Running AI models on a regular server can be slow and costly. When inference takes minutes instead of seconds, projects miss deadlines and budgets blow up. The Axelera AI Metis 4-Chip PCIe AI Accelerator Card, 16GB Active Cooling solves this by giving you dedicated AI power that fits in a single PCIe slot. It speeds up processing, cuts energy use, and lets you keep your existing hardware while adding a strong AI engine.
Axelera AI Metis 4-Chip PCIe AI Accelerator Card, 16GB Active Cooling with Axelera
Imagine a card that turns a slow workstation into a fast AI machine. That is what the Axelera Metis card does. It is a PCIe Gen 3.0 x16 accelerator that holds four Metis AI processing units and 16 GB of memory. Developers and engineers use it to run deep-learning inference without waiting for cloud resources. If you struggle with long training times or limited GPU slots, this card gives you the speed you need on-premise.
Key Features
- Four Metis AI processing units for parallel neural network execution.
- 16 GB system memory with options up to 64 GB for large models.
- Peak INT8 performance of 856 TOPS for high-speed inference.
- PCIe Gen 3.0 x16 single-slot HHHL form factor fits most servers.
- Active air cooling system maintains safe temperature from 0 C to 60 C.
- Power consumption between 30 W and 58 W for energy-efficient operation.
- 15 TOPS per watt efficiency improves total cost of ownership.
- Secure Boot and Root of Trust protect firmware and data.
- Voyager SDK and ready-to-use AI pipelines simplify development.
- Supports dozens of cameras on one board for large-scale vision tasks.
Specifications
| General |
|
| Product |
Metis 4-chip PCIe AI Accelerator Card |
| Form Factor |
PCIe single slot, HHHL (Half Height, Half Length) |
| Host Interface |
PCIe Gen. 3.0 x16 |
| AI Processing Unit (AIPU) |
4 × Metis AIPU |
| System Memory |
16GB (64GB available, please contact us) |
| Peak INT8 Performance |
856 TOPS |
| Operating Temperature |
0°C to 60°C |
| Thermal Solution |
Active air cooling |
| Typical Power Consumption |
30 W – 58 W |
| Security Features |
Secure Boot, Root of Trust |
| Performance |
|
| Energy Efficiency |
15 TOPS/W |
| AI Performance |
Up to 856 TOPS |
| Scalability |
~4× the performance of a single Metis PCIe |
| Camera Support |
A single board can run inference on dozens of cameras |
| Parallel Processing |
Supports multiple parallel neural networks |
| Software Stack |
Voyager® SDK |
| AI Models |
Wide range of end-to-end AI pipelines and models available out of the box |
| Quantization |
Advanced quantization tools for uncompromised prediction accuracy |
| Mechanical |
|
| Form Factor |
PCIe single slot, Half Height, Half Length (HHHL) |
| Packing List |
|
| Included Items |
Metis 4-chip PCIe card (as per configuration chosen in the order) with a Standard (full height) bracket affixed
1 × spare low profile (half height) bracketQuick-start flyer
|
