, look for testing partners with a track record of being “zeroth to market,” who have done this before and are active within IEEE, ETC, and other industry bodies. By working with a partner that’s actually answering the open questions - not just waiting for the dust to settle - you stand the best chance of making the right calls every step of the way.
