Every chip requires design tradeoffs including Performance, Power, Area, Cost, Scalability, Easy of Use, Thermals, BOM, etc.
MemryX products must meet 5 customer-centric criteria
MemryX MX3
Edge AI Accelerator
Each MX3 Chip offers
- >5 TFLOPs
- Activations: Bfloat16
- Weights: 4 / 8 / 16 bit
- Batch = 1
- ~10M parameters stored on-die
- USB 2.0/3.x and/or PCIe Gen 3 I/O
- ~1.0W average power
- MX-SDK with 1-click compilation
MX3 Form Factors
MX-SDK
The MemryX SDK is designed to allow developers to create systems and applications that optimally utilize MemryX AI accelerator chip(s). The goal of the SDK is to be simple, powerful, and flexible. You won’t find a complex code development environment or any requirement to change already trained AI model(s). Instead, you will be able to quickly simulate & deploy trained AI model(s) while choosing the ideal balance of performance, power, latency, and utilization for any application.
