As part of today’s announcements, during Intel’s Q&A session following the prepared remarks, CEO Pat Gelsinger explained how Intel is going to revive its fortune when it comes to the leading computer products. One of Gelsinger’s mantras seems to be that undisputed leadership products bring undisputed leadership margins to those products, and that Intel must return to its old days for Intel to execute them.
In the past, through the 1990s, 2000s and into the 2010s, Intel’s manufacturing philosophy was called ‘Tick-Tock’. This means that the leading computer hardware for each product generation was a check mark (improvement of the process node) or a tock (improvement of microarchitecture). Each generation would alternate between the two, enabling Intel to take advantage of a familiar design on a new process node or use a mature node to enable a new performance-oriented design. The policy was outraged when delays to Intel’s 10 nm forced Intel more into a Tick-Tock-Optimization-Optimization-Optimization model.
CEO Pat Gelsinger said today that Intel needs to restore the core of the Tick-Tock model that enables repeated leadership in the CPU ecosystem, driven by a sound CPU roadmap. Part of this is re-establishing discipline in Intel’s ranks to deliver continuous micro-architecture updates and process node updates on a regular expected cadence. Pat said as part of the call that Intel would look at a confirmed annual improvement of the process node, and consequently there could be quite a few ticks in the future, with the pressure on more tweaks.
In addition to these comments, Pat Gelsinger also said that Intel’s CPU cards have already been built in through 2021, 2022 and 2023. The company therefore wants ‘undisputed CPU leadership performance’ after 2024/2025, which is traditionally the fastest processor for single-wire and multi-wire workload. This is certainly a commendable goal, but Intel will also have to adapt to a changing landscape of chiplet processor designs (coming in 2023), which should improve the accelerators in the field (GNA already available), and also what it means for leadership performance to have – in the modern era, achievement of leadership does not mean much if you also drive a lot of Watts. Intel has said that its 7nm process is now comfortably on track to deliver Meteor Lake, a client CPU that uses tiles / chipsets, in 2023, but we are probably looking for a 7nm variant or even external processes for a 2024/2025 product. Intel has also stated that it wants to consider the core of its leading computing on external foundry processes, although one could argue that it does not explicitly say ‘CPU’.
It is also noteworthy that Intel / Gelsinger does not refer to the disaggregated silicon as ‘chipsets’, preferring to use the term ’tiles’. This is because Intel’s tiles are long threads over 3D packaging technologies such as EMIB and Foveros, compared to packet-based multi-inter-connection that requires buffers as well as control material. Tiles by this definition are more expensive to implement than chiplets and have additional thermal considerations by having strong-strength silicone close together, so it will be interesting to see how Intel incorporates these new packaging technologies with the more cost-sensitive elements of its portfolio balancing. , such as client processors.
It is known that Intel’s micro-architecture teams did not wait idle for 10 nm to get through the pipe, with a number of designs ready to wait as the process node technology ages. With any luck, if Intel can get a 7 nm headwind, then 2024 could be fast and fast if it turns in 2024.