Computing · Transistor Count
Moore’s Law
Fifty-five years of doubling, and the lead has changed hands twice.
- 2,300 → 336B transistors, 1971–2026
- Doubling every ~2 years, 2.3 most recently
- Intel CPUs → AMD EPYC → Nvidia GPUs
- Leading-edge parts, not annual max
Intel’s 4004 carried 2,300 transistors in 1971, six years after Gordon Moore first described the trend. Nvidia’s Rubin carries 146 million times as many in 2026, and the doubling time has barely varied: two years to 2000, two years since, 2.3 years across 2020–2026. What has changed is who was setting it. Intel CPUs hold every point from the 4004 through the multi-core era. Then frequency scaling hit power limits and the frontier of integration passed to massively parallel silicon: AMD’s EPYC takes 2017, and Nvidia’s AI accelerators every point after. Blackwell packs 208 billion transistors on TSMC’s custom 4 nm-class 4NP process; Rubin puts 336 billion on a reported 3 nm-class node Nvidia has not named. These are representative leading-edge parts, not strictly the single highest-count chip shipped in each year; some specialty processors carried more, but the curve follows the mainstream frontier.
| Year | Value | Note | Projected |
|---|---|---|---|
| 1971 | 2,300 transistors | Intel 4004 | |
| 1974 | 4,500 transistors | Intel 8080 | |
| 1978 | 29,000 transistors | Intel 8086 | |
| 1982 | 134,000 transistors | Intel 286 | |
| 1985 | 275,000 transistors | Intel 386 | |
| 1989 | 1,200,000 transistors | Intel 486 | |
| 1993 | 3,100,000 transistors | Pentium | |
| 1997 | 7,500,000 transistors | Pentium II | |
| 2000 | 42,000,000 transistors | Pentium 4 | |
| 2006 | 291,000,000 transistors | Core 2 Duo | |
| 2010 | 1,170,000,000 transistors | ||
| 2014 | 2,600,000,000 transistors | ||
| 2017 | 19,200,000,000 transistors | AMD EPYC | |
| 2020 | 54,200,000,000 transistors | Nvidia A100 | |
| 2024 | 208,000,000,000 transistors | Nvidia Blackwell B200 | |
| 2026 | 336,000,000,000 transistors | Nvidia Rubin |