Raptor Series
Gen Al Processor
Products
Viper Series
The Neuchips Viper series is an enterprise-focused offline AI solution featuring the Raptor N3000 Gen AI processor, advanced data compression, and 64GB onboard memory. It specializes in secure, efficient RAG applications while maintaining complete data privacy through local processing.
Overview
| Neuchips Products | Viper Series Gen Al PCIe Card | |||||
|---|---|---|---|---|---|---|
| Offline Al Solution | | | | | | |
| Domain Focus Al Application |
|
|
|
|
|
|
| System Integrator | HW & SW System Integration Service | Domain Specific API Management | |||||
| Software Service Provider / Application Interface | Device Management Platform | Fine Tuning | Document Partner | RAG... etc. | |||||
| Open Soruce Al Models | Gen Al/ LLM Models (Llama3.2 | Mistral NeMo | Phi 3.5 | TAIDE | Breeze) | |||||
| System Hardware Manufacturer | PC | IPC | Workstation | Server | |||||
Easy-to-Use SDK
| User Application | ||
| Software Development Kit (SDK) | ||
| Compiler | Neuchips PyTorch Extension | |
| FFP8 quantization | ||
| Operator optimization | ||
| Graph partition | ||
| Graph optimization | ||
| Memory planning | ||
| Shared Kernel | Custom kernels | |
| Runtime | ||
| Neuchips Raptor Series | ||
|---|---|---|
Transform Your Enterprise AI
Securely and Seamlessly
Securely and Seamlessly
The Viper GenAI PCIe card represents a breakthrough in enterprise AI solutions, combining powerful hardware acceleration with practical business functionality. At its core, it features the Raptor Gen AI processor with specialized embedding engines that deliver 10x efficiency improvements. The system's 32 GB onboard memory serves as a dedicated vector database, enabling secure local data processing while significantly reducing CPU overhead. By supporting multiple open-source LLM models and offering seamless integration with existing infrastructure, Viper enables enterprises to implement AI capabilities without compromising data security or requiring extensive system overhauls. This makes it particularly valuable for organizations seeking to leverage AI technology while maintaining complete control over their sensitive data and operational processes.
Specification
| Viper Series Specification | |
|---|---|
| Specification | Description |
| Support LLM Model | Llama3.2, Mistral NeMo, Phi3.5, Breeze, TAIDE |
| Total Board Power | Min. = 25W Default = 45W Max. = 75W |
| Thermal Solution | Active and Passive Cooling Available |
| Mechanical Form Factor | HHHL-SS (half-height, half-length, single-slot) |
| Memory Type | LPDDR5 |
| Memory Size | Up to 64GB |
| Memory Clock | 6400 Mbps |
| Ambient Operating Temperature | 0°C to 50°C |
| Storage Temperature | -40 °C to 75°C |
| Operating Humidity | 5% to 85% Relative Humidity |
| Storage Humidity | 5% to 95% Relative Humidity |
Key Features
Embedded Vector Processing Engine
- Built-in with efficient embedding engine
- Reduces communication overhead between the card and host CPU
- Provides 10x improvement in efficiency for vector similarity searches
- Features 64GB LPDDR5 memory capacity, supporting up to 12B LLM models in a single chip and card configuration
32GB On-Device Vector Storage
- Functions as an on-premise vector database
- Reduced cooling requirements
- Supports extensive vector data storage directly on the board
- Uses innovative quantization technique to compress vector data by 4x
LLM Model Support
- Compatible with pre-trained LLM models
- Supports Llama, Mistral, Phi, TAIDE, and Breeze
- Works with multiple language such as English and Traditional Chinese
Low Power Design
- Total board power of only 45W in average
- Operates in temperatures from 0°C to 50°C
- Support both passive and active thermal solution
Key Benefit
Maximize Savings, Minimize AI Costs
Transform AI Operation Instantly Efficiently
Efficiently Compact, No Hardware Hassle
Streamline Operations, Maximize Output
Offline Security, Zero Data Leaks
Low Power, High Sustainability
Ecosystem Alliance
- AIR-510
- MIC-770
