Nvidia has pushed back the release for the Kyber rack-scale architecture to contain its Rubin Ultra chips by over a year to 2028, according to news reports on Monday.
The Kyber rack is a server cabinet that aggregates 144 of the firm's more proficient AI chips into functioning as a single powerful computer, that provides enough horsepower to AI companies in order to train and operate their cutting-edge models.
ALSO READ: 'AI No Death Knell': Why Accenture, Cognizant, IBM, Other Software Shares Rallied Up To 7.5%
The rack is designed in such a fashion that the graphics processing units in compute trays are placed in a vertical manner rather than horizontally to improve density and bring down latency.
It was initially set to release in 2027 with the Vera Rubin Ultra, Nvidia's next-generation rack-scale system.
This delay was due to hurdles faced when manufacturing an important circuit board which is central to the system, reports said.
According to SemiAnalysis report, "the Kyber NVL144 rack architecture has been delayed to 2028 as the PCB midplane remains challenging from a manufacturability standpoint."
This is a key printed circuit board which has multiple layers and can connect electronic modules inside a system. The company has also delayed the NVL576, a bigger system that couples eight modules with one another through optical connections.
ALSO READ: Nvidia Shares In Focus As AI Chips Bellwether Plans Raising $20 Billion Via Bond Sale
The reports on the development noted that Nvidia is seemingly facing challenges in matching its speedy release schedules with its manufacturing capabilities.
Nvidia had also reportedly suggested melding two of its current generation racks together to achieve a comparable computing power. The idea was rejected by their customers who deemed the design unwieldly and costly.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
