UPDATED 15:03 EDT / DECEMBER 01 2020

CLOUD

AWS previews ultra-efficient AI instances for neural network training

Amazon Web Services Inc. today previewed an upcoming cloud compute instance series that will enable companies to train artificial intelligence models in its cloud with up to 40% better price-performance than with graphics cards. 

The instance series is expected to become available in the first half of 2021. AWS announced it at its virtual re:Invent event today alongside several other new compute offerings optimized for use cases such as running databases and rendering video content.

AWS’ planned AI instances will be based on Intel Corp.’s Gaudi chips (pictured), which are specially optimized for training neural networks. Intel absorbed the product line as part of its $2 billion acquisition of semiconductor startup Habana Labs Ltd. last year. Instances will be available with up to eight Gaudi units, which, according to a blog post published today by Habana executive Eitan Medina, will provide enough computing power to train an AI on 12,000 images per second. That’s when using the industry-standard ResNet-50 model. 

But while impressive, the number of images that the chips can process per second may not be the data point that ends garnering the most attention among AWS customers. The cloud giant said that it expects the Gaudi-powered instances to provide an up to 40% better price-performance ratio for deep learning model training than the GPU-powered instances it currently offers. In other words, it could cost up to 40% less for enterprises to train AI models on AWS.

The cloud giant is introducing the Gaudi instances at an opportune time. AI models are getting more complex, partially because enterprise machine learning initiatives are maturing and partially because research conducted by the likes of OpenAI is facilitating bigger neural network architectures. As neural networks grow in complexity, the amount of computing power necessary to train them is increasing and fueling demand for more efficient training infrastructure. 

The announcement of the upcoming instances is also notable because it represents a big win for Intel. The chipmaker has in recent quarters worked to expand beyond its core focus area of central processing units to other areas, among them the AI chip market. The fact that cloud leader AWS is adopting Intel’s Gaudi chips, and that it expects those chips to provide highly competitive price-performance, could send a strong signal to other potential customers.

AWS’ adoption of Gaudi advances a broader strategy by the cloud giant to expand the number of chip options available via its cloud. As part of the same effort, the company has developed AWS Trainium, a homegrown AI training processor that was also announced today at re:Invent. 

AWS Trainium, AWS said, will provide the “best price performance for training ML models in the cloud.” The Amazon.com Inc. subsidiary didn’t go into too much detail about the chip’s technical specifications. However, it did disclose that AWS Trainium will be suitable for training a wide range of models including neural networks used for image classification, translation, voice recognition, natural language processing and product recommendation, among others.

The chip is expected to become available to customers in the second half of 2021. 

Image: Intel

A message from John Furrier, co-founder of SiliconANGLE:

Your vote of support is important to us and it helps us keep the content FREE.

One click below supports our mission to provide free, deep, and relevant content.  

Join our community on YouTube

Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.

“TheCUBE is an important partner to the industry. You guys really are a part of our events and we really appreciate you coming and I know people appreciate the content you create as well” – Andy Jassy

THANK YOU