On Tuesday, January 10th, Intel launched its 4th Generation Xeon processors, also known as Sapphire Rapids (scalable processors), Sapphire Rapids HBM (CPU with high-bandwidth memory), and Ponte Vecchio (data center GPU) and several partnerships leveraging Sapphire Rapids to enhance data center performance.
The launch comes amidst chatter about four delays in the release of the two Sapphire Rapids offerings and Ponte Vecchio. According to Data Center Knowledge research, Intel pushed back the launch four times in the last two years. Industry analysts say Sapphire Rapids wasn’t worth the wait. Here’s Omdia analyst Manoj Sukumaran's take on the impact of the Sapphire Rapids delays:
- Control of server the market: Since Intel controls 70% of the server market their CPU delays had repercussions along the entire supply chain.
- Still lagging behind AMD: While the Intel Data Center GPU Max Series is indeed a monster co-processor, I believe the AMD MI300 announcement outshined the GPU Max Series launch.
- Server vendors put on hold: Everyone wanted Intel to launch it [Sapphire Rapids] ASAP. Both of these products [Sapphire Rapids and Ponte Vecchio] have slipped many delivery schedules before and server vendors had their platforms ready to ship waiting for the CPU to come from Intel.
Intel’s CEO has been candid about the delays, sharing with The Verge that the project was too complex from the start.
"That project was started five years ago, so it's in-flight. I can't just reset the methodology of a product that began five years ago," said Gelsinger to The Verge last October. "[Sapphire Rapids] had way too much complexity in it, with three major new systems, or interfaces, in that design... and there were no backups on any of them."
Also of note is Intel’s first foray into chiplet-based design. Though chiplet technology has been around since 1965, AMD’s Lisa Su led the firm’s revival primarily on the back of chiplet-centric designs. This approach allowed AMD to compete head-to-head with Intel and gain ground on their market share even in the midst of bankruptcy woes.
That’s because of the flexibility chiplets provide semiconductor designers. From a core chiplet design, chip companies can swap out chiplets to meet specific market needs, creating several products from one core design.
With this chiplet approach, Intel now boasts a boost in performance from its previous generation of Xeon processors.
Intel’s chiplets include high-bandwidth memory (HBM) for the CPU portion of the Sapphire Rapids lineup. “The most interesting launch Intel had today is the The Xeon CPU Max Series which has HBM on the package,” says Sukumaran. “I believe this is the most differentiated product launched today. Compute and memory workloads would definitely benefit from this memory architecture and is a clever move from Intel.”
Cisco Servers Leveraging Sapphire Rapids
Cisco announced its 7th Generation Unified Computing System (UCS) series and its X-series servers. The servers are powered with Intel’s Sapphire Rapids processors and industry analysts believe this is an incremental step forward for both companies.
“This new platform will have higher performance and footprint density compared to previous generation servers, which will translate to improved cost efficiencies for the typical enterprise user,” says Baron Fung, research director at The Dell’Oro Group. “Most of the server vendors have announced new servers based on the 4th generation Sapphire Rapids, but these enhancements will be particularly beneficial to Cisco’s X-series blade modular platform, which prioritizes high densities.”
Highlights of the processor strength in Cisco’s servers include Sapphire Rapids’ DDR5, PCIe5, and CXL. These new standards allow memory and interfaces to expand capabilities without hindering server performance.
“CXL is the most notable feature for the 4th Gen servers, as it could potentially enable new architectures that will improve utilization for composable infrastructure,” says Fung. “Even though it is not currently apparent how the Cisco UCS platform will leverage CXL in the upcoming release, it will pave the way for more efficient architectures for Cisco’s converged platform, Hyperflex.”
Nvidia Announces Accelerated Hopper Systems Based on Intel’s Sapphire Rapids Processors
In coordination with Intel’s Sapphire Rapids announcement, Nvidia shared news of a partnership with Intel to leverage the Sapphire Rapids processors in its H100 GPUs and its H100 DGX servers.
Touting the partnership as the path to greener data center operations, Nvidia says combining Xeon Scalable with its NVIDIA ConnectX-7 will lead to lower data center power consumption.
Nvidia promises new systems from partners that combine their H100 GPUs with the Sapphire Rapids CPUs. The announced partners include ASUS, Cisco, Dell Technologies, Hewlett Packard Enterprise, and Supermicro.