• Home
  • Hardware
  • NVIDIA Announces Its Most Advanced Chip Ever Built Specially for Artificial Intelligence: Here Are Its Incredible Features

NVIDIA Announces Its Most Advanced Chip Ever Built Specially for Artificial Intelligence: Here Are Its Incredible Features

NVIDIA announced the GH200, the most advanced chip on which artificial intelligence systems can be trained and used.
 NVIDIA Announces Its Most Advanced Chip Ever Built Specially for Artificial Intelligence: Here Are Its Incredible Features
READING NOW NVIDIA Announces Its Most Advanced Chip Ever Built Specially for Artificial Intelligence: Here Are Its Incredible Features

NVIDIA, one of the hardware architects of the artificial intelligence industry as well as performance hardware for work and play, announced its new chip to be used to run productive artificial intelligence. The chip is the GH200, the new member of the artificial intelligence chip series GH.

According to the announcement shared by the company, NVIDIA GH200 can be used for vector, recommender and large scale language models, which can handle the most complex and burdensome generative AI processes. On the other hand, the chip will carry the same graphics processor as the company’s current most powerful artificial intelligence chip, the H100.

Same graphics processor but higher memory will be offered:

Although the chip will carry the same processor as the H100 graphics processor, it will offer 3.5 times more memory (282 GB) than this unit. Combined with the performance offered by the fourth-generation Tensor cores and Transformer Engine in the H100, this memory will offer higher processing power than a standard user can imagine.

Also, it’s not just the amount of memory that will change. New HBM3e memory will also be made available with the GH200. Although no data has been shared about this memory yet, it will be 50% faster than the previous generation.

NVIDIA’s H100 graphics unit can deliver data transfer rates up to 3.35 TB per second. Comparing this to today’s most powerful graphics processor for standard consumers, the NVIDIA GeForce RTX 4090 supports transfers up to 1TB per second. In the GH200, this number will increase to 10 TB per second.

Moreover, NVIDIA’s artificial intelligence chips are not chips used alone. Thanks to the company’s “NVLink Switch System” technology, 256 processors can be used in conjunction with each other at the same time. More than 850 TB of data can be transferred per second (according to H100 series data). Thus, artificial intelligence systems with trillions of parameters are operated and trained.

NVIDIA has yet to reveal the price of the GH200. However, it should be noted that the H100 series graphics units are sold for about 40 thousand dollars today.

Comments
Leave a Comment

Details
137 read
okunma55022
0 comments