There were times when AMD and ATi and Nvidia (and with them many other brands) sent a new generation of graphics cards to the market every year. It wasn’t such a major problem, chips about millions of transistors, a new manufacturing process (sometimes more) each year, and memories whose transfer speeds grew so fast that most manufacturers didn’t care about its solution at all.
This era is long gone. Today we have key generations of production processes for 3 years and more, intergenerational quality shifts are getting lower, the speed of memory is growing within individual technological generations cosmetically and at the expense of consumption. Transistor budgets are in the tens of billions, and we are slowly reaching hundreds. It should come as no surprise that both AMD and Nvidia do not see meaning in the annual release of a new generation, as the annual shift in architecture is not significant, and virtually zero year-on-year differences in production process and memory cannot help achieve a generational shift. We have a generation for 2-3 years, which sometimes interweaves some refresh, when the yield of the process increases so much that it pays to keep some percentage of active stream processors active or to increase the clock frequency a bit.
Intel probably sees the situation differently. Already during his Architecture Day, he presented his plans in a context from which it could be deduced that the new generation would go on the market every year. It wasn’t just a mistake, as Bryce confirmed on Twitter that “If all goes well, it could [2025] be at the time Intel ARC ‘Druid’ arrives. “
- 2022: Alchemist
- 2023: Battlemager
- 2024: Celestial
- 2025: Druid
Given the alphabetical designation of each generation and the release of the Alchemist generation in the first quarter of 2022, this confirms that Intel counts one generation of graphics cards each year. However, it should be added that this does not necessarily mean a new architecture every year. Intel can combine the deployment of its processes with TSMC processes and each year, for example, use a hair better process than in the previous year (eg 6nm TSMC in 2022, 5nm TSMC or Intel 7 in 2023, Intel 4 in 2024, TSMC 4 / 3nm in 2025 ap.). After all, it is rumored that the first three generations will not differ fundamentally architecturally, and the Druid will be a big change. Finally, there is the condition mentioned by Bryce: “If all goes well…” The statement is therefore relatively cautious and it cannot be said that “Intel promises…”. Only that “Intel is planning…”
–
