What Are the EC2 Intel and AMD All-Core Turbo Clock Speeds?

Amazon EC2 offers a wide range of virtual machine instances powered by custom-tuned Intel Xeon Scalable and AMD EPYC processors, each delivering specific sustained all-core turbo clock speeds tailored to diverse workload needs. Understanding these frequency tiers across generations helps engineers choose the right instance family for latency-sensitive, compute-bound, or high-frequency database applications. While general-purpose and memory-optimized instances usually prioritize core density and power efficiency with moderate clock speeds between 2.9 GHz and 3.5 GHz, specialized high-frequency instances push sustained all-core turbo clocks up to 4.5 GHz. This guide breaks down the exact sustained all-core turbo frequencies across 5th through 7th generation Intel and AMD EC2 instance families, as well as specialized high-frequency configurations.

How EC2 Defines Sustained All-Core Turbo

On physical x86 processors, turbo boost frequencies dynamically fluctuate based on thermal headroom, power limits, and the number of active cores. In a multi-tenant cloud environment like Amazon EC2, unpredictable clock drift can cause inconsistent application performance. To mitigate this, AWS collaborates with Intel and AMD to procure custom-SKU processors or configure firmware power profiles that guarantee a predictable sustained all-core turbo frequency under heavy load.

When an EC2 instance documentation specifies an all-core turbo frequency, it represents the clock speed the processor maintains when all vCPUs on the socket are actively executing standard integer or floating-point instructions. The sole exception on Intel processors occurs during heavy AVX-512 vector workloads on older generations (such as Skylake and Cascade Lake), which trigger thermal downclocking offsets, though newer Sapphire Rapids and Emerald Rapids processors virtually eliminate this penalty for most workloads.

The effective compute throughput \(P\) of a single core executing a workload with an average instructions-per-cycle metric of \(\text{IPC}\) at a sustained clock frequency \(f\) is expressed as:

\[P = f \times \text{IPC}\]

Because \(\text{IPC}\) improves by roughly \(10\%\) to \(20\%\) with each successive microarchitecture, comparing clock speed \(f\) alone across generations does not tell the whole story; however, within the same generation or for strict single-threaded serialization bottlenecks, clock speed remains a critical differentiator.

Intel Xeon Scalable Generations on EC2

AWS deploys custom Intel Xeon Scalable processors across its general-purpose (M), compute-optimized (C), and memory-optimized (R) families. Across generations, Intel has steadily increased both architectural efficiency and sustained all-core clock speeds.

5th Generation: Skylake-SP and Cascade Lake

In the 5th generation, AWS split clock speeds noticeably between general-purpose/memory-optimized instances and compute-optimized instances:

• m5 and r5 Instances: Powered by Intel Xeon Platinum 8175M (Skylake) or 8259CL (Cascade Lake) processors. These run at a 2.5 GHz base clock and sustain an all-core turbo of 3.1 GHz.
• c5 Instances (Original Skylake sizes): Powered by Intel Xeon Platinum 8124M processors with a 3.0 GHz base clock, sustaining 3.4 GHz across all cores.
• c5 Instances (Cascade Lake expansion sizes like c5.12xlarge and c5.24xlarge): Upgraded to custom 2nd Gen Intel Xeon Scalable (8275CL) processors that sustain 3.6 GHz across all cores and reach up to 3.9 GHz on single-core workloads.
• m5n and r5n Instances: Built on Cascade Lake processors (8259CL), sustaining 3.1 GHz across all cores.

6th Generation: Ice Lake-SP

With the introduction of 3rd Gen Intel Xeon Scalable (Ice Lake) processors—specifically the custom Intel Xeon Platinum 8375C—AWS unified the clock speed profile across its core families:

• m6i, c6i, and r6i Instances: All three families share the same 3.5 GHz sustained all-core turbo clock speed. Unlike previous generations where C-series instances ran faster than M- and R-series instances, m6i and r6i match c6i clock-for-clock at 3.5 GHz.

7th Generation: Sapphire Rapids

For 4th Gen Intel Xeon Scalable (Sapphire Rapids) instances, AWS introduced the custom Intel Xeon Platinum 8488C processor:

• m7i, c7i, and r7i Instances: These processors feature a sustained all-core turbo frequency of 3.2 GHz. While the raw all-core clock speed is 300 MHz lower than Ice Lake's 3.5 GHz, Sapphire Rapids compensates with a significant IPC uplift, DDR5 memory bandwidth, and opportunistic turbo boost up to 3.8 GHz (and up to 4.0 GHz on fewer cores) when not all cores on the socket are saturated.

AMD EPYC Generations on EC2

AWS suffix labels ending in a denote instances powered by AMD EPYC processors. AMD EPYC instances in EC2 typically feature one physical core per vCPU in newer generations (disabling Simultaneous Multithreading on 7th generation instances) and offer distinct clock behavior compared to Intel.

5th Generation: Naples and Rome

• m5a and r5a Instances: Powered by 1st Gen AMD EPYC 7571 (Zen 1 / Naples) processors, delivering an all-core frequency up to 2.5 GHz.
• c5a and c5ad Instances: Despite sharing the "5" generation number, c5a skipped Zen 1 and launched directly on 2nd Gen AMD EPYC 7R32 (Zen 2 / Rome) processors, delivering a sustained all-core turbo frequency of 3.3 GHz.

6th Generation: Milan

• m6a, c6a, and r6a Instances: Powered by 3rd Gen AMD EPYC 7R13 (Zen 3 / Milan) processors. Similar to how AWS unified Intel clocks in the 6th generation, all three AMD 6th-gen families sustain 3.6 GHz across all cores, edging out their Intel Ice Lake counterparts (3.5 GHz) by 100 MHz.

7th Generation: Genoa

• m7a, c7a, and r7a Instances: Powered by 4th Gen AMD EPYC 9R14 (Zen 4 / Genoa) processors. These instances push the sustained maximum frequency to 3.7 GHz across all cores. Furthermore, because m7a, c7a, and r7a map 1 vCPU to 1 full physical processor core (no SMT/Hyper-Threading), every vCPU receives uncontested access to the core's execution pipelines at 3.7 GHz.

High-Frequency Specialized Instances

For workloads bounded by single-thread latency or per-core software licensing fees—such as Electronic Design Automation (EDA), relational databases (Oracle, SQL Server), financial modeling, and gaming servers—standard 3.2 GHz to 3.7 GHz clocks may not suffice. AWS provides specialized z1d, m5zn, and x2iezn instances engineered specifically for extreme sustained clock speeds.

z1d Instances (4.0 GHz)

Introduced during the 5th generation, z1d instances use custom Intel Xeon Platinum 8151 processors that sustain 4.0 GHz across all cores simultaneously. They pair this high frequency with up to 384 GiB of memory and fast local NVMe SSD storage.

m5zn and x2iezn Instances (4.5 GHz)

Both m5zn and x2iezn utilize the custom 2nd Gen Intel Xeon Scalable (Cascade Lake) 8252C processor, which holds the record for the highest sustained all-core turbo clock speed in EC2 at 4.5 GHz:

• m5zn: Offers a 4 GiB-to-1 vCPU memory ratio alongside 100 Gbps networking, designed for high-frequency simulation and financial trading workloads.
• x2iezn: Pairs the same 4.5 GHz all-core turbo processor with a massive 32 GiB-to-1 vCPU memory ratio (up to 1.5 TiB of RAM), purpose-built for in-memory databases and large-scale EDA physical verification where per-core license costs dominate total infrastructure spend.

Choosing Between High Clock Speed and Newer Generations

When selecting an EC2 instance based on frequency, raw gigahertz should always be weighed against microarchitectural generation:

1. Pure Single-Thread Clock vs. IPC: While an m5zn instance runs at 4.5 GHz (Cascade Lake), a 7th-generation c7a (3.7 GHz Zen 4) or c7i (3.2 GHz all-core / 3.8 GHz opportunistic Sapphire Rapids) often achieves higher actual work per second due to two generations of IPC improvements, larger L2/L3 caches, and DDR5 memory.
2. SMT vs. Non-SMT Contention: On Intel instances (c6i, c7i) and older AMD instances (c6a), 2 vCPUs share 1 physical core via Simultaneous Multithreading. On 7th-generation AMD instances (m7a, c7a, r7a), each vCPU is a dedicated physical core running at 3.7 GHz, eliminating sibling-thread resource contention under 100% load.