According to an excerpt from a market report, AMD is reportedly hiking pricing for some users of its EPYC data centre processors by 10% to 30%, a new reality in the face of chip shortages that have resulted in less visibility for future shipments.
AMD declined to comment on the subject since it has a policy of not commenting on client pricing. Intel’s next-generation Sapphire Rapids chips could be delayed until Q3 2022, according to the report, which means Intel could miss the re-scheduled timeframe it set last year.
Dolly Wu, the VP and GM of Datacenter/Cloud at Inspur Systems (the world’s second-largest server provider), is quoted in a report by Mizuho Securities’ Managing Director Jordan Klein.
According to the article, AMD has hiked pricing by 10 to 30%, but the impact varies by client, with major cloud customers paying less. According to the source, AMD’s clients are just accepting the higher prices because there is no indication of when more CPUs would be available.
In the data centre, AMD’s Milan has the best peak performance and, maybe more crucially, the best performance-per-watt. Milan’s greater core counts also provide superior performance density, allowing data centres to squeeze in more performance per server while lowering operational costs, all of which is appealing enough to justify the price increases.
AMD’s price hikes are unsurprising, considering the company’s reliance on outside firms for both lithographic (wafer) and packaging (OSAT) capacity, the latter being a particular source of pain in the industry. Due to EPYC’s multi-chip design, which scales up to nine die in a single package, both of these variables may have a greater impact on EPYC supply than on AMD’s other types of chips. Manufacturing prices have risen at practically every point in the supply chain, and AMD’s price hikes are more likely a result of the firm passing those higher costs on to its consumers than a way to boost margins.
According to the article, Intel has increased Ice Lake production and expects a 50 per cent rise in supply year over year in 2022. In addition, Intel isn’t boosting prices for its Ice Lake chips to maintain market dominance. Both of these problems may contribute to AMD’s market share advances being stifled.
However, according to the article, Intel’s full-scale Sapphire Rapids volume launch will be pushed back until ‘perhaps’ Q3 2022. (Intel had previously projected the formal launch in Q2 2022). Furthermore, the business does not believe Sapphire Rapids will aid Intel much, anticipating that Intel’s market share will be further eroded as a result of Sapphire Rapids” materially higher’ BOM pricing.
Intel’s Sapphire Rapids features a multi-chip architecture fabricated on the ‘Intel 7’ process and linked together by a plethora of EMIB connections that enable better throughput and latency than other packaging techniques, but at a higher cost. Even though Intel conducts most of its packaging in-house, due to material shortages, packaging remains a production bottleneck. Sapphire Rapids will be the company’s first widespread use of its EMIB technology in a high-volume range like Xeon, which could contribute to the expected price hike.
Intel CEO Pat Gelsinger has stated that Sapphire Rapids should bring Intel’s data centre chips closer to parity (Sapphire Rapids will win some and lose some), but that he does not expect Intel to acquire a clear lead over AMD until the next generation of data centre chips is released. Wu, on the other hand, believes that Milan and Genoa will maintain their performance advantage over Xeon, allowing AMD to continue its meteoric rise in the data centre.
Because AMD’s supply of silicon is more limited than Intel’s, the company’s ability to increase market share more quickly is hampered. Despite this, Intel faces difficult competition as TSMC ramps up 7nm capacity and AMD’s 5nm Genoa arrives later this year. As a result, the struggle for server sockets will heat up throughout the year.