P3 instances are the next-generation of EC2 general-purpose GPU computing instances, powered by up to 8 of the latest-generation NVIDIA Tesla V100 GPUs. These new instances significantly improve performance and scalability, and add many new features, including new Streaming Multiprocessor (SM) architecture for machine learning (ML)/deep learning (DL) performance optimization, second-generation NVIDIA NVLink high-speed GPU interconnect, and highly tuned HBM2 memory for higher-efficiency.
G3 instances use NVIDIA Tesla M60 GPUs and provide a high-performance platform for graphics applications using DirectX or OpenGL. NVIDIA Tesla M60 GPUs support NVIDIA GRID Virtual Workstation features, and H.265 (HEVC) hardware encoding. Each M60 GPU in G3 instances supports 4 monitors with resolutions up to 4096x2160, and is licensed to use NVIDIA GRID Virtual Workstation for one Concurrent Connected User. Example applications of G3 instances include 3D visualizations, graphics-intensive remote workstation, 3D rendering, application streaming, video encoding, and other server-side graphics workloads.
The new NVIDIA Tesla V100 accelerator incorporates the powerful new Volta GV100 GPU. GV100 not only builds upon the advances of its predecessor, the Pascal GP100 GPU, it significantly improves performance and scalability, and adds many new features that improve programmability. These advances will supercharge HPC, data center, supercomputer, and deep learning systems and applications.
P3 instances with their high computational performance will benefit users in artificial intelligence (AI), machine learning (ML), deep learning (DL) and high performance computing (HPC) applications. Users include data scientists, data architects, data analysts, scientific researchers, ML engineers, IT managers and software developers. Key industries include transportation, energy/oil & gas, financial services (banking, insurance), healthcare, pharmaceutical, sciences, IT, retail, manufacturing, high-tech, transportation, government, and academia, among many others.
GPU-based compute instances provide greater throughput and performance because they are designed for massively parallel processing using thousands of specialized cores per GPU, versus CPUs offering sequential processing with a few cores. In addition, developers have built hundreds of GPU-optimized scientific HPC applications such as quantum chemistry, molecular dynamics, and meteorology, among many others. Research indicates that over 70% of the most popular HPC applications provide built-in support for GPUs.
P2 instances use NVIDIA Tesla K80 GPUs and are designed for general purpose GPU computing using the CUDA or OpenCL programming models. P2 instances provide customers with high bandwidth 25 Gbps networking, powerful single and double precision floating-point capabilities, and error-correcting code (ECC) memory, making them ideal for deep learning, high performance databases, computational fluid dynamics, computational finance, seismic analysis, molecular modeling, genomics, rendering, and other server-side GPU compute workloads.
P3 Instances are the next-generation of EC2 general-purpose GPU computing instances, powered by up to 8 of the latest-generation NVIDIA Volta GV100 GPUs. These new instances significantly improve performance and scalability and add many new features, including new Streaming Multiprocessor (SM) architecture, optimized for machine learning (ML)/deep learning (DL) performance, second-generation NVIDIA NVLink high-speed GPU interconnect, and highly tuned HBM2 memory for higher-efficiency.
Amazon EC2 F1 is a compute instance with programmable hardware you can use for application acceleration. The new F1 instance type provides a high performance, easy to access FPGA for developing and deploying custom hardware accelerations.
There are two cases where developers would choose EI over Inf1 instances: (1) if you need different CPU and memory sizes than what Inf1 offers, then you can use EI to attach acceleration to the EC2 instance with the right mix of CPU and memory for your application (2) if your performance requirements are significantly lower than what the smallest Inf1 instance provides, then using EI could be a more cost effective choice. For example, if you only need 5 TOPS, enough for processing up to 6 concurrent video streams, then using the smallest slice of EI with a C5.large instance could be up to 50% cheaper than using the smallest size of an Inf1 instance.
AWS Neuron is a specialized SDK for AWS Inferentia chips that optimizes the machine learning inference performance of Inferentia chips. It consists of a compiler, run-time, and profiling tools for AWS Inferentia and is required to run inference workloads on EC2 Inf1 instances. On the other hand, Amazon SageMaker Neo is a hardware agnostic service that consists of a compiler and run-time that enables developers to train machine learning models once, and run them on many different hardware platforms.
Trn1 instances are a good fit for your natural language processing (NLP), large language model (LLM), and computer vision (CV) model training use cases. Trn1 instances focus on accelerating model training to deliver high performance while also lowering your model training costs. If you have ML models that need third-party proprietary libraries or languages, for example NVIDIA CUDA, CUDA Deep Neural Network (cuDNN), or TensorRT libraries, we recommend using the NVIDIA GPU-based instances (P4, P3).
Amazon EC2 allows you to choose between Fixed Performance Instances (e.g. C, M and R instance families) and Burstable Performance Instances (e.g. T2). Burstable Performance Instances provide a baseline level of CPU performance with the ability to burst above the baseline.
T2 instances provide a cost-effective platform for a broad range of general purpose production workloads. T2 Unlimited instances can sustain high CPU performance for as long as required. If your workloads consistently require CPU usage much higher than the baseline, consider a dedicated CPU instance family such as the M or C.
Amazon EC2 T4g instances are the next-generation of general purpose burstable instances powered by Arm-based AWS Graviton2 processors. T4g instances deliver up to 40% better price performance over T3 instances. They are built on the AWS Nitro System, a combination of dedicated hardware and Nitro hypervisor.
T4g instances deliver up to 40% better price performance over T3 instances for a wide variety of burstable general purpose workloads such as micro-services, low-latency interactive applications, small and medium databases, virtual desktops, development environments, code repositories, and business-critical applications. Customers deploying applications built on open source software across the T instance family will find the T4g instances an appealing option to realize the best price performance within the instance family. Arm developers can also build their applications directly on native Arm hardware as opposed to cross-compilation or emulation.
Until December 31, 2023, all AWS customers will be enrolled automatically in the T4g free trial as detailed in the AWS Free Tier. During the free-trial period, customers who run a t4g.small instance will automatically get 750 free hours per month deducted from their bill during each month. The 750 hours are calculated in aggregate across all Regions in which the t4g.small instances are used. Customers must pay for surplus CPU credits when they exceed the instances allocated credits during the 750 free hours of the T4g free trial program. For more information about how CPU credits work, see Key concepts and definitions for burstable performance instances in the Amazon EC2 User Guide for Linux Instances.
Q: Who is eligible for the T4g free trial All existing and new customers with an AWS account can take advantage of the T4g free trial. The T4g free trial is available for a limited time until December 31, 2023. The start and end time of the free trial are based on the Coordinated Universal Time (UTC). The T4g free trial will be available in addition to the existing AWS Free Tier on t2.micro/t3.micro. Customers who have exhausted their t2.micro (or t3.micro, depending on the Region) Free Tier usage can still benefit from the T4g free trial.
Q: What is the regional availability of T4g free trial The T4g free trial is currently available across these AWS Regions: US East (Ohio), US East (N. Virginia), US West (N. California), US West (Oregon), South America (Sao Paulo), Asia Pacific (Hong Kong), Asia Pacific (Mumbai), Asia Pacific (Seoul), Asia Pacific (Singapore), Asia Pacific (Sydney), Asia Pacific (Tokyo), Canada (Central), Europe (Frankfurt), Europe (Ireland), Europe (London), and Europe (Stockholm). It is currently not available in the China (Beijing) and China (Ningxia) Regions.
As part of the free trial, customers can run t4g.small instances across one or multiple Regions from a single cumulative bucket of 750 free hours per month until December 31, 2023. For example, a customer can run t4g.small in Oregon for 300 hours for a month and run another t4g.small in Tokyo for 450 hours during the same month. This would add up to 750 hours per month of the free-trial limit.
Q: Is there an additional charge for running specific AMIs under the T4g free trial Under the t4g.small free trial, there will be no Amazon Machine Image (AMI) charge for Amazon Linux 2, RHEL and SUSE Linux AMIs that are available through the EC2 console Quick Start for the first 750 free hours per month. After 750 free hours per month, regular On-Demand prices, including AMI charge (if any), will apply. The applicable software fees for AWS Marketplace offers with AMI fulfillment options is not included in the free trial. Only the t4g.small infrastructure cost is included and covered under the free trial. 1e1e36bf2d