Specifications
CPU, EIC7700X
4x SiFive Performance P550 RV64GC, 1.4 GHz to 1.8 GHz
CPU Class
Cortex-A75-class RISC-V performance
NPU, EIC7700X
19.95 TOPS INT8 with BF16 support
NPU, EIC7700
13.3 TOPS INT8 at 1.4 GHz
NPU, EIC7702 / 7702X
26.6 TOPS and 39.9 TOPS with 8x P550 cores
GPU
Imagination AXM-8-256 — OpenGL ES 3.2, Vulkan 1.2, OpenCL 2.1, Android NN HAL
Vision DSP
Single DSP cluster with 512 INT8 SIMD
Vision Engine
HAE 2D blit, crop, resize and normalisation plus 3-layer OSD
Memory
Up to 32 GB 64-bit LPDDR4/4x/5 (up to 64 GB on EIC7702/7702X)
Cache
32 KB + 32 KB L1, 256 KB L2, 4 MB L3 with SECDED ECC
Video Decode
H.265 up to 8K @ 50 fps or 32 channels of 1080p30
Video Encode
H.265 up to 8K @ 25 fps or 13 channels of 1080p30
Storage
eMMC 5.1, 2x SDIO 3.0, SATA III 6 Gb/s, SPI NOR flash
Expansion
4-lane PCIe 3.0 with root-complex and endpoint support
Networking
2x GMAC with RGMII Gigabit Ethernet
Camera Inputs
Up to six camera interfaces via MIPI D-PHY v2.1 / C-PHY v1.2
Display Output
HDMI 2.0 with HDCP 1.4/2.2 plus 4-lane MIPI-DSI TX
USB
2x USB 3.0/2.0 with dual-role data
Power Consumption
Typically 8 W running CNN workloads
Operating Temperature
−20 °C to +105 °C
Packages
FC-CSP 17 x 17 mm or FC-BGA 23 x 23 mm
Security
TEE, TRNG, ECDSA, RSA4096, AES, SM4, DES, HMAC, CRC32, 16 KB OTP
NPU Frameworks
PyTorch, TensorFlow, PaddlePaddle and ONNX
Ecosystem Hardware
Sipeed LM5A SoM, SiFive HiFive Premier P550, StarPro64 boards
Overview
The ESWIN EIC7700X is a quad-core 64-bit RISC-V SoC built for edge AI vision. Its CPU cluster uses four SiFive Performance P550 cores running the RV64GC instruction set at up to 1.8 GHz — roughly Cortex-A75-class throughput — backed by 32 KB + 32 KB L1, 256 KB L2 and 4 MB of shared L3 cache with SECDED ECC protection.
AI acceleration comes from a 19.95 TOPS INT8 neural accelerator with BF16 support, a vision DSP cluster carrying 512 INT8 SIMD lanes, and a dedicated vision engine for 2D blit, crop, resize and normalisation with three-layer OSD. A separate Imagination AXM-8-256 GPU supplies OpenGL ES 3.2, Vulkan 1.2 and OpenCL 2.1 support plus Android NN HAL, so the same silicon can drive graphics-capable edge devices.
The multimedia and I/O budget is unusually generous for an edge part: H.265 decode to 8K at 50 fps or 32 simultaneous 1080p30 streams, H.265 encode to 8K at 25 fps, up to 32 GB of LPDDR5, eMMC 5.1 and SATA III storage, 4-lane PCIe 3.0, dual Gigabit Ethernet and up to six camera inputs. Typical consumption is about 8 W with a CNN running, over a −20 °C to +105 °C range. ESWIN scales the same design into the EIC7700 (13.3 TOPS), the 8-core EIC7702 (26.6 TOPS) and the EIC7702X (39.9 TOPS with up to 64 GB memory) for AI-PC-class and generative edge workloads.
Key Benefits
Full RISC-V stack with four SiFive P550 cores and a PyTorch/TensorFlow/PaddlePaddle/ONNX-compatible NPU; high vision throughput decoding 32x 1080p30 streams on one chip; four family tiers from 13.3 to 39.9 TOPS sharing one architecture; low thermal load at roughly 8 W typical in a 17 x 17 mm package; and industrial temperature range to +105 °C for sealed or fanless enclosures.
Applications
Industrial quality inspection and intelligent sorting; video analytics and behaviour recognition; secure and face identification; large language model inference at the edge; speech synthesis; image and video generation on AI-PC-class designs; and multi-camera vision gateways.
Request a Quote — ESWIN EIC7700X — QUAD-CORE RISC-V EDGE AI SOC WITH 19.95 TOPS NPU
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →