Specifications
Accelerator
Rockchip RK1820 / RK1828 AI coprocessor
AI Performance
20 TOPS (INT8)
Precision
INT4, INT8, INT16, FP8, FP16, BF16
On-chip DRAM
2.5 GB (RK1820) / 5 GB (RK1828) 3D-stacked
Memory Bandwidth
1024 GB/s
Form Factor
M.2 2280
Interface
Standard M.2 M-Key
Host Controllers
RK3568, RK3576, RK3588, RK3572
LLM Support
Up to 3B (RK1820) / 7B (RK1828) parameters
Throughput
100+ tokens/s (Qwen, LLaMA2 class models)
Process Node
20 nm
Onboard CPU
Triple-core RISC-V 64GCB
Typical Power
~5 W
Cooling
Passive cooling supported
Operating Systems
Linux, Android
Use Case
Offline GenAI inference without cloud dependency
Overview
The Forlinx RK1820 and RK1828 M.2 computing cards are dedicated AI accelerator modules built on Rockchip's RK182x NPU family. Each delivers 20 TOPS of INT8 compute with mixed-precision support across INT4, INT8, INT16, FP8, FP16 and BF16, and they fit the industry-standard M.2 2280 M-Key footprint.
Rather than relying on host DDR, the cards carry 2.5 GB (RK1820) or 5 GB (RK1828) of 3D-stacked high-bandwidth DRAM running at up to 1024 GB/s. That on-module memory removes the edge memory-wall bottleneck, allowing the RK1820 to run 3B-class and the RK1828 7B-class LLMs and VLMs at more than 100 tokens per second, entirely offline.
The cards act as AI coprocessors for Rockchip hosts including the RK3568, RK3576, RK3588 and RK3572, are served by Linux and Android drivers, and draw only about 5 W so they can be passively cooled in fanless edge systems.
Key Benefits
20 TOPS INT8 with broad mixed-precision support; 2.5 GB / 5 GB 3D-stacked DRAM at 1024 GB/s eliminates the memory wall; 100+ tokens/s offline LLM/VLM inference without cloud dependency; and a standard M.2 2280 M-Key interface with ~5 W passive cooling for drop-in upgrades.
Applications
Local private LLM and VLM deployment, industrial edge AI, generative-AI appliances, computer vision gateways, robotics, kiosks and fanless embedded platforms.
Request a Quote — FORLINX RK1820 / RK1828 M.2 COMPUTING CARD — 20 TOPS EDGE AI ACCELERATOR
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →