When companies deal with proprietary systemsoften others try offer the same but open for the community. This is the case of NVIDIA with its NVLinka system for connect multiple GPUs used in data centers that need scalable systems for AI. To offer a system of high speed and low latency in AI accelerators that can compete with NVIDIA, they have aligned until 8 great companies of computing that will develop ULink.
This Ultra Accelerator Link will be promoted by AMD, Broadcom, Cisco, Google, HP Enterprise, Intel, Meta and Microsoft. This ULink will offer an open standard for users to AI accelerators can communicate effectively and quicklyand so that it can be develop easier integration with more flexibility and scalability in data centers that require multiple GPU connections to deliver higher performance in AI tasks.
The first version of these specifications will allow the connection of up to 1,024 accelerators and the direct loading and storing between connected memory by these accelerators. A complete version of these first specifications should be available for the third quarter of this year, by the UALink Promoter Group. With this new connection it is intended open these types of connections to all manufacturers and developers and promote an open ecosystem in which solutions are offered that will benefit the industry.
End of Article. Tell us something in the Comments!
Juan Antonio Soto
I am a Computer Engineer and my specialty is automation and robotics. My passion for hardware began at the age of 14 when I broke down my first computer: a 386 DX 40 with 4MB of RAM and 210MB of hard drive. I continue to give free rein to my passion in the technical articles I write for Geeknetic. I dedicate most of my free time to video games, contemporary and retro, on the more than 20 consoles I have, in addition to the PC.
Add Comment