![]() 51.2 Tbps Switch ASICs to Drive the Migration to 800 Gbps and 1.6 Tbps Starting in 2H22 NVIDIA continues to drive products and solutions at a rapid pace. They continued to push the boundaries and accelerate the pace of innovation in the data center through the recent GTC announcements. With the launch of Spectrum-4, we got to see NVIDIA’s first Ethernet Switch announcement and what they were busy developing since acquiring Mellanox. Keeping with the accelerating theme, NVIDIA clearly accelerated the pace of Innovation and proudly showcased Spectrum-4 at GTC. Market Background Over the next five years, 51.2 Tbps ASICs will be responsible for over $20B in Data Center Switching revenue. About 2X the size of the 100 Gbps upgrade cycle and 4X the size of the 400 Gbps upgrade cycle. This class of ASIC is critical to allowing data centers to scale as more workloads become hardware accelerated. Hardware acceleration can come via the server, NiC, or DPU and causes networking bandwidth to double every year compared to the more traditional growth of 30-40%. NVIDIA’s 51.2 Tbps ASIC NVIDIA’s Spectrum-4 announcement included many new and first-to-market capabilities. It was the first 51.2 Tbps ASIC announced in the market and the first switch that can do 64 ports of 800 Gbps from a single ASIC. It also supports 1.6 Tbps speeds, another first and the preferred speed for the two most prominent Hyperscalers. In addition, NVIDIA announced Spectrum-4 as a fully integrated Switch that will sample later in 2022 and begin shipping in 2023. The chip will be produced at TSMC in the 4N process. NVIDIA did over $1B in Ethernet Switch and NIC revenue in 2021. It's All About The SERDES NVIDIA will be using its own, homegrown, 112 Gbps SERDES. SERDES has always been a challenge in switch development, with many ASIC companies having to integrate 3rd party SERDES. However, we view vertically integrated as a source of differentiation and will become more common in the future as ASICs increase in speed. SERDES will also be a key to adding additional IP blocks and photonics to the networking ASIC, as those markets evolve. The Speeds And Feeds Race NVIDIA’s Spectrum-4 introduced several features that help it distinguish and route traffic. These traffic classes are often described as elephant and mouse flows. The key at a market level is that AI and accelerated computing put a considerable burden on the network. So AI networking switches need to keep pace with this new class of ‘elephant flow’ traffic and not slow or drop packets. At the same time, accelerated traffic is growing as a percent of traffic. Multiple 100G ports are already standard at the RFP level for AI workloads. 12.8 Tbps and 25.6 Tbps ASICs can not keep pace with a fully loaded AI cluster. Beyond AI While the announcement focused on NVIDIA examples, such as connecting Spectrum-4 to NVIDIA Certified OVX OEM servers, we view the market as benefiting from NVIDIA taking networking seriously. Cloud customers have always requested multi-vendor at the ASIC and system level, and the ability to collapse networking tiers with 51.2 Tbps will be a critical driver to adoption. We, therefore, expect Spectrum-4 to apply to the whole data center, not just focused on the high-end or NVIDIA-connected solutions. We are excited to see this switch show up in 2H22 in the customer’s hands.
0 Comments
Marvell to Add Innovium to Its Portfolio to Enhance High-End DC Switching
On August 3rd, Marvell announced its planned acquisition of Innovium for $1.1B. The addition has good synergies. Innovium had several design wins in the hyperscaler segment and Marvell provides additional scale, desire to expand custom, semi-custom, and merchant networking offerings in the Cloud. While the deal is pending regulatoriy approval with a planned close date by the end of 2021, we do not see any impediments to the deal closing or being delayed. We have been tracking merchant silicon performance for over four years and for the entire 12.8 Tbps sampling and production curve and see many positives to this deal. The current state of 12.8 Tbps in 2021 remains the tale of two sets of customers. Over 1M ports of 400 Gbps a quarter ship into three hyperscalers in each quarter already. These customers use a mix of copper and fiber, and Innovium’s market share is impressive as a challenger to Broadcom. The rest of the market is mainly in trials, delayed by COVID-19 and multiple semiconductor shortages. As the market moves to mainstream 12.8 Tbps adoption, those three hyperscalers will move to 25.6 Tbps, a chip that Innvoium is currently sampling. This market dynamic creates an opportunity for Marvell. Not only are there synergies, but Marvell obtains a more substantial portfolio for what lies ahead in networking (chiplets, onboard optics, and silicon photonics) and back-to-back opportunities to win ASIC share. Marvell’s recent product announcements and Inphi acquisition highlight several go-to-market opportunities and new scale as a data center supplier that Innovium can now participate in. 650 Group's early hyperscaler executive leadership interviews indicate positive feedback regarding the announced deal. While this is a consolidation amongst ASIC suppliers, we view the long-term state of Marvell with Innovium as leading to more OEM/ODM ASIC choices and increased innovation and R&D for this market. The deal paves the way for the market to be more robust when we discuss 51.2 Tbps which is not that far out. It’s important to note that the future state of data center switching involves significant innovation and increasing importance inside the data center. Ethernet switches will be critical to deploying AI/ML and other future workload technologies to move beyond plumbing to a key interconnect technology. Turning to vendors, we see an increased interest in using ASICs as a way to differentiate product offerings, and our OEM/ODM interviews indicate a similarly positive view of this announcement. Product launches in 1H21 were slow as vendors took a conservative approach to launch new products and not ship. We view the timing as suitable for OEM/ODM as they look to early 2022 for new product launches. We look forward to learning more when the deal closes and during Marvell’s analyst day later this year. - Alan Weckel, Founder and Technology Analyst at 650 Group.
Today Intel had a major Data Center event in San Francisco. It was a multi-hour announcement showcasing all the different products Intel is/will be launching.
Some interesting background that Intel talked about was that only 2% of worlds data has been analyzed and a 5G will be a major driver move compute to the edge. They also touted that over 50% of AI workloads are inference and runs best on Intel (X86). Noticeably absent at the beginning of the presentation was Intel’s work on training. We found the most interesting parts of the announcement being that AWS talked about custom versions of Intel’s CPU and a Up to 14X inference improvement from just July 2017 in its XEON processors. Overall there is a 30% gen/gen improvement in XEON, the biggest jump in 5 years. While staying at 10NM, Intel is able to continue to squeeze performance gains out of the server. It is important as we hit the limits of process geometry that everything be accelerated, especially with Intel having a similar view of AI workloads and that’s what we saw with Intel. New Optane memory, new persistent memory, faster Adapter cards which will lead to more Smart NIC announcements, and a 10nm FPGA. It was clear at the event that large cloud provers like AWS, Azure, and Tencent are looking at all avenues to increase performance and reduce power consumption via software and hardware advancements. Some interesting highlights included AWS touting over 100 unique instances that leverage Intel processers with more SAP instances running on AWS than anywhere else and Formula 1 using of 65 years of historic race data to train its models in order to make real-time race predictions. New CTO - It was great to see and hear Juniper’s new CTO, Bikash Koley at NXTWORK. The message was very clear that new networks need to be built, not only from the speeds and feeds technology point of view, but from the operator point of view. Scale and simplicity are very important and only going to be more critical as billions of IoT devices dump traffic and data on the network. Juniper is looking to take the network it is building for some of the hyperscalers and Service providers and take help the tier 2/3 Cloud as well as the enterprise customer adopt a cloud architecture for the future.
Contrail – The big announcement was OpenContrail joining the Linux foundation. What we saw in talking to customers and listening to the talks was that Contrail has significant commercial adoption, definitely larger than perceived in the marketplace. We see the largest hyperscalers and service providers as having their own controllers, but see Juniper leveraging its expertise in this area into smaller Cloud and service providers. This will give Juniper the opportunity to build, operate, and transfer for many customers. We were impressed with the caliber of customers that Juniper has. Juniper has premier Cloud, Service Provider, and Enterprise customers. We enjoyed listening to Twitter’s VP of engineering on stage as an example. We see Juniper as one of just a handful of vendors that can support a customer base with this breath and complexity. We note Juniper mentioned complexity and simplicity throughout the sessions; we only see networking getting more complex, especially as we move beyond 100 Gbps. It is up to vendors to help the human scale with that complexity and we saw Juniper give many good examples during NXTWORK 2017. Today Broadcom announced Trident 3. The companies third major release of a chip that drove the merchant silicon revolution in the data center and started the white box movie in the Cloud. With Trident 3, all of Broadcom’s data center switching ASICs now support speeds of at least 3.2 Tbps per chip.
Trident 3 is impressive, but a few things about it really caught my eye. First, Trident 3 will offer five different skus, two of which are really focused on campus switching. One could see a 48-port 2.5 Gbps switch out of the X3 version of Trident 3 next year. We believe the Trident family moving into the campus will be significant for the industry once products begin to ship. Second, native 25 Gbps ports. Trident is the most popular of Broadcom’s ASICs, especially in the enterprise, and with Trident 3, we expect the market to quickly move away from 10 Gbps/40 Gbps products and towards 25 Gbps/100 Gbps products. This product aligns well with our forecasts for this transition which we are excited to be publishing shortly. We still don’t see a bandwidth need in most enterprises for 25 Gbps, but the ability to future proof at the customer level and the ability to consolidate skus at a vendor level will make this compelling. Third, we see a potential for both switch vendors and customers to benefit from one family of ASICs from the campus all the way to the data center. While it is too early to know the impact of this right after the announcement, we look forward to conducting interviews over the next few months to define this impact. Extreme Networks has been very busy announcing acquisitions lately. Yesterday, it announced plans to acquire Brocade's Data Center switch business and on March 7, it announced plans to acquire Avaya's switch business. Additionally, it completed its acquisition of the Zebra Wireless LAN business just last October 2016.
When the two switch acquisitions are completed - some time in the summer of 2017 - the company will be vaulted higher to become one of the major share leaders in switching. To put the acquisitions into context, assuming the acquisitions had been completed last quarter, using our 4Q16 Ethernet Switch Market Report revenue share figures, the company's share would have been:
In 2016 Data Center switch market share leaders by revenue were (alphabetically) Arista, Cisco, Juniper and White Box vendors. In the North America in the Campus Switch market by revenue, considering Extreme had acquired the switch divisions of Brocade and Avaya it announced in March 2017, market share leaders were Cisco, Extreme and HPE. Starting this year, we expect that networking customers will get used to seeing Extreme in the top share lists when considering vendors. |
CHRIS DePUY
|