- Newest
- Most votes
- Most comments
You can find info at Amazon EC2 R8g Instances page.
R8gd and R8g are listed together in R8g section. For example, r8gd.large is listed to have 1 x 118 NVMe SSD
R8gd is the R8g memory-optimized family built on AWS Graviton4 and the Nitro System, and it adds locally attached NVMe SSD instance storage for workloads that need low-latency, high-throughput local disks. Depending on size, R8gd instances provide instance-store NVMe capacity up to 11.4 TB of local NVMe SSD block storage, and are positioned for memory- and I/O-intensive applications such as relational/NoSQL databases, in-memory caches and real-time analytics.
Performance and architectural improvements:
Compared with Graviton3-based predecessors, Graviton4-based R8gd/R8g instances deliver noticeable generational gains: It offers up to 30% better general compute performance, up to 40% higher performance for I/O-intensive database workloads, and up to 20% faster query results for I/O-intensive real-time analytics; these improvements are driven by Graviton4’s microarchitectural changes plus a much stronger memory subsystem. R8g instances use DDR5-5600 memory, larger caches and expanded vCPU/memory configurations (up to about 3× the vCPU and memory of the prior generation in the largest sizes), and significantly higher memory bandwidth, all of which combine to reduce database latency and improve throughput for memory- and I/O-bound workloads.
NVMe SSD (instance-store) details:
The NVMe SSDs in R8gd are physically attached instance-store NVMe devices presented as block devices inside the guest OS; capacity and drive counts vary by instance size. For example, smaller sizes begin with single NVMe devices (r8gd.medium = 1 × 59 GB, r8gd.large = 1 × 118 GB, r8gd.xlarge = 1 × 237 GB), mid sizes have larger single drives (r8gd.8xlarge = 1 × 1900 GB) and the largest sizes combine multiple 1.9 TB drives (r8gd.48xlarge and r8gd.metal-48xl present 6 × 1900 GB devices - 11.4 TB total). Because these are instance-store NVMe devices, they are fast local block devices (excellent IOPS and low latency compared with remote volumes) but their capacity and the number of devices are fixed by instance size.
Network, EBS bandwidth, and configurability:
R8gd instances also offer high networking and EBS throughput—AWS documents up to 50 Gbps network bandwidth and up to 40 Gbps of Amazon EBS bandwidth on the largest sizes-plus the ability to adjust (weight) how the instance’s available bandwidth is allocated between network and EBS by up to 25% to tune for your workload. Elastic Fabric Adapter (EFA) is supported on the largest 24x/48x and metal sizes, which can be useful for tightly-coupled HPC or MPI-style tasks and some distributed database/analytics patterns. These networking and bandwidth options are important when you design replication, distributed analytics, or data-ingest pipelines that pair local NVMe scratch with remote persistent storage.
Operational caveats and recommended approach:
Two practical cautions to call out: (1) instance-store NVMe is ephemeral - data on those SSDs persists only for the life of the instance and is lost if the instance is stopped, hibernated, terminated, or the host fails — so do not use instance-store as the sole location for irreplaceable data (back up to EBS/S3/EFS as needed); and (2) AWS’s public instance-type documentation lists capacities and device counts but does not publish OEM part numbers, per-drive endurance ratings, or guaranteed per-drive IOPS/latency figures, so exact micro-specs can vary and will affect real application behavior. For those reasons you should benchmark the R8gd size(s) you intend to use (fio or your application-level tests) to validate sustained IOPS, throughput and latency for your workload and to determine whether instance-store NVMe or an alternative (EBS provisioned IOPS / bare-metal with local NVMe, or a replicated architecture) is the better fit.
Sources:
[+] https://aws.amazon.com/ec2/instance-types/r8g/
[+] https://aws.amazon.com/about-aws/whats-new/2025/04/amazon-ec2-c8gd-m8gd-r8gd-instances/
[+] https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/nvme-detailed-performance-stats.html
answered a year ago
Relevant content
asked 7 years ago
asked 3 years ago

I wouldn't ask if I could find. Obviously the link you have provided has nothing relevant to question. Even the more detailed AI reply is much better what is in that link (but it doesn't answer to the question either). I am explicitly asking about the specs of NVME drives used in those instances. For example, lots of NVME SSD disks which I weren't allowed to list here are all NVME SSD disks but they have very different specs like their random and sequential IOPS, read\write.