Huawei may be battling a tough case against the American trade ban. However, that hasn’t stopped the company from developing and launching multiple products. In addition to the recently announced alternative to Google’s Android OS, Huawei has launched two interesting components to the increasingly attractive and competitive Artificial Intelligence (AI) space. While the Huawei Ascend 910 is reportedly the most powerful AI-focused processor, MindSpore is supposed to be all-scenario AI computing framework. Together, the processor and framework should help developers to quickly train their AI-based platforms, claims the company.
“Ascend 910 and MindSpore marks a new stage in Huawei’s AI strategy,” noted Eric Xu, Huawei’s Rotating Chairman at a launch event. Ascend 910 is used for AI model training. Together, the duo is about two times faster at training AI models than other mainstream training cards using TensorFlow, claimed Huawei. The company has some ambitious plans for the processor and framework. With the claimed versatility, the duo can offer their ready-to-deploy services to edge computing, and on-vehicle computing for autonomous driving, and training.
Huawei Ascend 910 AI Processor Specifications And Features:
The Ascend 910 is a new AI processor that belongs to Huawei’s series of Ascend-Max chipsets. The Ascend 910 delivers 256 TeraFLOPS for half-precision floating-point (FP16) operations. For integer precision calculations (INT8), it delivers 512 TeraOPS. What’s even more surprising is the power draw efficiency of the Huawei Ascend 910 processor. While the company planned to keep the power draw below 350W, the Ascend 910 consumes even lower power at 310W.
We’ve ‘chosen’ to take A.I. to the next level. Introducing Huawei Ascend 910, the world's most powerful AI processor, and MindSpore, an all-scenario AI computing framework. https://t.co/tfu6dvaIgQ #Ascend910 #MindSpore pic.twitter.com/qqMrY7pv3T
— Huawei (@Huawei) August 23, 2019
Last year, Huawei announced its AI strategy and full-stack, all-scenario, AI portfolio. The company launched the Ascend 310 AI processor. Huawei also launched ModelArts that provides full-pipeline model production services. The Ascend 310 AI processor is Huawei’s first commercial AI System on a Chip (SoC) in the Ascend-Mini Series. The processor consumes just 8W power but still manages to deliver 16 TeraOPS in integer precision (INT8) and 8 TeraFLOPS in half-precision (FP16). The processor also packs a 16-channel FHD video decoder. Needless to mention, this makes the Huawei Ascend 310 the most powerful AI SoC, ideally suited for edge computing. Currently, the processor is found in Huawei’s Mobile Data Center (MDC), which is deployed in shuttle buses, new-energy vehicles, and autonomous driving, as it is best suited for smart transportation and smart grid applications.
Huawei MindSpore All-scenario AI Computing Framework Features:
In addition to the Huawei 910 AI processor, the company also launched MindSpore, an AI computing framework that reportedly supports development for AI applications in all scenarios. The framework is based on three core principles: Easy development, efficient execution, and adaptable to all scenarios. The framework is dedicated to offering a secure, pervasive AI. It deals with gradient and model information that has already been processed. Simply put, MindSpore does not process the data itself. This secures the data even in cross-scenario environments.
Huawei Ascend 910 Provides a NVIDIA AI Training Alternative in solutions designed to scale from IoT to large training clusters https://t.co/Eb4unKsWjU @Huawei @nvidia #DeepLearning #AI pic.twitter.com/S7Lpot2rSQ
— STH (@ServeTheHome) August 25, 2019
To top that, MindSpore has built-in model protection technology. This feature ensures all models remain secure and trustworthy. The versatile framework is adaptable to all scenarios. In other words, it can work across all devices, edge, and cloud environments, and can ensure on-demand dynamic cooperation between them. It appears Huawei has perfected a new “AI Algorithm As Code” design concept that can be offered to developers, who can, in turn, develop advanced AI applications with ease and train their models quickly.