Feather M4 vs ESP32-S3
Cortex-M4F (SAMD51) at 120 MHz against Xtensa LX7 (ESP32-S3) at 240 MHz, on real hardware, same C kernels, clocks verified.
Feather M4
Cortex-M4F (SAMD51)
- Family
- Arm Cortex-M
- Board
- Feather M4 Express
- Clock
- 120 MHz
- float / double
- 32-bit / 64-bit
ESP32-S3
Xtensa LX7 (ESP32-S3)
- Family
- Xtensa
- Board
- ESP32-S3 DevKit
- Clock
- 240 MHz
- float / double
- 32-bit / 64-bit
Typical gap
1.48×
ESP32-S3 faster, geometric mean of 30 cells
Biggest win for Feather M4
2.44×
double divide
Biggest win for ESP32-S3
2.87×
double modulo
Clock for clock
1.35×
Feather M4 needs fewer cycles per op (2.00× clock gap)
Operation by operation
Every cell
How many times faster Feather M4 is than ESP32-S3, operations per second. Blue: Feather M4 ahead. Orange: ESP32-S3 ahead.
| Add | Subtract | Multiply | Divide | Modulo | |
|---|---|---|---|---|---|
| uint8_t | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.64× 7.01 M/s · 11.5 M/s | 1/1.72× 6.36 M/s · 10.9 M/s |
| uint16_t | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.66× 6.12 M/s · 10.1 M/s | 1/1.81× 5.36 M/s · 9.72 M/s |
| uint32_t | 1/1.38× 9.71 M/s · 13.4 M/s | 1/1.38× 9.71 M/s · 13.4 M/s | 1/1.38× 9.71 M/s · 13.4 M/s | 1/1.68× 6.53 M/s · 11 M/s | 1/1.69× 6.19 M/s · 10.5 M/s |
| uint64_t | 1/1.27× 6.19 M/s · 7.87 M/s | 1/1.26× 6.19 M/s · 7.81 M/s | 1.03× 5.61 M/s · 5.45 M/s | 1/2.03× 885 k/s · 1.8 M/s | 1/1.97× 885 k/s · 1.74 M/s |
| float | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.41× 8.98 M/s · 12.7 M/s | 1/1.41× 8.98 M/s · 12.7 M/s | 1.36× 4.73 M/s · 3.48 M/s | 1/2.25× 866 k/s · 1.95 M/s |
| double | 1/2.54× 1.2 M/s · 3.04 M/s | 1/2.34× 1.19 M/s · 2.79 M/s | 1/1.17× 1.7 M/s · 2 M/s | 2.44× 1.26 M/s · 516 k/s | 1/2.87× 425 k/s · 1.22 M/s |