Amazon has made its AWS Graviton5 CPUs generally available, bringing a major upgrade to its custom cloud chip lineup. The new processor offers up to 25 percent better performance than Graviton4, doubles the core count to 192 cores, and adds support for faster DDR5 memory and PCIe Gen 6.
Graviton5 is built for demanding cloud workloads, including AI inference, databases, analytics, real time applications, and general compute. Amazon says it is the fastest and most efficient CPU it has built so far, continuing the company’s push toward custom silicon inside AWS.
The launch matters because cloud providers are trying to reduce the cost of AI infrastructure while improving performance. GPUs remain central to AI training and large scale acceleration, but CPUs are still critical for inference, data processing, service orchestration, and many production workloads. With Graviton5, AWS is trying to give customers more performance per dollar while keeping power use under control.
Graviton5 brings a major core count and cache upgrade
The biggest headline is the jump to 192 cores. That gives Graviton5 far more parallel processing capacity than earlier generations and makes it better suited for heavy cloud workloads.
Amazon also says the chip has a much larger cache. The L3 cache is five times larger than the previous generation, and each core has access to 2.6 times more L3 cache than Graviton4. More cache can reduce delays when applications need repeated access to data, improving response times in workloads that depend heavily on memory movement.
| Feature | AWS Graviton5 |
|---|---|
| CPU cores | 192 |
| Process technology | 3nm |
| Performance uplift | Up to 25 percent over Graviton4 |
| AI and ML inference | Up to 35 percent faster |
| Application performance | Up to 35 percent faster |
| Database performance | Up to 30 percent faster |
| Memory support | DDR5 8800 |
| I/O support | PCIe Gen 6 |
| Available instances | Amazon EC2 M9g and M9gd |
The memory upgrade is also important. Graviton5 supports DDR5 8800, which Amazon describes as the fastest DDR5 memory in the cloud. Faster memory should help with larger datasets and memory intensive services.
AWS is positioning Graviton5 for AI and enterprise workloads
Graviton started as Amazon’s Arm based cloud CPU experiment, but it has now become one of AWS’s most important custom silicon projects. Graviton5 shows how far that strategy has moved from basic efficiency to high performance cloud computing.
Amazon says AI and machine learning inference can run up to 35 percent faster on Graviton5 compared with Graviton4. That is important as more companies deploy AI services that need to serve requests quickly and cheaply.
The chip also targets databases, distributed applications, analytics, gaming, and enterprise workloads. For customers running large cloud deployments, even moderate performance and efficiency gains can translate into meaningful cost savings.
M9g and M9gd instances are the first to use Graviton5
Graviton5 is now available through Amazon EC2 M9g and M9gd instances. The M9g instances are aimed at general purpose cloud workloads, while M9gd adds local SSD storage for workloads that need high speed local capacity.
Amazon says M9gd instances can offer up to 11.4TB of local SSD storage and up to 30 percent higher IOPS than the previous generation. That makes them useful for databases, caching layers, analytics tasks, and applications where local storage speed matters.
Network and storage bandwidth have also improved. Amazon says Graviton5 based instances offer up to 15 percent higher network bandwidth and 20 percent higher Amazon EBS bandwidth on average across instance sizes. The largest instances can offer up to twice the network bandwidth.
Major customers are already using Graviton
Amazon says more than 120,000 customers are building on Graviton. Meta is expected to deploy tens of millions of Graviton cores for agentic AI workloads, while Uber and Snowflake are also using the platform.
That customer base matters because custom cloud chips only succeed if major workloads actually move to them. Graviton5 gives AWS another way to keep customers inside its ecosystem while offering better price performance than traditional third party server CPUs in many cases.
It also strengthens Amazon’s position against Microsoft, Google, Oracle, and other cloud providers that are investing heavily in custom AI and cloud infrastructure.
Graviton5 shows why cloud CPUs still matter in the AI era
The AI market is often focused on GPUs and accelerators, but CPUs remain essential. Every AI system still needs CPUs to manage data, schedule workloads, run services, handle networking, and support inference pipelines.
Graviton5 is Amazon’s answer to that broader need. It gives AWS a larger, faster, and more modern custom CPU that can support AI workloads without relying only on expensive accelerators.
For customers, the main appeal will be practical: better performance, lower cost, improved efficiency, faster memory, and stronger cloud infrastructure. For Amazon, Graviton5 helps reduce dependence on outside chip suppliers while giving AWS more control over pricing and performance.
With 192 cores, DDR5 8800 support, PCIe Gen 6, and stronger AI inference performance, Graviton5 is one of Amazon’s most important cloud CPU launches yet. It shows that the next phase of AI infrastructure will not be built only around GPUs. Custom CPUs are becoming just as important to the economics of cloud computing.
Discussion (0)
Be the first to comment.