| Developers: | Nvidia |
| Date of the premiere of the system: | 2022/03/25 |
| Branches: | Electrical and Microelectronics |
| Technology: | Processors |
Content |
The main articles are:
2025: Differences between the original and counterfeit
In the booming GPU (GPU) market, it is important to distinguish between original and fake products - especially in the high-end model segment such as Nvidia H100. This document discusses the main characteristics of the named product, common purchase errors and practical tips for recognizing genuine accelerators (Nvidia HGX H100 4 GPU 80GB). Read more here.
2022: Announcement of the Hopper silicon architecture and the GPU based on it H100
On March 25, 2022, NVidia announced a number of technologies focused on the use of artificial intelligence. First of all, this is the Hopper silicon architecture, created on its basis by the GPU H100 and the Grace CPU "super processor."
According to the company, the architecture and superprocessor are named after Rear Admiral of the Navy USA Grace Hopper (Grace Hopper), the creator of the first ever compiler computer programming language and one of the first high-level programming languages. COBOL
The Hopper architecture and GPUs are H100 designed to optimize the learning process of machine models. The fourth generation NVLink bus allows you to connect up to 256 GPU H100, providing a nine-fold increase in bandwidth compared to previous generations.
The H100 processor itself contains 80 billion transistors. At the same time, it supports PCIe Gen5 and HBM3; thanks to this, the memory bandwidth reaches 3 terabytes per second. Nvidia claims that the H100 is three to six times higher than the previous generation A100 processor in floating point computing, depending on bit size.
Depending on the form factor, the characteristics of the GPU will differ slightly. So, the older version has 15,872 CUDA cores.
Both versions have 80 GB of memory, and this is a HBM3 with a bandwidth of 2 or 3 TB/s. The TDP of the older version is 700 W, while the modification with PCIe 5.0 has half the figure.
The performance of the SXM5 version reaches 30 TFLOPS (FP64), 60 TFLOPS (FP32), and the PCIe 5.0 version has 24 and 48 TFLOPS, respectively. It is also worth noting the NVLink interface, which in the same order provides a data transfer rate of 900 or 600 GB/s.
As for the Grace CPU, these are actually two processors connected by the NVLink-C2C bus. Grace is designed for use in "hyper-scale high-performance computing and AI" complete with H100 or on its own. The processor has 144 Arm cores, and its bandwidth is 1 terabyte per second[1].
The Hopper architecture also brings fourth-generation tensor cores to the market, which are nine times faster than the previous generation in a number of machine learning tasks. Separately, Nvidia noted a set of DPX instructions that will help developers write code to speed up dynamic programming algorithms in various industries, speeding up workflows for disease diagnosis, quantum modeling, graph analysis and routing optimization.
Hopper-based systems will be available to Nvidia partners in the third quarter[2].
Notes
| The site content is translated by machine translation software powered by PROMT. The machine-translated articles are not always perfect and may contain errors in vocabulary, syntax or grammar. Read original article If you find inaccuracies or errors in the results of machine translation, please write to editor@tadviser.ru. We will make every effort to correct them as soon as possible. |