[ad_1]
The new NVIDIA A100 comes to fight in the HPC market. Just an hour ago we got the news that AMD was launching its Instinct MI100 graphics. This promised to be the best in the world for computing, but soon after NVIDIA reacted with this new 80GB A100, which promises exactly the same. Was it the most ephemeral leadership for AMD? Let’s see it.
NVIDIA A100 80 GB, a monster for computing and artificial intelligence: the throne of GPUs resists AMD
The new graphics enjoy all the advantages of the Ampere architecture that made the A100 so powerful, and its biggest advantage is in the memory bandwidth and amount of it. For the first time, 2 TB / s of bandwidth exceeded, with double the capacity (80GB compared to the original A100).
As they say, both AMD and NVIDIA proclaim themselves the throne of computing with artificial intelligence. And it seems that both will find their key segment, no doubt NVIDIA will dominate in those cases where an exaggeratedly massive amount of data needs to be processed. Its fast 80GB will be able to give quick results in AI training with billions of users and products, for example, or train the GPT-2 (which generates long text in natural language) without more than one GPU.
Check out our news on the Radeon Instinct MI100 24GB.
The graph uses all the technologies of Ampere and its predecessor, such as MIG (Multi-instance GPU), which allows you to partition the graph up to 7 instances with 10GB of memory each, fully insulated safely to perform multiple tasks simultaneously or extreme performance in specific applications such as sparse arrays.
NVIDIA DGX Station A100 platform, an impressive supercomputer in a small
The other version of NVIDIA was his DGX station second generation. This is basically an awesome server with two CPUs AMD Epyc Rome 7742 of 64 cores each (it is clear who dominates you) e 4x A100 80GB GPU, with a total memory of up to 320 GB, which can also be divided into up to 28 separate instances to run 28 different and completely isolated jobs, and without affecting performance.
These GPUs will be interconnected by NVLink and the truth is the equipment is the closest thing to a supercomputer in terms of server size, and without special cooling needs. Awesome.
We advise you to read the best graphics cards on the market.
It is clear that the battle has served between AMD and NVIDIA. The latter was enjoying an extremely comfortable position, but now he began to respond with agility before the continued progress of its main competitor.
Source link
