University Scholars, Nvidia and IBM Join the forcesTo facilitate the work of machine learning. The researchers plan to connect GPUs directly to SSDs.

The idea is called super accelerator memory (BaM) and consists of connecting graphics processors with large amounts of SSD memory, which will remove the so-called bottleneck in the machine learning process.

BaM uses the GPU cache along with a library where graphics threads can directly request data stored on NVMe SSDs. All this without the involvement of the wizard, which will act as a limitation.

The goal of BaM is to expand the memory capacity and increase the effective bandwidth of the GPU, which will allow easy on-demand access to large data structures. BaM can be used primarily in data centers and supercomputers, with many graphics cards and disks.

The technical details are quite complicated, but if you are interested in them, you can read about this solution here. In the end, the project will be made available on an open source basisWhich is really great news for people who are interested in machine learning.

We invite you to our new YouTube channeltvtechIt is dedicated to issues related to new technologies. There you will find many tips on problems with computers, consoles, smartphones and player equipment.

Reservation

GRY-OnLine SA receives commission from selected stores whose offers are shown above. However, we have done our best to choose only fun promotions – First of all, we want to publish the best deals for equipment, gadgets and games.

See also  Apple M1 benchmarks and how to translate them into real use