- Huawei wants UB-MEH to unify the fragmented interconnection standards in massive groups
- The UB mesh design combines the closed spine with multidimensional meshes at the scalability frame
- Traditional interconnections grow too expensive in large -scale implementations
Huawei has revealed open source plans its UB-MEH interconnection, a system destined to unify how processors, memory and network equipment communicates in massive data centers.
The UB mesh design combines a spine based on the level of the data room with multidimensional meshes within each rack.
When combining these topologies, Huawei states that it can maintain costs under control, even when system sizes are reduced in tens of thousands of nodes. He also hopes to solve the problem of climbing the workloads of AI, where latency and hardware failures represent barriers.
Replacement of fragmented standards with a single frame
The movement is presented as a way to replace multiple overlapping standards with a single frame, potentially remodeling how large -scale computer infrastructure is built and operates.
In simple terms, Huawei wants to replace the current combination of different connection rules with a universal system, so everything joins easily and more economically.
“Next month we have a conference, in which we are going to announce that the UB-SMEH protocol will be published and revealed to anyone as a free license,” said HENG Liao, head scientist of Hysilicon, Huawei’s processor arm.
“This is a very new technology; we are seeing efforts of competitive standardization of different camps. […] Depending on how successful we are in the implementation of real systems and demands of partners and customers, we can talk about turning it into some kind of standard. “
One of the central arguments behind UB-WSH is that traditional interconnections grow too expensive at scale, and eventually cost more than the accelerators that must be connected.
Huawei indicates its own manifestations, where an implementation of 8,192 nodes was used as evidence that costs do not need to increase linearly.
This is framed as essential for the future of AI systems built with millions of processors, high -speed network devices and massive storage matrices, such as the largest SSD systems used in cloud storage operations.
UB-MESH is part of a broader idea that Huawei calls the supernodium. This refers to a central data scale cluster where CPUs, GPUs, memory, SSD units and switches can function as if they were inside a single machine.
The bandwidth statements of more than one Terabyte per second per device and the latency of submicroseconds are being positioned as proof that the concept is not only possible but necessary for next generation computing.
However, standards such as PCIE, NVLink, Uualink and Ultra Ethernet already have a support of multiple companies in the semiconductor and networks industries.
The question now is whether the industry will accept a new protocol backed by Huawei or will continue to favor standards already backed by a wider range of companies.
Huawei’s proposal, although ambitious, places customers in the position of adopting a protocol owned and controlled by a supplier.
Even with the open source license, there are concerns about long -term interoperability, governance and geopolitical risks.
That said, Huawei’s technical potential sounds impressive, but its movement demands a degree of confidence and adoption throughout the industry that has not yet assured.
Via Hardware Toms
