SAKURA-II AI Accelerator
SAKURA®-II supports the latest Generative AI solutions from Vision to multi-model, and is driven by DNA, a neutral IP core with runtime reconfigurable interconnects between compute units, achieving exceptional parallelism and AI efficiency.
Key Benefits
Optimized for Generative AI
Supports multi-billion parameter Generative AI models like Liama 2. Stable Diffusion, DETR, and VIT within a typical power envelope of BW
Efficient AI Compute
Achieves more than 2x the AI compute utilization of other solutions, resulting in exceptional energy efficiency
Enhanced Memory Bandwidth
Up to 4x more DRAM bandwidth than competing AI accelerators, ensuring superior performance for LLMs and LVMs
Large DRAM Capacity
Support for up to 32GB of DRAM enabling efficient processing of complex vision and Generative AI workloads
Real-Time Data Streaming
Optimized for low-latency operations with Batch=1
Arbitrary Activation Function Support
Hardware-accelerated approximation provides enhanced adaptability
Advanced Precision
Software-enabled mixed-precision provides near FP32 accuracy
Efficient Data Handling
Integrated tensor reshaper engine minimizes host CPU load
Sparse Computation
Reduces memory footprint and optimizes DRAM bandwidth
Power Management
Advanced power management enables ultra-high efficiency modes
Sakura-II Offering
Model Resources
Model Zoo: Pre-trained optimized AI inference models
Support for popular Generative AI models, including Llama-2, Stable Diffusion, Whisper, DETR, DistillBert, DINO and ViT
Post training model calibration and quantization
Best For
Transportation/Autonomous Vehicles
Defense/Aerospace
Security
5G Communications
Augmented & Virtual Reality
Smart Manufacturing/Robotics
Smart Cities
Smart Retail
Drones & Robotics
