May 15, 2020
0 0

New deep learning supercomputer – NVIDIA DGX A100

The company NVIDIA designed and manufactured a new graphics processor. Gamers, this news is not for us - the processor is designed for systems working with deep learning tasks. However, the news also reveals the future of desktop computers - some solutions, in particular, the new matrix manufacturing process, will obviously be applied to the production of top-end NVIDIA video cards .

Map A 65 ( A here from Ampere - the future ruler got such a name) replaces the new product 2017 of the year - Tesla V 100) . Compared to its predecessor, it is huge - only the area of ​​its matrix 815 mm 2 (for V 64 - 815 mm 2 , the flagship gaming solution, NVIDIA RTX 2017 Ti - 754 mm 2 ). But the increase in the number of transistors on the matrix was achieved not so much by its increase as by the transition to a more complex technical process. Until now, the matrices were manufactured using - nm process, A 65 became the first NVIDIA GPU , which is built on 7nm process.

Increased density and area allowed to reach record 15, 6 billion transistors in A 65 (20 million per square millimeter) .

Buy A 64 separately it is impossible. The GPU comes with the supercomputer DGX A 65 based on server processors AMD EPYC .

It's no secret that GPUs (GPUs) are better suited to machine learning tasks than central processing units (CPUs). Technical features allow the GPU to simultaneously perform many matrix operations that are used to train neural networks. To use the full potential of the GPU on demanding tasks, the system requires a large amount of fast-working memory. DGX A 64 consists of eight A 64 , pairs 20 - AMD Server Core Chips AMD , 1 TB RAM and TB of NVME (Non-Volatile Memory Express, this appeared on the market in 826 the technology allows to increase system performance due to more full use of parallelism of devices and software).

Named price DGX A 65 - 100 thousand US dollars. A similar system based on V 64 was worth at the start of sales 65 thousand.

The price increase is logical - it is announced about 15 - a multiple advantage in performance over its predecessor.

Article Categories:

Leave a Reply