![]() ![]() ![]() Let’s start with what we know about the Ampere GA100 GPU. The bulk of the computing in this AI workflow is being done by the GPUs, and we will show the dramatic impact of this in a separate story detailing the new Nvidia DGX, HGX, and EGX systems based on the Ampere chips after we go through the technical details we have gathered about the new Ampere GPUs. There are still CPUs in these systems, but they are relegated to handling serial processes in the code and managing large blocks of main memory. And thus, now the massive amount of preprocessing as well as the machine learning training and the machine learning inference can now be done all on the same accelerated platforms. With the Ampere chip, Nvidia has also announced that it has been working with the Spark community to accelerate that in-memory, data analytics platform with GPUs for the past several years, and it is now also ready. Nvidia has created a single GPU that can not only run HPC simulation and modeling workloads considerably faster than Volta, but also converges the new-fangled machine learning inference based on Tensor Cores onto one device. The Ampere chip is the successor not only to the “Volta” GV100 GPU that was used in the Tesla V100 accelerator announced in May 2017 (aimed at both HPC and machine learning training workloads) but, as it turns out, the chip is also the successor to the “Turing” TU102 GPU used in the Tesla T4 accelerator launched in September 2018 (aimed at graphics and machine learning inference workloads). Here is the most important bit, right off the bat. All of the feeds and speeds and architectural twists and tweaks have not yet been divulged, but we will tell you what we know and do a deep architecture dive next week when that data is available. The in-person GPU Technical Conference held annually in San Jose may have been canceled in March thanks to the coronavirus pandemic, but behind the scenes Nvidia kept on pace with the rollout of its much-awaited “Ampere” GA100 GPU, which is finally being unveiled today. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |