Compute
AI-class compute on the board
The HK800CA CPU field reads AMD Ryzen AI 9 365. The TOP800B runs Core Ultra 5/7 at 28W for desktop-class inference headroom.
Industry route · Edge AI
Inference where the cameras are, not where the cloud is.
Edge AI moves the model to the site: NVR rooms, camera cabinets, gateways at the network edge. The board needs compute class, an accelerator slot and a way to report back. These three models quote those fields on their datasheets.
Ryzen AI 9 365
M.2 PCIe 5.0
4G/5G backhaul
What the inference node asks for
What the application demands
An inference node is sized by the model it runs and the site it sits in. The three checks below come before the CPU selection.
Accelerator
Bandwidth for the model
The RP7003AW quotes an M.2 2280 slot on PCIe 5.0 x4. The TOP800B adds a PCIe 4.0 x4 slot beside its M.2 2280 NVMe.
Memory and link
Buffers and a way home
DDR5 to 96GB on the TOP800B and RP7003AW holds multiple models and video buffers. A B-Key slot carries 4G/5G at sites without wired uplink.
Recommended boards
Three nodes for the inference fleet.
CPU fields, slots and memory are quoted from the model pages. The AI-capability reading stops where the datasheet stops.
Ryzen AI
HK800CA
Why: Ryzen AI 9 365 on the CPU field with three 4K-capable outputs and dual 2.5G LAN; on-device inference without a separate accelerator.
Ryzen AIDDR5 64GBAT mode
View board
Core Ultra
TOP800B
Why: Core Ultra 5/7 at 28W, DDR5 to 96GB, a PCIe 4.0 x4 slot and M.2 2280 NVMe; B-Key 4G/5G with Micro-SIM.
HDMI 2.1 8KPCIe 4.0 x44G/5G
View board
12th / 13th Gen
RP7003AW
Why: M.2 2280 on PCIe 5.0 x4, DDR5 to 96GB on 13th Gen and 2× CAN; a B-Key takes 4G/5G or an SSD.
PCIe 5.02× CANDC 9–36V
View board
Deployment flow
Size the node around the model.
Inference type, batch size and latency budget set the CPU class; site power and uplink set the board.
Talk through the workloadFit matrix
Match the model, then the node.
Accelerator slots, memory and backhaul against the three recommended models. Values are quoted from the model pages.
3 models · 2 requirements
Edge AI fit
Slot generation, memory and the cellular path decide the node. Confirm revision and configuration before design freeze.
Swipe to compare →| Model | Accelerator & storage | Memory & backhaul | Why it fits |
|---|---|---|---|
| HK800CA | M.2 2280 NVMe PCIe 4.0 x4 | DDR5 64GB · dual i226-V 2.5G | Ryzen AI 9 365 on the CPU field for on-site inference nodes. |
| TOP800B | PCIe 4.0 x4 slot · M.2 2280 NVMe | DDR5 96GB · B-Key 4G/5G + Micro-SIM | Desktop-class headroom for an NVR or inference box, cellular backhaul included. |
| RP7003AW | M.2 2280 PCIe 5.0 x4 | DDR5 to 96GB (13th Gen) · B-Key 4G/5G or SSD · 2× CAN | PCIe 5.0 bandwidth for accelerator caches and a cellular slot for camera sites. |
Source: HK800CA, TOP800B and RP7003AW datasheets
Compare the full lineupQuestions
Edge AI, answered.
Compute class and backhaul. The two answers that size an inference node.
The HK800CA carries AMD Ryzen AI 9 365 on its CPU field. The TOP800B runs Core Ultra 5/7 at 28W for desktop-class inference headroom.
Start a project
Send the workload and the site power.
We size the node within 24 to 48 hours, from CPU class to accelerator slot and backhaul.
GEOLiNK / Shenzhen