Источник изображения: Igor

NVIDIA GA102 Gaming Graphics Processor Will Be Almost Quarter More Compact than A100


The size of the GPU crystal directly affects both the level of marriage during mass production and the cost of the product. The A100, presented by NVIDIA in May, has no function blocks to accelerate ray tracing, so Ampere architecture gamers will use other GPUs. GA102 will be one of them.

Image Source: Igor's LAB

Image Source: Igor’s LAB

One of the leading NVIDIA A100 development experts, Jonah Alben, in a recent podcast on the company’s corporate blog pages, admitted that the flagship A100 got an area of ​​826 mm2), close to the limits of technological capabilities of TSMC in the framework of the existing 7-nm process technology. If the developers realized that they did not fit into these limits, then they would have to re-develop the A100.

The German resource Igor’s LAB made comments on the area of ​​the GA102 GPU. According to the source, the area of ​​this crystal will slightly exceed 627 mm2. It turns out that, compared with the A100, a decrease in the crystal area by 24% will be achieved. Most likely, the NVIDIA Ampere family will also have more compact consumer-grade GPUs. Otherwise, they couldn’t be the basis for the gaming video cards of the mass price segment.

An illustration appeared on the pages of Igor’s LAB explaining the layout features of GA102-based gaming graphics cards using the PG132 series circuit board. Unlike PG133, which will be used for reference cards of the Founders Edition series, such a printed circuit board has a strictly rectangular shape and a length of not more than 213 mm. There is enough space for the GA102 GPU, and for GDDR6X memory chips located around according to the “4 + 3 + 4 + 1” formula, and for power elements, as well as a pair of additional power connectors.

Finally, the source calls for a proper interpretation of the characteristics of older gaming cards of the Ampere family related to energy consumption. So, the TGP value (350 W) determines the overall energy consumption of the graphics card, and the GPU itself accounts for no more than 230 watts. The difference between these values ​​is distributed among other components: the main part is GDDR6X memory and power elements responsible for powering the GPU.

If you notice an error, select it with the mouse and press CTRL + ENTER.

Leave a Comment