AWS r6a.24xlargevsAWS r6i.24xlarge

r6a.24xlarge pairs 96 vCPUs with 768GB of RAM at $5.4432/hr On-Demand (about $3919/mo at 24×7). r6i.24xlarge pairs 96 vCPUs with 768GB at $6.0480/hr (~$4355/mo). r6a.24xlarge is 11% cheaper per hour than r6i.24xlarge ($0.6048/hr gap).

Both are generation-6 memory-optimized instances, but they run on different silicon: **r6a.24xlarge** is AMD EPYC (x86_64), **r6i.24xlarge** is Intel Xeon (x86_64). AMD variants (suffix `a`) are typically 10% cheaper than Intel siblings at comparable single-thread performance. Graviton variants (suffix `g`) are usually 20–40% cheaper but require ARM64-compatible binaries — most modern Linux stacks are fine, but verify any compiled extensions, native modules, or third-party binaries before migrating. Same vCPU/RAM ratio, same network performance class, different processor.

On raw price-per-performance, the two are r6a.24xlarge delivers ~52% more single-thread Sysbench score per dollar (749 vs 492 points per $1/hr). That's the cleanest signal we have for "which one runs your workload faster per dollar," but it only matters if your workload is single-thread-bound; for parallel workloads the multi-core scores (217304 vs 148619) are what to weigh. Spot pricing flips many of these comparisons — when r6a.24xlarge drops to $1.2186/hr and r6i.24xlarge drops to $1.9371/hr, the cheap-per-hour winner can swing meaningfully.

In practice, pick r6a.24xlarge when your workload is closer to memory-optimized (memory-bound work — in-memory databases, real-time analytics, large caches). Pick r6i.24xlarge when it's closer to memory-optimized (memory-bound work — in-memory databases, real-time analytics, large caches). When neither side is obviously right, the cheaper hourly rate usually wins for fault-tolerant batch workloads, while the higher single-core score usually wins for latency-sensitive web traffic. The regional pricing tables linked from each instance page below show where each is currently cheapest — sometimes a >20% regional gap flips the comparison entirely.