Skip to main content

On This Page

AWS Graviton5: 60% Performance Boost with 20-30% Cost Savings for EC2 Workloads

2 min read
Share

These articles are AI-generated summaries. Please check the original sources for full details.

AWS Graviton: The Best Price Performance for Your AWS Workloads

At AWS re:Invent 2025, AWS unveiled Graviton5, its most powerful custom chip, delivering up to 60% performance improvements. Sudhir Raman and Ali Saidi highlighted that 90,000+ customers have achieved 20-30% cost savings and significant carbon footprint reductions.

Why This Matters

The technical reality of cloud workloads often diverges from idealized models. While Graviton5’s 192-core architecture and 192MB L3 cache promise scalability, real-world migrations face challenges like concurrency bottlenecks and cache inefficiencies. Atlassian’s failed Graviton3 migration (2023) cost six months of effort, underscoring the risks of inadequate testing. The cost of not adopting Graviton extends beyond performance: customers like Snowflake report 57% lower carbon emissions per workload, while suboptimal choices risk missed sustainability goals and higher operational costs.

Key Insights

  • “Atlassian’s failed Graviton3 migration, 2023”: Initial tests showed regression due to unoptimized Java workloads, halting the project.
  • “Graviton5’s 192-core architecture optimized for scale-up workloads like databases”: DDR5-8800 support and 5.3x larger L3 cache address memory-bound applications.
  • “Atlassian’s 32% performance gain on Jira with Graviton4”: Migration to Graviton4 with THP optimization reduced latency by 12% and costs by 25%.

Practical Applications

  • Use Case: Atlassian’s Jira/Confluence achieved 32% faster performance and 25% cost reduction on Graviton4.
  • Pitfall: Overlooking cache hierarchies (e.g., L3 misses) can negate Graviton’s benefits, as seen in Atlassian’s initial Graviton3 failure.

References:


Continue reading

Next article

Kiro Powers Launch at AWS re:Invent 2025: Specialized AI for Development

Related Content