Axelera AI introduces Metis AI in card, board and vision system formats
Accelerating computer vision at the edge, the Metis AI platform has been announced by Axelera AI, a start up based in the Netherlands.
It consists of the Metis AI processing unit (AIPU) chip and the software stack Voyager software development kit. Axelera AI offers AI acceleration cards, boards and vision systems to accelerate computer vision at the edge designed to accelerate developers through the evaluation and development cycles and shorten time-to-revenue.
The Metis AIPU is a quad core architecture. Each core can execute all layers of a standard neural network without external interactions delivering up to 53.5 TOPS of AI processing power for a compound throughput of 214 TOPS. This makes it, claims Axelera AI, the most powerful chip for edge devices, outperforming currently available products.
The cores can either be combined to boost throughput or operate independently on the same neural network to reduce latency. Another option is for it to process concurrently different neural networks for applications featuring pipelines of neural networks. Axelera’s SRAM-based D-IMC engine accelerates matrix-vector multiplication operations, for an energy efficiency of 15 TOPs/W at INT8 precision. FP32 iso-accuracy is achieved without the need to retrain the neural network models, added Axelera AI.
The four cores are integrated into an SoC, comprising RISC-V controller, PCIe interface, LPDDR4X controller and a security complex connected via a high speed network on chip (NoC). The NoC also connects the cores to a multi-level shared memory hierarchy of more than 52Mibyte of on chip, high speed memories. The LPDDR4 controller connects to external memory which enables support for larger neural networks.
A PCIe interface provides a high speed link to an external host, which will offload full neural network applications to the Metis AIPU.
The PCIe card is powered by four Metis AIPUs and delivers up to 856 TOPs, powered by four Metis AIPUs.
The software stack, Voyager SDK, includes an integrated compiler, a runtime stack and optimisation tools that enable customers to import their own models. Application templates speed up development; customers can choose from Axelera’s curated templates in the Model Zoo. This is a repository for neural networks which can be customised, fine-tuned, deployed as-is. Customer’s own pre-trained neural networks can also be imported.
The platform is claimed to be the first to offer hundreds of TOPS performance at “an edge price point”: the 214 TOPS Edge M.2 acceleration card is $149, corresponding to $0.70 per TOPS.
“The Metis AIPU . . . demonstrates the effectiveness of Axelera’s D-IMC engine by achieving matrix-vector multiplications with high precision and constant time complexity, which in tandem with the advanced quantisation algorithms, the large on-chip SRAM and the RISC-V technology are well-suited to address the stringent operational requirements of the AI applications at the edge,” said Axelera AI’s co-founder and CTO Evangelos Eleftheriou.
“Axelera’s end-to-end integrated software stack, built on open, industry-standard frameworks and APIs, addresses the user pain points of AI application development at the edge,” he added.
The Metis AI Platform enables customers to rapidly develop and deploy computer vision applications. The Voyager SDK provides pipelines for models, which a user may customise for an application, generate optimised code for, and deploy to Metis-enabled edge devices.
The platform is intended for a range of environments, from fully embedded use cases to distributed processing of multiple 4K video streams across networked devices.
The Voyager SDK automatically quantises and compiles neural networks that have been trained on different frameworks, generating code which runs on the Metis AI platform. Optimised networks running on Metis AIPU are indistinguishable from those running on systems with floating-point units, said Axelera AI.
The early access program opens today. The Metis platform will be exhibited at CES 2023 next month in Las Vegas, USA.
