What happened
Nanyang Singtech, an NTU-incubated deep-tech company, unveiled the SingNova-H Studio, a new AI PC. This workstation features a RISC-V dataflow architecture system-on-chip, delivering 200 TOPS (INT8) and 32 TFLOPS (FP16) with 128 GB of LPDDR5 ECC unified memory. It enables on-device execution of large language models up to 70 billion parameters, ensuring data locality and privacy, per the company's announcement. The design overcomes "memory wall" limitations through control-data collaborative optimisation for high-concurrency, low-latency processing.
Why it matters
Local AI inference capabilities will shift procurement strategies for enterprises and security architects managing sensitive data. The SingNova-H Studio's 200 TOPS and 128 GB unified memory enable running 70B parameter models entirely on-device, reducing cloud dependency and eliminating network latency and data privacy risks. This RISC-V dataflow architecture offers a mechanism to bypass memory bottlenecks, providing high-concurrency, low-latency execution for edge AI workloads. Platform engineers should evaluate such local solutions for cost reduction and enhanced data control, particularly as Asian chipmakers intensify their focus on AI hardware.
