Amazon Web Services (AWS) has launched EC2 instances, which they claim are specifically designed for deep learning training.
The new Amazon EC2 Trn1 instances are powered by AWS Trainium chips, a second-generation ML chip developed by AWS, following on from its AWS Inferentia chips.
The cloud giant claims that these new instances are effective in large-scale distributed training of complex deep learning techniques, such as natural language processing and image recognition.
What do users get?
Trn1 instances are available in two versions, each of which is powered by up to 16 AWS Trainium chips with 128 vCPUs.
These instances might boast up to 512 GB of high-bandwidth memory and 3.5 petaFLOPS of TF32/FP16/BF16 compute power. NeuronLink is a chip that connects to different devices. When it comes to scaling workloads, magnets are assumed to be ineffective.
Trn1 instances are the first EC2 instances to allow up to 800 Gbps of elastic fiber Adapter network bandwidth for high-throughput network communication. AndTrn1 instances come with up to 8 TB of local NVMe SSD storage for ultra-fast access to large datasets.
AWS claims that its Trainium chips include specic scalar, vector, and tensor engines that are intended for deep learning algorithms.
Trainium chips include support for a broad spectrum of data types, including FP32, TF32, BF16, and UINT8, Stochastic rounding, as well as custom operators written in C++ and dynamic tensor shapes.
The sameAWS Neuron SDKas AWS Inferentia are part of AWS Trainium, which might make the transition to AWS Trainium easier.
Where can I sign up?
In certain areas such as the AWS US East (N. Virginia) and US West (Oregon), you may launch Trn1 instances today.
These Trn1 instances may be deployed using AWS Deep Learning AMIs, and container images may be downloaded through managed services such as Amazon SageMaker, Amazon Elastic Kubernetes Service (Amazon EKS), Amazon Elastic Container Service (Amazon ECS) and AWS ParallelCluster.
To learn more, visitAmazon''s Trn1 instances page (opens in a new tab) on this page.