- Sources: primary, discussion
- Summary: AWS made the memory-optimized R9g and R9gd instances generally available on 2026-08-31, stating up to 25 percent higher compute performance per vCPU over Graviton4-based R8g, DDR5 at 8800 MT/s up from 5600 MT/s, a 5x larger L3 cache, up to 2x higher network and EBS bandwidth at the largest sizes reaching 100 Gbps network and 72 Gbps EBS on the 48xlarge, and up to 3x higher packet-processing performance. Both families offer 11 sizes from medium to metal-48xl, topping out at 192 vCPUs and 1536 GiB, support Instance Bandwidth Configuration to shift 25 percent of allocation between EBS and VPC networking, and carry the Nitro Isolation Engine introduced with C9g and M9g, which AWS describes as using formal verification of isolation. R9gd adds local NVMe SSD, AWS states most applications migrating from R8g need no code changes, and availability is US East (N. Virginia, Ohio), US West (Oregon), and Europe (Frankfurt).
- Why it matters: The memory bandwidth and cache figures are what move workloads sized by memory rather than by vCPU count, and AWS states the move from R8g needs no code changes.
send feedback on this story