For convenience, manufacturers usually sort their chips into several categories. After that, the chips are sent to the product manufacturers according to their requests. For example, the most successful chips are usually used in the most productive and expensive solutions based on them. And less successful, but still corresponding to the declared characteristics, as a rule, are used in more affordable solutions.

Perhaps one of the most difficult is the binning (sorting) process in GPU manufacturing. There are a lot of video card manufacturers on the market, and each of them may have different requirements for chips for a particular video card model, even within the same series. NVIDIA and AMD have to spend a lot of effort to select the best GPUs and weed out the lower quality ones.

According to Igor’s Lab, NVIDIA has provided three categories of GPU binning for the latest series of GeForce RTX 30xx video cards based on the Ampere architecture. The “Bin 0” category includes chips that correspond to the characteristics and performance level originally laid down in them, but no more; higher quality processors get into “Bin 1”; and in “Bin 2” there are only elite chips, with the most outstanding characteristics and performance level. As a result, GPUs in the Bin 2 category run cooler than Bin 0 and Bin 1 and also exhibit higher overclocking potential.

In practice, binning also means very close coordination between the developer (in this case NVIDIA) and the manufacturing contractor (Samsung with its 8nm 8N process technology).

According to the source, as part of the pilot production of Ampere GPUs, the South Korean giant issued 30% of the chips in the Bin 0 category, 60% in the Bin 1 category, and 10% in the Bin 2 category. This is quite a good figure for Samsung, given the short duration of the test production. Probably, in the future, the company will be able to increase the production of better quality chips.