(ORDO NEWS) — NVIDIA announced the first graphics chip of the new series based on Ampere architecture. This was reported by the portal VentureBeat. The model received the designation A100. Representatives of the company said that their new product “demonstrates the biggest leap in performance compared to previous generations of GPUs.”
NVIDIA A100 will be manufactured on the basis of a seven-meter process technology. The main purpose of the chip is to accelerate the processing of tasks that are associated with artificial intelligence, scientific calculations and the analysis of large amounts of data. According to the developers, the new chip will increase the computing speed for AI by about 20 times compared with previous generations of graphics processors.
On a chip with an area of 826 mm 2 there are more than 54 billion transistors. In addition, the NVIDIA A100 has six Samsung HBM2 memory stacks with a total capacity of 40 GB. The total number of CUDA cores is 10,368: of which 3,456 are reserved for double-precision floating-point calculations.
High-performance servers based on new chips will be released by companies such as Atos, Fujitsu, Dell, Gigabyte. However, NVIDIA itself will be the first to launch such a system: it will receive the designation DGX A100 and will include eight A100 chips connected via the NVLink bus, two 64-core central processors from AMD, 1 TB of RAM and a 15 TB solid-state drive. The cost of such a set will be 200 thousand dollars.
Previously, Adobe created an algorithm that could detect faces that were edited in a photo editor on images, and errors in Radeon video card drivers prevented volunteers from looking for alien civilizations.
Contact us: [email protected]