4. Hardware developments

4.1 Microarchitectural generations

Over these ten years, six micro-architectural generations have followed one another. In Nvidia's terminology, this corresponds to the notion of "Compute Capability", which varies from version 1 to version 7 and defines the available functionalities, which we'll detail in the following sections.

The table 2 shows the main features of the successive "high-end" generations of Nvidia GPUs. Successive technology nodes, with the spectacular growth in the number of transistors available, enable an increase in the number of multiprocessors and cores on the one hand, and the size of the various memories on the other (L1 cache and shared memory first separated, then merged,...