The performance part. MIPI, H.264, high-speed USB, no radio.
- CPU
- Dual-core RISC-V with FPU and AI extensions
- Clock
- Up to 400 MHz
- Low-power core
- Low-power RISC-V core, 40 MHz
- On-chip SRAM
- 768 KB L2MEM + 32 KB LP
- PSRAM
- 16 or 32 MB in-package, up to 64 MB external
- Flash
- External, up to 64 MB
- Wi-Fi
- None — use a companion chip
- Bluetooth
- None — use a companion chip
- 802.15.4
- No
- GPIO
- 55
- USB
- USB 2.0 High Speed (480 Mbps) + Full Speed OTG + Serial/JTAG
- Deep sleep
- ~25 µA
- Package
- QFN 9×9
- Reference dev board
- ESP32-P4-Function-EV-Board ($56)
Stands out for
MIPI-DSI and CSI at 1080p, image signal processor, H.264 encoder, JPEG codec, 2D accelerator, 10/100 Ethernet, 3× TWAI
Watch out for
No radio, no in-package flash, most expensive and most power-hungry
Typical designs
Touch HMI panels, video doorbells, network cameras, USB capture devices, robotics
The chip that gave up the radio
Every other part in this family is defined by its radio. The P4 is defined by not having one. There is no Wi-Fi, no Bluetooth, no 802.15.4 — and the die area, power budget and pin count that would have gone to a radio went into application performance and I/O instead.
The result is by some distance the most capable Espressif application processor:
- Two RISC-V cores at up to 400 MHz with a single-precision FPU and AI extensions, plus a separate 40 MHz low-power core
- 768 KB of on-chip L2 memory, plus 16 or 32 MB of PSRAM in the package and up to 64 MB external
- MIPI-DSI display output and MIPI-CSI camera input, both handling up to 1080p
- An image signal processor (ISP) for the camera pipeline — demosaicing, white balance, noise reduction in hardware
- A hardware H.264 encoder and a JPEG codec
- A PPA (pixel processing accelerator) and 2D-DMA for blitting, scaling and rotation
- USB 2.0 High Speed at 480 Mbps, with an integrated HS PHY — plus a separate Full Speed OTG controller and a USB Serial/JTAG unit. Three independent USB controllers.
- 10/100 Ethernet MAC (RMII), three TWAI (CAN) controllers, SD/MMC
- 55 GPIOs, sixteen of them in the low-power domain
- I3C support — uniquely in the family — and LP-I2C, LP-SPI, LP-UART
- 14 touch channels, two ADCs, voice activity detection
What this actually enables
The distinction between the P4 and the S3 is the difference between a screen and a display.
An S3 drives a parallel-RGB panel at perhaps 800×480, decodes JPEGs in software with SIMD help, and runs LVGL acceptably if you are careful about what you redraw. A P4 drives a 1920×1080 MIPI-DSI panel, composites through hardware, and has enough PSRAM bandwidth to double-buffer it. Espressif’s own ESP32-P4-Function-EV-Board ships with a 7-inch 1920×1080 touchscreen, which tells you the target.
Similarly for cameras: an S3 with a DVP sensor gets you VGA stills and a slideshow-rate stream. A P4 with a MIPI-CSI sensor, the ISP and the H.264 encoder gets you an actual network camera or a UVC webcam at 480 Mbps over USB.
The radio problem, and how it is solved
You cannot ship a connected P4 product without something else providing the radio. Espressif’s answer is ESP-Hosted: a companion ESP32 chip runs the Wi-Fi and Bluetooth stacks and presents them to the P4 over SDIO, SPI or UART, with the P4 running the TCP/IP stack locally.
The standard pairing is P4 + C6, which gets you Wi-Fi 6, Bluetooth 5.3, Thread and Zigbee. Both are RISC-V, both use ESP-IDF, and Espressif supports the combination directly. Other combinations make sense for specific needs: P4 + C5 for 5 GHz, P4 + H2 for a low-power Thread link into building automation, P4 + E22 for tri-band Wi-Fi 6E when you are pushing video.
Budget for this in cost and board area. A P4 at $4.00 plus a C6 at $1.85 is a $5.85 silicon bill before memory and passives — several times the cost of a single C6 solution. The P4 is the right answer only when you genuinely need what it does.
Power
Deep sleep is around 25 µA — an order of magnitude worse than the C and H series, and unremarkable for a chip of this class. The 32 KB LP SRAM and the 40 MHz low-power core do let you run meaningful background logic while the main cores are down, which is more useful here than a lower floor would be. The P4 is a mains-powered or large-battery chip. Do not plan a coin-cell product around it.
Security
Strong: AES with DPA resistance and pseudo-round protection, hardware SHA, RSA, ECC and HMAC, an eFuse-backed ECDSA digital signature peripheral, APM/TEE isolation and a Key Manager. The combination of 400 MHz, 768 KB of SRAM and a hardened AES block also makes the P4 the best platform in the family for software post-quantum cryptography, since no embedded MCU accelerates ML-KEM or ML-DSA in hardware today.
Modules and boards
Bare chips are ESP32-P4NRW16X ($4.00, 16 MB PSRAM) and ESP32-P4NRW32X ($4.50, 32 MB), in a QFN 9×9 package. There is no in-package flash on any variant — flash is always external.
The reference board is the ESP32-P4-Function-EV-Board ($56 as the P4X variant): 7-inch 1920×1080 MIPI-DSI touchscreen, 2 MP MIPI-CSI camera, microphone, speaker, Ethernet and a microSD slot. ESP32-P4-EYE is the vision-focused kit, with a 2 MP MIPI-CSI camera, a 1.54-inch display, a fill light and a battery connector.
Third-party boards have arrived quickly: Waveshare’s ESP32-P4-Nano and ESP32-P4-Module-DEV-KIT are the most widely available, and the P4 is beginning to appear on MIPI-DSI panel boards aimed at Home Assistant dashboards.
Choose it when
- You are driving a high-resolution MIPI display — anything beyond ~800×480
- You need a real camera pipeline: MIPI-CSI, ISP, H.264 or JPEG in hardware
- You need USB High Speed (480 Mbps) — UVC capture, fast mass storage
- You need many GPIOs (55), multiple CAN buses (3), or Ethernet
- You need the most compute available in the family and can add a radio separately