By AMD intends to unite its CDNA and RDNA architectures into a “new DNA,” or “unified DNA,” that will combine the architectures utilized by its data center and consumer graphics architectures, according to a company executive.
In addition to AMD’s modification to its consumer Radeon GPU ambitions to target mainstream but not flagship enthusiast goods, the objective is to include more developers in the fold.
AMD is attempting to assemble a core group of software developers who can create applications for Radeon technology, similar to Steve Ballmer’s well-known campaign to hire developers. According to Jack Huynh, senior vice president and general manager of Computing and Graphics at AMD, if the company focuses on the small percentage of buyers who will purchase a flagship GPU, it will not draw in as many developers as it would if it were to promote mainstream products during a breakfast meeting for reporters on Friday at the Berlin IFA show. Although Huynh initially referred to the new coordinated effort as “new DNA,” he also used the term “unified architecture,” or UDNA.
Huynh stated, “We have a CDNA architecture [for data centers] and RDNA for the consumer side. This is part of the big transformation at AMD.” “From now on, we will refer to it as New DNA. To consolidate it, there will be a single, uniform design for both the commercial side and Instinct, AMD’s enterprise GPU architecture. The developers will have it so much easier. They have to make a decision today, and it will not get better.
We forked it because, although you get the recommended issues and sub-optimizations, it becomes quite challenging for these engineers, especially as our data business grows. Now that we have unified it, we might introduce it. That will be of great assistance to it. Because, as I already mentioned,
Huynh noted that the process will take time. But for that reason, he remarked, “I am telling the team right now.” “We need to build to that scale right now.”
Huynh claimed that AMD had developed RDNA with “mistakes.” “We have to reset the matrix on the optimizations every time we update a subsystem’s memory,” he stated. “That is not what I want to do.”
Huynh stated that he did not want to alter the memory hierarchy and thereby lose all of the optimizations while creating new RDNA generations. “It is quite feasible; it just needs a lot more work and careful planning to accomplish. However, it is the course.
According to Toms Hardware, AMD has not provided robust AI support for its consumer GPUs, but with these modifications, AMD may be able to take on Nvidia’s CUDA architecture and DLSS more directly. Because of CUDA’s extensive developer support and the sheer power of Nvidia GPUs, Nvidia has been a preferred architecture in consumer AI generation applications, such as AI art. However, when will this take place? And will it work? It appears that AMD will be providing further details on this proposal in the coming weeks.
Further reporting was done by Adam Patrick Murray. Murray was accommodated at a hotel by Lenovo. IFA at Qualcomm’s request; Qualcomm covered his accommodation and travel expenses. The editorial content was not influenced by either company.
