NVIDIA CEO Jen-Hsun Huang Heats Up GTC 2020 by Pulling Ampere-Based AI Solution Out of the Oven
This week, NVIDIA will unveil its new Ampere architecture GPUs at GTC 2020. The Ampere architecture is focused on cloud computing, deep learning, AI, and other related technologies. Prior to the presentation, NVIDIA CEO Jen-Hsun Huang unveiled the new DGX A100 platform. The humorous touch is that he pulls this professional solution out of an oven.
We can see in the image a high-level solution with characteristics that will be revealed in a few days. This supercomputer would be based on solutions based on the GA100 and that will offer great performance. We do not know what is hidden under the heatsinks, but they will undoubtedly be solutions that will give a lot to talk about.
Jen-Hsun Huang takes an Ampere solution out of the oven
The previous generation of these specialty solutions for advanced computing implemented x16 Volta GPUs. For this solution we see only 8 heatsinks, which makes us think of only x8 Ampere GPUs. They have surely managed to improve performance and not exactly by little, with less GPU and therefore a much more efficient platform.
Ampere, theoretically will be manufactured in the high-end in 7nm by TSMC and the low-end in 8nm by Samsung. Turing is based on 12nm lithography, so the jump in efficiency and performance would be quite significant. According to the previous data, in Ray Tracing we would talk about up to x4 times more performance. We do not know in other fields, but come on, the performance jump can be spectacular.
Buying the current Ampere-based solution for AI, Deep Learning, and other applications, and upgrading it to Volta, will be a massive leap forward. Of course, we're talking about Volta, not Ampere, so the performance and power improvements could be staggering. NVIDIA is very comfortable in the professional segment, and it seems unlikely that AMD will come close.
