SUSE Labs Conference 2019 - Accelerate your AI Cloud infrastructure: a virtualization perspective
None
AI/DL(Deep Learning) has become quite popular today, also comes with a lot of DL frameworks. Among them, a hardware accelerator is necessary when dealing with huge data processing. Also, in order to relieve the pain of framework setup for researchers, cloud deployment becomes an urgent request. Based on such background, this session will give a full explanation on how hardware accelerators(GPU, FPGA, ASICs) are used for AI/DL and their virtualization implementation.
First part, we will have a look on several hardware accelerators and their upstream status for DL support: hardware like GPU from Nvidia and AMD, Nervana from Intel, TPU from Google, FPGAs and some AI SOCs. We will see the how they accelerate and the difference between them.
Second part, we will focus on virtualization implementation for them. First, we will have a update on the GPU Virtualization from last year's session, add AMD part and some user response. Then, we will have a quick look on AI Chips and FPGAs: how to put them in a Virtualization environment, the difficulties, and so on.
Last, we will see how to migrate this infrastructure to a containerization environment. And most importantly, let us discuss how SUSE could help all of them.
Liang Yan
https://labsconf.suse.cz/labs2019/tal...