MgO will participate in the development of hbmnext memory standard, which is expected to be released in 2022

Nowadays, there are two kinds of video memory widely used: gddr and HBM. The latest ones are gddr6 and hbm2e. Recently, MGG said that it would participate in the development of the standard for hbmnext video memory. It is expected that hbmnext will be available in 2022, but no specific technical details have been determined yet. < / P > < p > SK Hynix was the first manufacturer to launch hbm2e video memory, and it had mass production as early as July this year. SK Hynix’s hbm2e has the characteristics of large capacity, high speed and low power consumption. It uses TSV silicon perforation technology to stack eight 16GB chips vertically, so as to achieve a single capacity of 16GB, twice that of the previous generation hbm2. In theory, it can stack 12 chips and single chip capacity 24GB. < / P > < p > if four combinations are used, the total capacity is 64GB, the total bit width is 4096 bit, and the total bandwidth is more than 1.8tb/s. However, hbm2e of MgO did not appear until recently. Two specifications are provided. One is four stack, single die capacity is 8GB, and the other is eight stack and single die capacity is 16GB. The corresponding single capacity is 8GB and 16GB, and the data rate is 3.2gbps or higher. When combined with 1024 bit width, the total bandwidth is 410gb / s. If the four chips are combined, a total capacity of 64GB and a total bandwidth of 1.64tb/s can be obtained. < p > < p > in gddr memory, MgO recently revealed that the RTX 3090 of NVIDIA’s next generation RTX card will carry the latest gddr6x memory. However, the gddr6x memory of RTX 3090 is still not the strongest compared with NVIDIA’s latest A100 accelerator card. The A100 accelerator uses 5120bit hbm2 video memory with a bandwidth of 1.6tb/s, easily surpassing the RTX 3090. But after all, the bandwidth required by the game graphics card is not as high as that of the professional graphics card, and the computing performance can not be compared with the Professional Computing card. Baidu continues to work on quantum computing to lay a solid foundation for new infrastructure construction