A research project of the Cybear Security Lab.
Modern processors contain internal hardware structures, such as caches and branch predictors, that programs cannot see directly but that affect how code runs. Security research has shown that these structures can be used in ways their designers did not intend.
Published results in this area are usually measured on one or two processors. This project asks how well those results hold up across different Intel and AMD processor generations. We rebuild published techniques and measure them the same way on every machine, so the results can be compared.
The work is defensive. Its goal is to tell defenders and hardware designers which behaviours are general and which depend on a specific processor.
Our local testbed covers a few Intel and AMD processors. To draw general conclusions we need more processor generations, especially recent server parts. The measurements require:
- Bare-metal nodes. Virtual machines add noise and hide the hardware behaviour we measure.
- Root access. We pin work to single cores, fix the CPU frequency, and read hardware performance counters.
- A range of Intel and AMD CPUs. Different generations are the point of the study.
- Each experiment is a small, single-node program that runs on our own allocated machine.
- Experiments use no network traffic beyond our own SSH access and file transfer.
- Nothing targets other users, other tenants, or any outside system.
- Runs are short and repeated many times for statistical confidence, so we expect modest, bursty use of each node.
Active, 2026–2027. Code and results will be released alongside publications.
Cybear Security Lab — via the GitHub organization CybearsecurityLab.