Amazon Web Services has made a new impact on cloud computing with the announcement of Graviton5, its fifth generation of custom-designed ARM processors for Amazon EC2. The company describes it as its “most powerful and efficient CPU to date,” with a clear message: more performance, better prices, and lower energy consumption for increasingly complex workloads.
Up to 192 cores and 5 times more cache
Graviton5 is designed to maximize cloud workload scaling:
- Up to 192 cores per chip, the highest CPU density available today on EC2.
- 5x larger L3 cache than Graviton4, with 2.6 times more accessible L3 cache per core.
- Manufactured using a 3nm process, which helps improve both performance and energy efficiency.
According to AWS data, the new M9g instances based on Graviton5 deliver up to a 25% higher computing performance than the previous generation, with additional improvements in memory, networking, and storage:
- Up to 15% more network bandwidth.
- Up to 20% more bandwidth for Amazon Elastic Block Store (EBS) on average.
li>Double the network bandwidth on the largest instance sizes.
In practice, this translates to applications that respond faster and can process more data per instance, with fewer servers needed to do the same work.
Designed for real-time gaming, large databases, and analytics
The design of Graviton5 aims to reduce the “distance” between cores and the memory they need. AWS claims up to 33% less latency in core-to-core communication, which is especially beneficial for:
- Online and real-time games, where every millisecond of latency matters.
- High-performance databases (OLTP and OLAP).
- Big data analytics and log processing.
- Electronic Design Automation (EDA) and CPU-intensive engineering workloads.
The expanded L3 cache helps keep “hot” data closer to the processor, reducing memory trips and speeding up complex queries and calculations. For those running Java applications, microservices, database engines, or analytics platforms, this cache boost can make a significant difference in tail latencies (p95/p99).
Lower power consumption, same or greater performance
AWS emphasizes that the improvement is not just raw performance but also efficiency. Graviton5 is designed to deliver more performance per vCPU and per watt than its predecessors, a key factor in a landscape where organizations must balance cost, performance, and sustainability goals.
The company notes that, for the third consecutive year, over half of the new CPU capacity added to AWS is based on Graviton chips, and that 98% of the top 1,000 EC2 customers are already using these instances, including names like Adobe, Airbnb, Atlassian, Epic Games, Formula 1, Pinterest, SAP, Siemens, Snowflake, and Synopsys.
Some use cases highlighted by Amazon include:
- Airbnb: up to 25% performance improvement over other architectures of the same generation and 20% over Graviton4 on search workloads, with better p95 latencies for users.
- Atlassian (Jira/Confluence): internal tests show 30% more performance and 20% lower latency compared to Graviton4.
- SAP HANA Cloud: 35% to 60% performance increase on OLTP queries using M9g versus the previous generation.
- Siemens and Synopsys: improvements of up to 30-35% in verification times and EDA tools, critical in semiconductor design.
Security: Nitro and the new isolation engine
Like its predecessors, Graviton5 relies on the AWS Nitro System, which offloads network, storage, and virtualization functions to dedicated cards (Nitro Cards) so that nearly all CPU and memory resources are available for customers’ workloads. This architecture also supports a “zero operator access” design: no one at AWS can log into an EC2 server or read its memory.
The innovation in this generation is the Nitro Isolation Engine, a minimalistic software component subjected to formal verification, meaning mathematical proofs that demonstrate it behaves exactly as specified. Amazon presents it as a new cloud security standard, providing additional guarantees of workload isolation between tenants and from AWS infrastructure itself.
Availability: M9g in preview, C9g and R9g in 2026
The first Graviton5-based instances are the Amazon EC2 M9g, aimed at Mainstream workloads, now available in previews for interested customers.
AWS also announced:
- C9g: compute-optimized instances aimed at simulations, rendering, and scientific processing.
- R9g: memory-optimized instances for in-memory databases, distributed caches, and big data analytics.
Both C9g and R9g will arrive throughout 2026, expanding the Graviton5 family to more workload profiles.
What does Graviton5 mean for enterprises?
For many organizations, the message is twofold:
- More performance for the same or less cost: if AWS’s promises hold true in production, companies can consolidate instances, reduce processing times, and simultaneously lower their EC2 bills.
- A clear path toward ARM in the data center: with each generation, Graviton reinforces that 64-bit ARM architecture is not only an efficient mobile option but a serious alternative to x86 in the cloud, even for critical enterprise workloads.
For technical teams, the practical question will be assessing compatibility (ARM binaries, containers, libraries) and measuring real-world performance on their specific workloads, something many large customers have been doing since Graviton2 and Graviton3.
Frequently Asked Questions about AWS Graviton5
1. What is AWS Graviton5, and how does it differ from Graviton4?
Graviton5 is the fifth generation of AWS’s custom ARM processors for Amazon EC2. It offers up to 192 cores per chip, five times larger L3 cache, and up to 25% higher performance compared to Graviton4, along with improvements in network and storage bandwidth and energy efficiency.
2. What types of EC2 instances use Graviton5?
Currently, Graviton5 is available in preview for M9g general-purpose instances. AWS has announced that C9g (compute-optimized) and R9g (memory-optimized) instances will be launched in 2026.
3. Which workloads should consider migrating to Graviton5?
Graviton5 is especially suitable for backend applications, microservices, databases, data analytics, EDA, online gaming, and any workload that benefits from high core density, large cache, and low network and memory latencies. Migration is generally easier if applications already run in containers or on mature ARM-supported runtimes (e.g., Java, Node.js, Python, Go).
4. What cost and energy efficiency benefits does Graviton5 offer?
AWS promotes Graviton5 as its best price-performance and energy-efficient CPU. In theory, this can reduce the number of instances needed and associated power consumption, helping organizations meet sustainability goals without sacrificing performance.
via: Amazon