After almost a year and a half of delays, Intel has begun to ship its 4th Generation Xeon Scalable processors, code-named Sapphire Rapids, to customers, and it has set January 10, 2023 as the formal launch date.
The launch is a formality because, according to an Intel spokesperson, the new Xeons are already shipping to customers—OEMs—now, but it falls to those OEMs to announce their product release plans.
CEO Pat Gelsinger said during the company’s earnings call last week that the company was ramping up production for launch and that he expected the new Xeons to see the fastest ramp to one million units ever.
The challenge for Intel wasn’t in design, it was manufacturing. This will be the first generation of chips using Intel 7 fabrication, an advanced 10nm design that took years to get right.
4th Generation Xeon Scalable is ambitious, with a new micro-architecture, up to 60 cores per chip, plus support for DDR5 memory, PCI Express Gen 5, CXL 1.1, and HBM2E memory.
It also brings in special-purpose accelerators, something not seen in the Xeon before. These include:
- Advanced Matrix Extensions (AMX): x86 instructions specifically for AI
- Dynamic Load Balancer (DLB): provides load balancing and scheduling across CPU cores.
- Intel Data Streaming Accelerator (DSA): reduces power consumption and bottlenecks when moving large datasets around.
- Intel In-Memory Analytics Accelerator (IAA): compresses and decompresses data in memory when moving it around, reducing memory usage.
- Intel QuickAssist Technology (QAT): accelerates crypto and compression tasks, speeding up security functions.
Moving up the announcement about the chips may have to do with two bits of news.
First, AMD will announce the release of Genoa, the third generation of its Epyc server chips, with 96 cores on November 10. Taiwanese research firm TrendForce estimates AMD’s x86 server CPU market share to be approximately 15% in 2022 and will rise to 22% in 2023.
Second, tech marketing-analysis firmTrendForce issued a report that predicted the launch of Sapphire Rapids had been moved from the fourth quarter of this year to the first half of next year due to manufacturing issues.
TrendForce said Intel‘s production yield rate for Sapphire Rapids was estimated to be in the range of 50 percent to 60 percent. What that means is for every wafer Intel made, half were unusable.
It will be interesting to see whether Gelsinger meets the production ramp-up he promised.
