An in-depth look at Intel's new Ice Lake chips, including their development history and benchmark results
Dr. Ian Cutress / AnandTech :
Context & Ripple Effects
Ian Cutress's Ice Lake deep dive lands at the end of a four-year architecture drought: since the Skylake launch in 2015, Intel's client chips had been iterative refreshes on the same core design while the 10nm node slipped repeatedly. Ice Lake is the first mainstream shipment of that long-delayed process, so the benchmark numbers here are effectively an audit of what the delay cost and what it bought.
The piece also sets up the fork Intel faced in 2019: Ice Lake carried the new 10nm design into mobile, while the server roadmap still ran through 14nm parts like the second-generation Xeon Scalable (Cascade Lake). The related coverage closes that loop two years later, when 10nm finally reached the datacenter.
First-order effects
- Laptop buyers and OEMs get the first mainstream 10nm client silicon, making this the reference point for whether Intel's node problems translated into real-world performance or just slid further out.
- AMD's Ryzen mobile push now competes against a genuinely new Intel architecture rather than another Skylake derivative, raising the bar for Intel's own marketing of generational gains.
Second-order effects
- Desktop stays behind on 14nm — the coverage's later Comet Lake review shows Intel pushing clocks 'to new heights at the expense of power efficiency,' the classic stopgap when a node isn't ready, which splits Intel's lineup into two power-efficiency stories across segments.
- Every claim in these benchmarks becomes the baseline the eventual server part is measured against; the 2021 Ice Lake Xeon review frames its gains explicitly as the payoff of Intel's first 10nm server design.
Third-order effects
- If 10nm holds in volume, Intel's cadence reverts from architecture stagnation to a normal node-architecture rhythm — but the multi-year gap means competitors set expectations during exactly the window Intel was standing still, a textbook case of execution risk compounding into market-position risk.
- The client-to-server lag visible in this coverage (10nm mobile in 2019, 10nm Xeon in 2021) points to a structural pattern where each new Intel process debuts in low-power parts years before it can serve the higher-margin datacenter business.
The trend: Intel's post-Skylake era is one long node-recovery arc — 14nm refreshes holding the line until 10nm ships in client silicon first and reaches Xeon only years later.