Amazon Web Services (AWS) announced the general availability of Amazon Elastic Compute Cloud P4d instances, Amazon EC2: the new generation of GPU-enabled instances, compared to P3 instances of the previous generation, delivers up to three-time performance

The P4d instances are equipped with eight Nvidia A100 Tensor Core GPUs and 400 Gbps network bandwidth (16 times more than P3 instances).

By using the P4d instances with AWS’s Elastic Fabric Adapter (EFA) and Nvidia GPUDirect RDMA (remote direct memory access), Amazon AWS customers are able to

With the EC2 UltraClusters, customers can scale P4d instances to over 4,000 A100 GPUs (double more than any other cloud provider, according to Amazon) using non-block networking infrastructure

The increase in the performance of P4d instances accelerates the training times of machine learning models up to three times, ensures Amazon AWS, reducing training times from days to hours; moreover, the additional memory of the GPU helps the

Data is becoming increasingly abundant for cloud service customers to train models with millions and sometimes billions of parameters, such as those used for the processing of natural language for document synthesis and the answer to questions, detection and classification of objects for autonomous vehicles, and the

P4d instances are equipped with 8 Nvidia A100 Tensor Core GPUs that can offer up to 2.5 petaflop mixed-precision performance and 320 GB high bandwidth GPU memory across an EC instance P4d instances are the first in the industry, says Amazon AWS, to offer a network bandwidth of 400 Gbps with Elastic Fabric Adapter (EFA) and Nvidia GPUDirect RDMA network interfaces, to allow

Each instance P4d, has illustrated again Amazon AWS, also offers 96 vCPU Intel Xeon Scalable (Cascade Lake), 1.1 TB system memory and 8 TB local NVMe memory to reduce individual training times

By increasing performance more than twice the previous generation of P3 instances, P4d instances can reduce the training costs of machine learning models by up to 60%, Amazon AWS highlights, providing customers with greater efficiency than on-premise systems

HPC customers can also benefit from the increased processing performance of P4d instances and GPU memory for challenging workloads such as seismic analysis, drug discovery, DNA sequencing, material science and risk modelling

Leave a Reply

Your email address will not be published.

You May Also Like