ARM today announced Project Trillium, a suite of products meant to bring machine learning to mobile devices.
While Project Trillium will first target smartphones, Arm says it will eventually scale to both less and more powerful devices as needed. The two central pieces of Project Trillium are the Arm Machine Learning core and the Arm Object Detection core.
ARM says the ML core can push 4.6 trillion operations per second while keeping thermals and power consumption in check. The Arm OD processor was built specifically to help computers recognize people, pets, and other objects visible in the frame. It can handle real-time recognition while recording full HD video at 60 frames per second.
Together, the Arm ML and Arm OD provide massive efficiency boosts to accompanying CPUs, GPUs, and accelerators, and can outperform traditional programmable logic on today’s DSPs.
The Arm ML and Arm OD are currently sampling and Arm expects to ship them in bulk by the middle of the year.
The components will help companies put together their own mobile processor modules that can perform calculations similar to those being made by the Qualcomm Snapdragon 845 and Huawei Kirin 970.
The Arm ML processor is built from the ground-up, specifically for ML. It is based on the highly scalable Arm ML architecture and achieves the highest performance and efficiency for ML applications:
- For mobile computing, the processor delivers more than 4.6 trillion operations per second
- (TOPs) with a further uplift of 2x-4x in effective throughput in real-world uses through intelligent data management
- Unmatched performance in thermal and cost-constrained environments with an efficiency of over three trillion operations per second per watt (TOPs/W).
The Arm OD processor has been designed specifically to efficiently identify people and other objects with virtually unlimited objects per frame:
- Real-time detection with Full HD processing at 60 frames per second
- Up to 80x the performance of a traditional DSP, and a significant improvement in detection quality relative to previous Arm technologies.
