by

AWS cranks out Graviton5

Amazon Web Services has made its Graviton5 CPU generally available, giving cloud customers another Arm-based chip to chew through workloads. The new silicon is rolling out in Amazon EC2 M9g and M9gd instances.

AWS says Graviton5 delivers up to 25 per cent better performance than Graviton4. Web applications are claimed to run 35 per cent faster, AI and machine learning inference 35 per cent faster and databases 30 per cent faster.

The chip has 192 cores, is built on a 3nm process and cuts inter-core latency by up to 33 per cent. That gives AWS a tidy bit of marketing ammo, although customers will care more about the bill than the badge.

Graviton has come a long way since 2018, when Amazon launched its first custom cloud processor. Graviton2 followed in 2019 with up to 40 per cent better price performance than comparable processors.

Graviton3 arrived in 2021 and used 60 per cent less energy than comparable Amazon EC2 instances for the same performance. Graviton4 turned up in 2023 with three times the processing cores and better handling for large databases and analytics.

Graviton5, which arrived in 2025, doubles the core count to 192 and targets everything from real-time gaming to AI. The cloud giant is clearly not waiting for the usual server chip crowd to define its roadmap.

The I/O side gets a decent shove. Graviton5 supports DDR5-8800 memory, which AWS claims is the fastest DDR5 in the cloud, along with PCIe Gen6.

Each chip has an L3 cache five times larger than the previous generation. Each core gets access to 2.6 times more L3 cache, which should mean fewer stalls while workloads wait around for data like bored commuters.

AWS says the larger cache keeps frequently accessed data closer to the processor. In less vendor-polished English, that means quicker response times when the software is not tripping over itself.

Memory performance has improved, giving customers more room for larger datasets and memory-hungry applications. Network and storage bandwidth get a bump too, with up to 15 per cent higher network bandwidth and 20 per cent higher Amazon Elastic Block Store bandwidth on average across instance sizes.

The largest instances can reach up to twice the network bandwidth of the previous generation. That should help with distributed applications, faster backups and data transfers that do not crawl along like a hungover intern.

Meta has committed to deploying tens of millions of Graviton cores for agentic AI use. Uber and Snowflake are among the customers using the chips, while AWS says more than 120,000 customers are building on Graviton.

The M9g instances offer 25 per cent better compute performance than the previous generation. M9gd instances are aimed at outfits that need high-speed local SSD storage, with up to 11.4TB of capacity and 30 per cent higher IOPS than the previous generation.

AWS wants Graviton5 seen as a major jump for cloud computing, not just another slab of server silicon with a shinier press release. With 192 cores, PCIe Gen6, DDR5-8800 and a swollen cache, it at least has the spec sheet to make rivals twitch.

 

 

TOPICS:
AI inference  ·  Amazon EC2  ·  arm  ·  aws  ·  cloud computing  ·  DDR5-8800  ·  Graviton5  ·  M9g  ·  M9gd

Latest articles

Share

Featured articles

Hot topics

No results found.

Latest reviews