Profit or Poverty: NUMA - Vasilios Syrakis

An introduction to NUMA (Non-Uniform Memory Access) and why mastering it is a hard requirement for HFT micro-optimizations.

Published on April 17, 2026 4 minutes read • While modern software tries to hide hardware complexity behind abstractions, staff working in HFT must dig down to ensure that every nanosecond is spent executing trading logic, not chewed up by operating system overheads or noisy neighbouring processes. Learning the ins and outs of NUMA is not a micro-optimization, it is a requirement. Short history of NUMA Back in the day, CPUs were slower than memory. In the 1960s, processors began to overtake memory in speed and as a result found themselves stuck waiting for data to arrive from a memory…

saved by

• andy24 L

related reading

• What Every Programmer Should Know About Memorypeople.freebsd.org
• How to Optimize a CUDA Matmul Kernel for cuBLAS-like Performance: a Worklogsiboehm.com
• Inside NVIDIA GPUs: Anatomy of high performance matmul kernels - Aleksa Gordićaleksagordic.com
• What Every Programmer Should Know About Memoryakkadia.org
• Look Ma, No Bubbles! Designing a Low-Latency Megakernel for Llama-1B · Hazy Researchhazyresearch.stanford.edu
• Memory access is O(N^[1/3])vitalik.eth.limo
• Your RAM Has a 60 Year Old Design Flaw. I Bypassed It.youtube.com
• Production Twitter on One Machine? 100Gbps NICs and NVMe are fast - Tristan Humethume.ca
• Memory location matters for performancepythonspeed.com
• research!rsc: Hardware Memory Models (Memory Models, Part 1)research.swtch.com
• On the GPU Memory Hierarchygytis.co
• Implementing a fast Tensor Core matmul on the Ada Architecture | spatters.caspatters.ca