This article reports a novel GPU-based 2D digital image correlation system (2D-DIC) overcoming two major limitations of this technique: It measures marker-free, i.e. without sample preparation, and the sampling rate meets the recommendations of ASTM E606. The GPU implementation enables zero-normalized cross correlation (ZNCC) calculation rates of up to 25 kHz for 256 × 256 pixel ROIs. This high-speed image processing system is combined with a high-resolution telecentric lens observing a 10 mm field-of-view, coaxial LED illumination, and a camera acquiring 2040 × 256 pixel images with 1.2 kHz. The optics resolve the microstructure of the surface even of polished cylindrical steel specimen. The displacement uncertainty is below 0.5 μm and the reproducibility in zero-strain tests approximately 10-5 (1 σ) of the field-of-view. For strain-controlled testing, a minimum of two displacement subsets per image are evaluated for average strain with a sampling rate of 1.2 kHz. Similar to mechanical extensometers, an analogue 0-10V displacement signal serves as a feedback for standard PID controllers. The average latency is below 2 ms allowing for cycle frequencies up to 10 Hz. For strain-field measurement, the number of ROIs limits the frame rate, e.g., the correlation rate of 25 kHz is sufficient to evaluate 10 images per second with 2500 ROIs each. This frame rate is still sufficient to compare the maximum and minimum strain fields within a cycle in real-time, e.g. for crack detection. The result is a marker-free and non-contact DIC sensor suitable for both strain-controlled fatigue testing and real-time full-field strain evaluation.