The Chinese tech giant Huawei this week officially launched their long-awaited Ascend 910 processor which they claimed to be the world’s most powerful artificial intelligence (AI) processor.
Huawei said that Ascend 910 processor chip consumes 310W of power, which is less than what has originally expected a figure of 350W. Huawei is essentially promising faster AI computing but at lower and efficient power consumption rates with the new chip.
The Ascend 910 is a new AI processor that belongs to the company’s series of Ascend-Max chipsets. Along with the Ascend 910 processor, the company also announced an AI computing framework called MindSpore.
Huawei has said, the chipsets will be embedded in the company’s acceleration cars, cloud services, etc. “We have been making steady progress since we announced our AI strategy in October last year.”
“Moving forward we will continue to invest in AI chipsets. Our plan is to launch Ascend 320 in 2021 & Ascend 920 so that we have a complete AI computing infrastructure,” said Huawei’s rotating chairman.
Speaking about Ascend 910, the processor is said to deliver 256 TeraFLOPS for half-precision floating points (FP16) and 512 TeraFLOPS for integer precision calculations (INT8) respectively.
The chip apparently takes 310W of power which is less than what was originally expected. The company’s original specs for the chip projected 350W power consumption from the Ascend 910 chip.
This is important because higher power consumption also increases the cooling costs for a data center, which are already through the roof when it comes to AI computing.
The chipset is used for AI model training and in a typical training session based on ResNet-50.
The combination of Ascend 910 and MindSpore is about two times faster at training AI models than other mainstream training cards using TensorFlow.
Huawei’s AI computing MindSpore offers easy development, efficient execution and is adaptable to all scenarios. The framework is essential for enabling secure, pervasive AI.
MindSpore helps ensure user privacy. Furthermore, MindSpore has 20% fewer lines of core code than leading frameworks on the market, and it helps developers raise their efficiency by at least 50%.
“In a typical neural network for natural language processing (NLP), MindSpore has 20% fewer lines of core code than leading frameworks on the market, and it helps developers raise their efficiency by at least 50%,” Huawei said.
MindSpore also supports other processors as well as CPUs and GPUs in addition to Ascend.
“MindSpore will go open source in the first quarter of 2020. We want to drive broader AI adoption and help developers do what they do best,” said Xu.
Huawei’s AI portfolio covers all deployment scenarios, it includes the Ascend IP & chip series, chip enablement layer CANN, training & inference framework MindSpore, & application enablement platform ModelArts.
Huawei believes that AI will be used in almost every sector of the economy but there are a number of gaps to close before AI can become a true general-purpose technology.
If we talk about Huawei competition, Nvidia has already led the market, by repurposing its graphics processing units (GPUs) for use in data centers and the AMD, has done the same by announcing 7nm chips recently.
On asked what will the company’s strategy be in India, especially since the uncertainty regarding Huawei’s 5G rollout in the country, Xu said, “Our AI strategy is not necessarily tied to 5G. As for our AI strategy in India, it is no different from our overall strategy.”
Huawei will develop Atlas as well as MDC products based on Ascend processors, which can be provided to universities and other partners in India as they develop applications to address industry-specific challenges.
“Soon, we will have training and inference class services based on Ascend processors as well which can also be made available in the Indian market.”
More in AI