AWS Expands R8i to Milan
AWS made EC2 R8i and R8i-flex instances available in the Europe (Milan) region on August 7. The arrival brings eighth-generation, memory-optimized compute to teams operating in the region.
Both instance types run on custom Intel Xeon 6 processors available only on AWS. AWS says they deliver up to 15% better price-performance and 2.5x more memory bandwidth compared to previous-generation Intel-based instances.
Against the R7i generation directly, the improvement is 20% higher overall performance, with larger gains on specific workloads: up to 30% faster for PostgreSQL databases, up to 60% faster for NGINX web applications, and up to 40% faster for AI deep learning recommendation models.
The processor runs at a sustained all-core turbo frequency of 3.9 GHz, up from 3.2 GHz in the previous generation, and the instances carry a 4.6x larger L3 cache than their predecessors. L3 cache is fast on-chip memory that reduces how often the processor has to reach out to slower main memory. That matters most for database workloads and CPU-based inference, where the larger cache reduces latency directly.
Both types also use DDR5 7200MT/s memory modules, the faster memory standard behind the 2.5x bandwidth figure. The instances support always-on memory encryption, a detail relevant to teams with data-residency or compliance obligations in Italy.
Two Instance Families, Two Use Cases
The two families serve different operating profiles and are worth treating separately when planning capacity.
R8i is built for workloads that need sustained, maximum CPU performance or the largest available instance sizes. It comes in 13 sizes, including two bare metal options and a new 96xlarge size that reaches 384 vCPUs and 3 TB of memory. Bare metal means the operating system runs directly on physical hardware with no hypervisor in between, which matters for workloads such as SAP HANA that have strict licensing or performance isolation requirements.
R8i is SAP-certified and delivers 142,100 aSAPS, a performance benchmark used to size SAP systems. AWS states that figure is the highest among all comparable machines in on-premises and cloud environments.
R8i-flex is optimized for workloads that use memory heavily but do not consistently pin the CPU at full capacity. These are common in enterprise environments: mid-size caches, in-memory analytics, and multi-tier applications that spike rather than sustain. R8i-flex covers sizes from large to 16xlarge and delivers up to 5% better price performance and 5% lower prices compared to R8i by efficiently matching compute allocation to actual demand.
AWS describes it as able to reach full CPU performance 95% of the time, which covers most real workload patterns without requiring a dedicated high-CPU instance.
Both families use sixth-generation AWS Nitro Cards, which handle I/O offload in dedicated hardware separate from the main CPU. That architecture delivers up to 2x more network and EBS bandwidth compared to prior-generation instances. Amazon Elastic Block Store (EBS) is AWS's primary block storage service. R8i also supports Elastic Fabric Adapter (EFA) on the 48xlarge, 96xlarge, and metal sizes, enabling low-latency networking for tightly coupled distributed workloads.
A new bandwidth configuration feature lets teams shift allocation between network and EBS in 25% increments. For database workloads that are either storage-bound or network-bound at different points, this is a meaningful tuning option that did not exist in prior generations.
What This Means for Milan-Based Workloads
Teams operating in Europe (Milan) for data-residency, latency, or regulatory reasons now have a compute tier that was previously unavailable in the region. The practical effect depends on what they are currently running.
For SAP HANA environments, the SAP certification and aSAPS figure directly address the sizing exercise. SAP workloads are constrained by certified configurations, and the R8i's 142,100 aSAPS score is the highest among all comparable machines in on-premises and cloud environments.
For PostgreSQL and other relational databases, the 30% speed gain over R7i is tied to higher memory bandwidth and the larger L3 cache. Database queries that repeatedly scan large datasets in memory will see the most benefit. Teams benchmarking against existing R7i or M-series instances should run representative query workloads rather than synthetic tests to confirm actual gains in their environment.
For AI inference running on CPU rather than GPU, the Intel Advanced Matrix Extensions (AMX) with FP16 support enables a broader set of inference workloads without requiring a GPU instance. AMX is a hardware instruction set that accelerates matrix operations common in neural network inference. That matters for teams running recommendation models or lightweight inference tasks where GPU capacity is expensive or oversized.
Instances are available via Savings Plans, On-Demand, and Spot pricing. Teams that already have Savings Plans in place should check whether the commitments cover R8i in the Milan region or require adjustment.
What to Do Now
Cloud platform owners and application teams running memory-intensive workloads in Europe (Milan) should take three concrete steps.
First, identify every workload currently running on R7i or older Intel-based instances in the Milan region. Flag databases, in-memory caches, SAP systems, and inference jobs as priority candidates for evaluation.
Second, match each workload to the right family. Workloads with sustained high CPU or SAP certification requirements belong on R8i. Workloads with variable CPU that peak but don't sustain should go on R8i-flex for the additional price-performance benefit.
Third, run a cost model before launching. On-Demand pricing in a new region is the right starting point, but the savings from Savings Plans or Reserved Instances can be significant for workloads that run continuously. Teams running Oracle on RDS can find reservation mechanics and discount structures for R8i in the Amazon RDS for Oracle Adds Reserved Instances for R8i and M8i article.