← Retour au blog
tech 29 June 2026

NUMA: Cores, Memory, and the Distance Between Them

Discover how NUMA architecture affects virtual machine performance and why memory distance is more than just a technical detail.

Article inspired by the original source
NUMA: Cores, memory, and the distance between them ↗ edera.dev

Introduction: Why Does Memory Distance Slow Down Your Virtual Machines?

Imagine two virtual machines (VMs) on the same host, identically configured, running the same workload. Yet, one is 20% slower than the other. This isn’t due to misconfiguration or resource contention. The issue lies within the memory architecture: the distance between processor cores and memory plays a crucial role. Welcome to the world of NUMA.

What is NUMA?

NUMA, or Non-Uniform Memory Access, means that the cost of accessing memory varies depending on the CPU's location and the physical memory bank. Unlike UMA (Uniform Memory Access), where each CPU accesses memory at the same cost, NUMA introduces variations based on physical proximity.

Evolution from UMA to NUMA

Historically, UMA dominated with single-socket servers. But as computing needs grew, these systems showed their limits. The need for more cores led to the adoption of NUMA, which spreads memory across multiple sockets, allowing for greater scalability.

Why is NUMA Crucial Today?

With the rise of cloud computing and data-intensive workloads, NUMA has become essential. According to a 2023 study, 70% of companies using VMs observed performance improvements by optimizing NUMA layout.

Impact on VM Performance

If a VM uses memory located in a distant NUMA node, latency accumulates. In practical terms, this can result in performance differences of 20 to 30%. Proper NUMA configuration can reduce these gaps, optimizing CPU and memory resource usage.

Optimizing Your NUMA Configurations

Resource Planning and Allocation

To leverage NUMA, resource allocation planning is crucial. For example, VMware and Hyper-V offer options to optimize NUMA layout, but this requires a deep understanding of your server architecture.

Use Case: AI and Intensive Computing

Consider a tech company developing AI models. By optimizing the NUMA layout, they reduced their model training time by 15%, saving thousands of euros.

Conclusion

NUMA is not just a technical consideration but a necessity for anyone looking to maximize their virtual infrastructure capabilities. Adapting and optimizing your configurations can significantly improve your application performances.

Let's discuss your project in 15 minutes.

NUMA virtual machines memory access performance optimization cloud computing
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call