Last April, UNISOC introduced Tiger T310 quad-core Cortex-A75/A55 processor for 4G Phone, and the company has now added a new member to their Tiger family with T710 octa-core processor with four Cortex-A75 cores, four Cortex-A55 cores, and a dedicated NPU (Neural-network Processing Unit) for artificial intelligence applications at the edge in the industry, commerce, medical care, home, and education sectors.
The company did not setup a product page, so information is really limited. Known features & specifications so far:
- CPU – 4x Arm Cortex-A75 cores at up to 2.0GHz, 4x Arm Cortex-A55 cores @ 1.8GHz
- NPU – Heterogeneous dual-core architecture NPU delivering 2.5TOPS per watt
- GPU – Imagination Technologies PowerVR GM9446 GPU also found in MediaTek Helio P90.
The processor supports multiple AI framework formats, including TensorFlow, TensorFlow Lite, Caffe, and multiple AI data formats such as quantization (INT4, INT8, INT16) and floating-point (FP16). Tiger T710 also works with Android NNAPI, UNISOC provides a proprietary SDK for more efficient development and deployment of third-party AI applications.
That does not seem that impressive until you look at results from AI Benchmarks.
As of August 2019, Tiger T710 tops the AI benchmarks Mobile’s SoCs ranking with 28,097 points. That means for AI workloads it is faster than processors like Snapdragon 855 Plus or Huawei Kirin 980. It’s true at least for most benchmarks from AI Benchmark as sometimes Kirin 810 takes the lead.
The CPU and GPU, however, matches the performance of mid-range processors, so Tiger T710 really focuses on AI applications. You find more charts and benchmark on AI Benchmark blog post about the processor.
UNISOC Tiger T710 processor is expected to become available in 2020.
[Update: I’ve just noticed the company also launched Tiger T618 deca-core 4G mobile processor with two 2.0 GHz Arm Cortex-A75 and six 1.8 GHz Arm Cortex-A55 processors with Mali G52 MP2 GPU, 48MP camera support]