New Amazon EC2 P6E-GB200 UltraSERVERS Accelerated NVIDIA GRACE Blackwell GPU for the highest performance AI | Amazon Web Services

Today, we announce the general availability of the Amazon Elastic Compute Cloud (AMAZON EC2) P6-GB200 Ultrservers, accelerated by NVIDIA GB200 NVL72 to offer the highest GPU performance for training and inference AI. Amazon EC2 UltraSERVERS combines several EC2 instances by means of a reserved high -width and low latency connection in these cases.

Superchips Nvidia Grace Blackwell combine two high-performance NVIDIA Blackwell GPUs and NVIDIA Grace CPUs based on ARM architecture by connecting NVIDIA NVLink-C2C. Each Grace Blackwell Superchip releases 10 FP8 Compute (without Sparsity) and up to 372 GB HBM3E memory. With the Superchip architecture, the GPU and CPU are colored within a single computing module, which increases the bandwidth between the GPU and the Nordicnenly CPU compared to the instances of the current EC2 P5EN generation.

With the UltraSERVERS EC2 P6E-GB200, you can access up to 72 GPU NVIDIA Blackwell GPUs in one NVLink domain to use 360 FP8 Compute (without Sparsity) and 13.4 TB of high-high bandwidth (HBM3E). The P6E-GB200 Ultraservers, powered by AWS Nitro, are deployed in the EC2 ultra-clubs to scaling safely and reliably to tens of thousands of GPUs.

The UltraSERVERS EC2 P6-GB200 adds up to 28.8 Tbps of the total elastic fabric adapter (EFAV4). EFA is also associated with NVIDIA GPUDirect RDMA to allow communication with low latency GPU-GPUs between operating system bypass servers.

EC2 P6E-GB200 Ultraservers Special
UltraSERVERS EC2 P6E-GB200 are available in sizes from 36 to 72 GPU under NVLink. Here are the specifications for UltraSERVERS EC2 P6E-GB200:

The UltraSERVERS P6E-GB200 is ideal for the most computing and most modern AI workload, such as training and inference of border models, including a mixture of professional models and thinking models, on a trillion parameter scale.

You can build agent and generative AI applications, including answers to questions, generating code, video and images generation, speech recognition and more.

P6E-GB200 Ultraservers in Action
The UltraSERVERS EC2 P6E-GB200 can be used in the Dallas local zone via EC2 blocks for ml. Local zone Dallas (us-east-1-dfw-2a) is the expansion of the US East (N. Virginia).

If you want to book EC2 capacity blocks, choose Reservation of capacity On Amazon EC2. You can select Purchase capacity blocks for ml and then select your total capacity and specific how long you need an EC2 capacity block U-P6E-GB200x36 gold U-P6E-GB200x72 Ultraservers.

Once the capacity block is successfully planned, it is loaded in advance and its price will not change after purchase. Payment will be the ball for your birthday within 12 hours after buying the EC2 capacity blocks. If you want to learn more, visit the ML capacity blocks in the Amazon EC2 user manual.

If you want to run an instance with a block of purchased capacity, you can use the AWS Management Console, the AWS (AWS CLI) or AWS SDKS command line. On the software side you can start with AWS Deep Learning friends. These images are configured in advance with and tools that you probably already know and use: Pytorch, Jax and much more.

You can also smoothly manage the UltraSERVERS EC2 P6E-GB200 with different services. For example:

• Amazon Sagemaker Hyperpod has been administered, a durable infrastructure that automatically processes the provision and management of the UltraSERVERS P6E-GB200, replaces defective instances for the reconfigured alternative capacity with the same NVLink domain to maintain performance.
• Amazon Elastic Kubernetes Services (Amazon EKS) allows one group of managed nodes to exceed multiple ultrarvers P6-GB200 as nodes and automate their provision and control of life cycle in Kubernetes clusters. For the UltraSERVERS P6E-GB200, you can use the topological routing of EKS topology, enabling the optimal placement of fixed-connected components of distributed workloads within one ultraserver nvlink-connected instance.
• Amazon FSX for Luster file systems provides access to UltraServers P6-GB200 data when hanging GB/s throughput and millions of input/output operations per second (IOPS) required for extensive workload HPC and AI. For quick access to large data files, you can use up to 405 TB of local NVMe SSD storage or virtually unlimited cost -effective storage with Amazon Simple Storage Service (Amazon S3).

Now available
UltraSERvers Amazon EC2 P6E-GB200 are available today in the local zone of Dallas (us-east-1-dfw-2a) Via EC2 capacity blocks for ml. For more information, visit the Amazon EC2.

Try the UltraSERVERS Amazon EC2 P6E-GB200 in the Amazon EC2. If you want to know more, visit the Amazon EC2 P6E instance page, and send AWS Re: Post for EC2 or by the usual AWS support contacts.

– Channels