Clemens Lutz from DFKI published a Blogpost to the SIGMOD Paper 'Pump Up the Volume: Processing Large Data on GPUs with Fast Interconnects' for the ACM SIGMOD/PODS International Conference on Management of Data, Portland, USA in 2020. The paper is written by Clemens Lutz, Sebastian Breß, Steffen Zeuch, Tilmann Rabl, Volker Markl.
Recently, multiple GPU vendors have released or announced GPUs with interconnects that provide unprecedented connectivity to the CPU and main-memory. NVLink 2.0 by Nvidia , Infinity Fabric by AMD , and Compute Express Link by Intel [4, 5] are a new class of fast GPU interconnects. These promise bandwidth up to 150 GB/s, and offer a new solution to an age-old problem described in database literature called the data transfer bottleneck.
In our recent publication that was accepted at the ACM SIGMOD conference , we investigate the benefits and challenges of fast interconnects for processing data-intensive applications on GPUs. In this post, we will cover what fast interconnects are capable of today, and why they might change the game for GPU-enabled data processing.
Find the whole Blogpost here