FPGA-Powered AI Accelerator
Accelerating Conversational AI Algorithms
With AI algorithms changing constantly, don’t be dependent on a generic library with lackluster performance and minimal support.
Revolutionize Your ASR Capabilities with Industry-Leading Ultra-Low Latency.
Contact Us
Videos
ASR Solution Highlights
2,000 concurrent real-time audio streams
20x reduction in latency (<50 ms end-to-end latency)
Competitive accuracy
Retrainable solution to any language or specialized vertical market with a provided dataset
Replaces the equivalent processing power of over 20 CPU-based servers or 15 GPU cards
Slashes CapEx by up to 90% compared to CPU-based servers
Saves millions in OpEX with cloud costs
No RTL or FPGA knowledge required
Compatible with all major deep learning frameworks such as PyTorch
AC7t1500 FPGA Device Highlights
Built on TSMC's 7nm FinFET process
692K 6-input LUTs (1,522K 4-input LUT equivalents
2,560 MLP blocks capable of 41K Int8 operations per cycle
2D Network on Chip (NoC) capable of >20 Tbps
Up to 61 TOps int8 performance
Up to 4 Tbps of GDDR6 bandwidth
Target Applications
Call Centers and customer service centers
- Transcription
- Voice assistants
