Graviton5: AWS Next-Gen ARM Processor Preview

Graviton5: AWS’s Most Powerful ARM Processor Yet

AWS unveiled Graviton5 at re:Invent 2025, marking the company’s most ambitious leap forward in custom silicon development. For organizations already achieving up to 40% cost savings with Graviton-based instances, this next generation promises even more impressive performance gains while maintaining the price-performance advantage that made ARM processors a viable alternative to x86 in the data center.

The announcement represents AWS’s continued commitment to vertical integration in cloud infrastructure, following a trajectory that has seen each Graviton generation deliver substantial improvements over its predecessor. With major enterprises including Netflix, Pinterest, and Twitter already running production workloads on Graviton, the fifth generation arrives with high expectations.

The Evolution from Graviton1 to Graviton5

Understanding where Graviton5 fits requires looking at how AWS’s ARM processor family has evolved since its inception. The original Graviton processor, launched in 2018, was AWS’s first attempt at custom ARM silicon. While modest in performance compared to contemporary x86 offerings, it proved the concept was viable and set the stage for what would become a transformative product line.

Graviton2 changed everything. Released in 2020, it delivered 7x the performance of the original and established ARM as a legitimate choice for production workloads. AWS designed Graviton2 using 64-bit ARM Neoverse N1 cores, each with their own dedicated L1 and L2 caches, eliminating the noisy neighbor problems common in shared cache designs. This generation convinced many skeptical organizations to give ARM a serious look.

Graviton3 arrived in 2022 with 25% better compute performance, DDR5 memory support, and twice the floating-point performance of its predecessor. The use of advanced 5nm process technology also delivered up to 60% better energy efficiency, addressing growing sustainability concerns in cloud computing. This generation introduced support for bfloat16, making it suitable for machine learning workloads.

Graviton4 in 2024 pushed boundaries further with 96 cores per processor, support for up to 536GB of RAM, and 30% better performance than Graviton3. It introduced the R8g instance family with the highest memory-to-vCPU ratio in the Graviton lineup, making it ideal for memory-intensive database and analytics workloads.

Graviton Generation Comparison

Performance Benchmarks and Real-World Improvements

AWS has published preliminary benchmark data showing Graviton5 delivers approximately 35% better performance for general compute workloads compared to Graviton4. For specific workload types, the improvements are even more pronounced:

• Web serving: 40% improvement in requests per second
• Database operations: 35% higher throughput for PostgreSQL and MySQL
• Container workloads: 30% better pod density per instance
• ML inference: 50% improvement for transformer-based models
• Video encoding: 45% faster transcoding with FFmpeg

Graviton5 includes enhanced vector processing capabilities with expanded SVE2 (Scalable Vector Extension 2) support, making it particularly well-suited for AI/ML workloads that can benefit from ARM’s advanced SIMD instructions. This positions Graviton5 as not just a general-purpose processor but also a viable option for inference workloads traditionally handled by GPUs.

Cost Savings: Graviton vs Intel and AMD

The business case for Graviton continues to strengthen with each generation. Based on AWS pricing and customer reports, here’s what organizations can expect:

These savings compound significantly at scale. An organization running 1,000 instances could potentially save $500,000 or more annually by migrating from Intel-based instances to Graviton5, while achieving better performance.

Instance Types Using Graviton5

AWS launches new Graviton processors across multiple instance families to address different workload requirements. The G5 generation introduces the following instance families:

Migration Considerations and Compatibility

Moving to Graviton requires planning, but AWS has invested heavily in making the transition smooth. Most Linux distributions support ARM64 natively, and container workloads typically require only rebuilding images for multi-architecture support.

Key migration steps:

1. Audit your application dependencies for ARM64 compatibility
2. Build multi-architecture container images
3. Test thoroughly in a staging environment
4. Use AWS Compute Optimizer for instance recommendations
5. Implement gradual traffic shifting during migration

# Build multi-architecture container images
docker buildx create --use
docker buildx build --platform linux/amd64,linux/arm64 \
    -t myapp:latest --push .

# Update Terraform for Graviton instances
resource "aws_instance" "web" {
  # Change from m7i to m9g (Graviton5)
  instance_type = "m9g.xlarge"

  # Use ARM64 AMI
  ami = data.aws_ami.amazon_linux_arm.id
}

# Use AWS Compute Optimizer for recommendations
aws compute-optimizer get-ec2-instance-recommendations \
    --instance-arns arn:aws:ec2:us-east-1:123456789:instance/i-xxx

When to Use Graviton (and When Not To)

Ideal workloads for Graviton5:

• Containerized microservices and Kubernetes workloads
• Web applications, APIs, and serverless functions
• Open-source databases (PostgreSQL, MySQL, MongoDB, Redis)
• Big data processing (Spark, Hadoop, Presto)
• Machine learning inference and model serving
• Media encoding and processing pipelines
• CI/CD build agents and automation

Workloads that may require evaluation:

• Legacy applications with hard-coded x86 dependencies
• Windows workloads (Graviton is Linux-only)
• Software with binary-only x86 distributions
• Workloads requiring specific x86 instruction sets (AVX-512)
• Vendor software without ARM64 support

Getting Started with Graviton5

AWS provides several resources to help teams evaluate and migrate to Graviton:

• AWS Graviton Ready Program: Validates ISV software compatibility with certification badges
• Graviton Technical Guide: Detailed migration documentation and best practices
• AWS Compute Optimizer: Automated instance recommendations based on utilization patterns
• Graviton Challenge: AWS credits for testing migrations in your environment
• Graviton Fast Start: Consulting engagement for enterprise migrations

Start by identifying containerized or Linux-native workloads in your environment. These typically migrate with minimal effort and deliver immediate cost benefits. Use AWS Compute Optimizer to identify candidates and track the price-performance gains after migration. With Graviton5’s improved AI capabilities, even machine learning workloads that previously required GPUs may now be candidates for ARM-based inference.