Hyperthreading, commonly known as Simultaneous Multi-Threading (SMT), is a technology used in certain computer architectures to allow a single physical core to appear as two separate CPUs, often referred to as virtual CPUs (vCPUs), in user environments. This enables the operating system to schedule two different processes to run concurrently on these vCPUs. The CPU scheduler manages the execution pipeline to allocate compute time to each process effectively

In multi-tenant environments, where applications are not consistently CPU-intensive, hyperthreading (or SMT) enhances resource efficiency by minimizing CPU wait times for processes. In cloud applications, especially those sensitive to latency, sharing a CPU with another application that consumes more than 50% of available CPU time can lead to increased wait times for your process. This phenomenon, known as the "Noisy Neighbor" problem, results in increased tail latency, affecting user experience.

However, with Cloud Native Processors like Ampere CPUs, this issue is mitigated. Ampere cores are single-threaded, ensuring that no core is shared with another process. This guarantees that you can fully utilize the compute power for which you are paying, without interference from other processes.

• Looking beyond SMT in the Cloud: https://amperecomputing.com/blogs/looking-beyond-smt-in-the-cloud
• Adopting Ampere at Scale – How Uber Transitioned to Arm-based Compute in an x86 Fleet: https://amperecomputing.com/blogs/how-uber-transitioned-part-1
• Momento Migrates Object Cache as a Service to Ampere Altra: https://amperecomputing.com/assets/developer_story_2_629ebb9527.pdf