Amazon EC2 G4 Instances
The industry’s most cost-effective GPU instances for machine learning inference and graphics-intensive applications
Amazon EC2 G4 instances are the industry’s most cost-effective and versatile GPU instances for deploying machine learning models such as image classification, object detection, and speech recognition, and for graphics-intensive applications such as remote graphics workstations, game streaming, and graphics rendering. G4 instances are available with a choice of NVIDIA GPUs (G4dn) or AMD GPUs (G4ad).
G4dn instances feature NVIDIA T4 GPUs and custom Intel Cascade Lake CPUs, and are optimized for machine learning inference and small scale training. These instances also bring high performance to graphics-intensive applications including remote workstations, game streaming, and graphics rendering. These instances are also ideal for customers who prefer to use NVIDIA software such as RTX Virtual Workstation and libraries such as CUDA, CuDNN, and NVENC.
G4ad instances feature the latest AMD Radeon Pro V520 GPUs and 2nd generation AMD EPYC processors. These instances provide the best price performance in the cloud for graphics applications including remote workstations, game streaming, and graphics rendering. Compared to the G4dn instances they offer up to 45% better price performance for graphics-intensive applications.
Amazon EC2 G4dn Instances
G4dn instances, powered by NVIDIA T4 GPUs, are the lowest cost GPU-based instances in the cloud for machine learning inference and small scale training. They also provide high performance and are a cost-effective solution for graphics applications that are optimized for NVIDIA GPUs using NVIDIA libraries such as CUDA, CuDNN, and NVENC. They provide up to 8 NVIDIA T4 GPUs, 96 vCPUs, 100 Gbps networking, and 1.8 TB local NVMe-based SSD storage and are also available as bare metal instances.
Increase performance and reduce machine learning inference costs
G4dn instances are equipped with NVIDIA T4 GPUs which deliver up to 40X better low-latency throughput than CPUs, so more requests can be served in real time. Also, G4dn instances are optimized to be cost-effective for machine learning inference, which can represent up to 90% of overall operational costs for machine learning initiatives.
Cost-effective small scale training
G4dn instances are also useful for small-scale/entry-level machine learning training jobs for those businesses or institutions that are less sensitive to time-to-train. G4dn instances deliver up to 65 TFLOPs of FP16 performance and are a compelling solution for small-scale training jobs.
High performance for graphics intensive applications
G4dn instances have up to 1.8X better graphics performance and up to 2X video transcoding capability over the previous generation G3 instances. Customers can configure virtual workstations with access to NVIDIA RTX Workstations at no additional cost.
Powered by NVIDIA T4 GPUs
NVIDIA T4 GPUs accelerate diverse cloud workloads, including deep learning training and inference and graphics. Based on the new NVIDIA Turing architecture, T4 GPUs feature multi-precision Turing Tensor Cores and new RT Cores. Turing Tensor Core technology with multi-precision computing for ML powers breakthrough performance from FP32 to FP16 to INT8, as well as INT4 precisions. It delivers up to 9.3X higher performance than CPUs on training and up to 36X on inference.
High performance networking and storage
G4dn instances offer up to 100 Gbps of networking for applications requiring high throughput. G4dn instances also support Elastic Fabric adapter (EFA) that enables customers to run applications requiring high levels of inter-node communications at scale. These instances offer up to 1.8 GB of NVMe-based SSD storage for applications that require fast access to locally stored data.
RTX and gaming drivers
G4dn instances offer NVIDIA RTX and Gaming drivers to customers at no additional cost. RTX drivers can be used to provide high quality virtual workstations for a wide range of visually intensive workflows. The Gaming driver provides unparalleled graphics and compute support for game development.
Amazon EC2 G4ad instances
G4ad instances, powered by AMD Radeon Pro V520 GPUs, provide the best price performance for graphics intensive applications in the cloud. These instances offer up to 45% better price performance compared to G4dn instances, which were already the lowest cost instances in the cloud, for graphics applications such as remote graphics workstations, game streaming, and rendering that leverage industry-standard APIs such as OpenGL, DirectX, and Vulkan. They provide up to 4 AMD Radeon Pro V520 GPUs, 64 vCPUs, 25 Gbps networking, and 2.4 TB local NVMe-based SSD storage.
Highest performance and lowest cost instances for graphics intensive applications
G4ad instances are the lowest cost instances in the cloud for graphics intensive applications. They provide up to 45% better price performance, including up to 40% better graphics performance, compared to comparable instances for graphics applications such as remote graphics workstations, game streaming, and rendering that leverage industry standard APIs such as OpenGL, DirectX, and Vulkan.
Simplified management of virtual workstations at the lowest cost in the cloud
G4ad instances allow customers to configure virtual workstations with high-performance simulation, rendering, and design capabilities in minutes, allowing customers to scale quickly. Customers can use AMD Radeon Pro Software for Enterprise and high-performance remote display protocol, NICE DCV, with G4ad instances at no additional cost to manage their virtual workstation environments with support for up to two 4k monitors per GPU.
Dependability in third party applications
The AMD professional graphics solution includes an extensive Independent Software Vendor (ISV) application testing and certification process called the Day Zero Certification Program. This helps ensure that developers can leverage the latest AMD Radeon Pro Software for Enterprise features combined with the reliability of certified software on the day of the driver release.
Powered by AMD Radeon Pro V520 GPUs
AMD Radeon Pro V520 GPUs provide high performance acceleration for graphics such as virtual workstations, computer generated imagery (CGI), game streaming, and digital content creation (DCC). These GPUs are built on AMD’s RDNA architecture that is hyper efficient, with low latency and high CPU to GPU bandwidth needed to enable high quality workstation and gaming experiences. With an improved graphics pipeline, RDNA architecture is designed to render your games faster with higher performance per clock.
Local instance storage
G4ad instances offer up to 2.4 TB of local NVMe storage for fast data access, enabling customers to efficiently create photo-realistic and high-resolution 3D content for movies, games, and AR/VR experiences.
Professional grade graphics drivers
G4ad instances provide professional grade graphics drivers at no additional cost. These drivers can be used to provide the best virtual workstation experience for a wide range of visually intensive workflows and unparalleled graphics and compute support for game development.
Ubitus is an innovative company leading the way in building cloud gaming technology. Through their platforms, users can enjoy a AAA gaming experience on any device including smartphones, tablets, gaming consoles, smart TVs, and computers as long as they’re connected to a broadband network.
“We used AWS to partner with 505 Games to bring Control Ultimate Edition to a highly popular portable gaming device. With the new EC2 G4ad instances we have a lower cost structure, which will enable us to bring more games like Control Ultimate Edition to the mobile users globally.”
Wesley Kuo, CEO – Ubitus.
Duolingo is a free language education platform that has become the most popular way to learn languages online. Duolingo’s language learning scientists, machine learning engineers, and AI experts use data from over 300 million learners to constantly increase effectiveness of the platform.
"As our ML and research teams grew, we decided to update our existing Amazon ECS-based compute infrastructure to support Amazon EC2 P3 and G4 GPU-based instance types to better scale our development model. Amazon's ECS-optimized AMIs for GPU instances helped us get the new cluster up and running very quickly and we found that the G4 instances doubled our ML training speeds when compared to P2 instances, leading to a cost savings of 33%, while the P3 instances quadrupled the performance and provided a cost savings of 15%. Overall, the G4 instances are suitable for our general use cases since they provide a good balance of cost and performance, and the P3 instances are ideal when the additional speed is critical for a particular workload."
Max Blaze, Staff Operations Engineer - Duolingo
Nearmap was founded in Perth, Australia, in 2007 and has grown from a small startup to a leader in digital imaging. The company specializes in creating 2D and 3D images from aerial photos of landscapes, a valuable resource for a wide variety of industries. For Nearmap, the impact of upgrading to Amazon EC2 G4 instances was immediate and profound: the company was able to run three times as much data for the same cost as on Amazon EC2 G2 instances, in a process that felt effortless and whose integration felt seamless with the AWS tools it was already using.
“Our customers rely on us to deliver highly accurate 3D Reality Models computed from multi-angle aerial photography across massive coverage areas. We use around 870 thousand GPU cores per day. We used to run this pipeline on Amazon EC2 G2 instances but switched to Amazon EC2 G4 instances and reduced our costs by 67%.”
John Corbett, Director - Vision Systems
In 2018, Snap Inc. (Snap), known for its Snapchat messaging app, had an intriguing new idea: create a series of animated videos starring each user’s Bitmoji, the personalized cartoon avatar that is Snapchat’s signature feature. Each week, Bitmoji TV would debut new episodes consisting of silly, professionally scripted and animated 3- to 8-minute videos in which Bitmojis of users and their friends were the stars, doing everything from fighting off zombies to competing in a low-gravity “Moonlympics.” With Amazon EC2 G4 Instances, Snap was able to create a fun, bold kind of personalized entertainment that rendered quickly enough for millions to enjoy simultaneously while keeping compute costs down.
"With Amazon EC2 G4 Instances versus Amazon EC2 G3 Instances, we were getting a 50 percent boost for a 10 percent higher cost.”
Brad Kotsopolous, Software Engineer - Snap Inc.
Untold Studios deployed various AWS virtual workstation configurations, including G4 instances, which feature NVIDIA T4 GPUs and RTX technology, and are optimized for VFX and animation workflows. With expanded instance options to select from, Untold Studios can more accurately match virtual machine types to specific workloads, an approach that is beneficial both creatively and economically.
“G4 has had a huge impact on how we work. We can upgrade a whole fleet of workstations with one line of code, so artists can work faster, improving creativity within everything we do.”
Sam Reid, Head of Technology - Untold Studios
The workstations were first deployed with Amazon Elastic Compute Cloud (Amazon EC2) G3 Instances. But when Amazon EC2 G4 Instances—powered by NVIDIA T4 Tensor Core GPUs (NVIDIA’s latest generation of GPUs) and NVIDIA RTX technology, the latest technology for high-performance simulation, rendering, and design—became available, Hive VFX was able to upgrade easily and quickly.
“Upgrading from the G3 to G4 instance was simple and in minutes, artists had access to a new generation of NVIDIA GPUs that deliver the latest in computer graphics and performance. The workstation performance we get with G4 Instances is well above even some of the bigger studios’ and more than sufficient for our needs.”
Bernie Kimbacher, Founder - Hive VFX
The Doritos’ Super Bowl 2020 commercial showcased a dance off between 75-year-old actor Sam Elliott and 21-year-old rapper Lil Nas X, a competition made possible using artificial intelligence technology developed by a startup called Humen, the parent company behind Sway. This tiny company was able to earn a partnership with megacompany Doritos for a Super Bowl campaign and launch Sway, an app that topped the charts in the App Store, with the help of Amazon Web Services (AWS). To take on the colossal workload of launching a graphics-intensive application during the Super Bowl, Humen used Amazon Elastic Compute Cloud (Amazon EC2) G4 Instances—and achieved super results.
"With Amazon EC2 G4 Instances, I was able, in about 6 months, to make Sway capable of handling Super Bowl scale."
Jesse Myers, Engineer - Humen
“Visual Effects software relies more and more on GPU power, and when you need it, you really need it, but that’s not every day. As a small business, we can’t afford to buy a racecar to drive to the grocery store; we can’t afford to buy something and not use it every day. We also can’t buy 10 of them for one project with unique demands. But with G4 instances, we have access to a fleet of GPU-enabled racecars that are better and faster than anything we could buy, but without the overhead of paying for them when we don’t need them. Artists are excited to have access to racecars, and we’re happy to return them when they are not required anymore!”
Chris Healer, President - The Molecule
"At PureWeb, we strive to provide the world's best platform for real-time streaming of interactive content for enterprises. So when the Geneva International Motor Show was cancelled due to COVID-19, Volkswagen reached out to us with the innovative solution of providing a virtual showroom experience for their customers to adapt to this cancellation. By working closely with AWS, as well as our creative agency partners, we were able to get a virtual replica of the showroom, created in the Unity Game Engine, running on our platform using Amazon EC2 G4dn instances in just three weeks. As a result, thousands of users were able to enjoy the Volkswagen booth of the Geneva Motor Show from home. Given both the time and technical constraints on the project, we are confident that AWS is the only cloud provider that could have helped us get the job done, and serve up a user experience that is stable, scalable, cost-effective, and truly global."
Chris Jarabek, Senior Software Architect - PureWeb
Instance Size vCPUs Memory (GB) GPU Storage (GB) Network Bandwidth (Gbps) EBS Bandwidth (Gbps) On-Demand Price/hr* 1-yr Reserved Instance Effective Hourly* (Linux) 3-yr Reserved Instance Effective Hourly* (Linux) Single GPU VMs g4dn.xlarge 4 16 1 125 Up to 25 Up to 3.5 $0.526 $0.316 $0.210 g4dn.2xlarge 8 32 1 225 Up to 25 Up to 3.5 $0.752 $0.452 $0.300 g4dn.4xlarge 16 64 1 225 Up to 25 4.75 $1.204 $0.722 $0.482 g4dn.8xlarge 32 128 1 1x900 50 9.5 $2.176 $1.306 $0.870 g4dn.16xlarge 64 256 1 1x900 50 9.5 $4.352 $2.612 $1.740 Multi GPU VMs g4dn.12xlarge 48 192 4 1x900 50 9.5 $3.912 $2.348 $1.564 g4dn.metal 96 384 8 2x900 100 19 $7.824 $4.694 $3.130
* Prices shown are for US East (Northern Virginia) AWS Region. Prices for 1-year and 3-year reserved instances are for "Partial Upfront" payment options or "No Upfront" for instances without the Partial Upfront option.
Amazon EC2 G4dn instances are available in the US East (N. Virginia and Ohio), US West (Oregon and N. California), Canada (Central), Europe (Frankfurt, Ireland, London, Milan, Paris, and Stockholm), Asia Pacific (Hong Kong, Mumbai, Seoul, Singapore, Sydney, and Tokyo), Middle East (Bahrain), South America (São Paulo), Africa (Cape Town), China (Beijing and Ningxia), and GovCloud (US-East and US-West) AWS Regions.
Instance Size vCPU Memory (GB) GPU Instance Storage (GB) Network Bandwidth (Gbps) EBS Bandwidth (GBps)
1-yr Reserved Instance Effective Hourly* (Linux)
3-yr Reserved Instance Effective Hourly* (Linux)
g4ad.4xlarge 16 64 1 600 Up to 10 Up to 3
g4ad.8xlarge 32 128 2 1200 15 3
g4ad.16xlarge 64 256 4 2400 25 6
*Prices shown are for US East (Northern Virginia) AWS Region. Prices for 1-year and 3-year reserved instances are for “Partial Upfront” payment options or “No Upfront” for instances without the Partial Upfront option.
Get Started with G4dn Instances
Using pre-built AMIs and containers from AWS
Using Amazon Deep Learning AMIs or Deep Learning Containers, you can quickly launch Amazon EC2 instances pre-installed with popular deep learning frameworks and interfaces such as TensorFlow, PyTorch, and MXNet to train sophisticated, custom AI models, experiment with new algorithms, or to learn new skills and techniques. To learn more, visit the Amazon Deep Learning AMIs product page or the AWS Deep Learning Containers product page.
NVIDIA RTX Virtual Workstation AMIs support running up to four 4K displays per GPU for visualization apps. NVIDIA Gaming AMIs render videos at 60 FPS and support running a single 4K display on a per GPU basis.
Building your own custom AMI with NVIDIA drivers
The NVIDIA Data Center drivers provide the best performance for highly intensive computational workloads often found in Deep Learning, Data Science, and HPC.
NVIDIA RTX drivers can be used to enable NVIDIA RTX Virtual Workstation software. RTX vWS offers support for up to four 4K displays on a per GPU basis. NVIDIA Gaming drivers support the world's most advanced graphics cards, gaming solutions, and gaming technology. This driver offers support for a single 4K display on a per GPU basis.
Get Started with G4ad Instances
Using pre-built AMIs from AWS
AMD Radeon Pro Software for Enterprise drivers provide high performance graphics for virtual workstations, game streaming, rendering, and other graphics applications. To get started quickly, you can use AWS provided AMIs that include AMD Radeon Pro Software for Enterprise pre-installed. These AMIs are available on the AMD Radeon Pro Marketplace.
Building your own custom AMI with AMD drivers
AMD Radeon Pro Software for Enterprise drivers can be used to configure instances with powerful and reliable professional-grade graphics. Workstations provisioned with these drivers can support up to two 4K displays.