The recent NVIDIA transition to the partial opening of the Nvlink interconnex through the Fusion Nvlink program caused significant interest in the industry, because eight companies expressed interest in the development of semi-non-standard processors and silicone for the NVIDIA ecosystem. However, as it turns out, Nvidia has retained some key nvlink components closed from its partners, and may even decide that they do not grant a license for some products and secure their entry into the ecosystem, informs Digitimes.
Alchip Technologies, one of the few companies that allows you to get involved in the program, claims that although it can develop a full implementation of NVlink, warns that his clients may have difficulty implementing solutions if Nvidia refuses to access some layers of NVlink protocols. This is due to the fact that the green company has access to protocols (except for certain levels) goalked when controlling the entire software stack. In addition, NVIDIA orders a special license of pages that want to utilize NVSWITCES to connect to the graphic processor, which further complicates the adoption of this technology more or less broadly.
The Nvidia Nvlink program does not make Nvlink an open industry standard under any circumstances. They only choose companies among NVIDIA partners, they gain access to it. For now, this is a term and summary of IP companies; Designer of Alchip Technologies contract chips; Developer of specially built Solutions of the LABS ase communications; and designers of CPU Fujitsu, Marvell, MediaTek and Qualcomm. However, there are a bit: only processor designers seem to be able to implement NVlink on both hardware and software levels.
Technologies such as NVINLink NVIDIA usually utilize many protocols, including physical layer protocols (PHE), data link layer protocol, transport protocol, memory coherence protocol, nuclear surgery protocol and Numa protocol. While companies such as Alchip can access everything related to the hardware, the software layer, which is responsible for initiating and configuring links, is still controlled by NVIDIA, and therefore Alchip customers must obtain a license and approval.
In addition to limiting the availability of IP Fusion Nvlink to selected partners (which is not bad, so to speak), NVIDIA also requires the actual developers of the platform to implement at least one NVIDIA product (CPU, GPU, Switch) to their platform, which brings additional revenues, but slightly reduces the flexibility of such inventions.
Such strongly controlled access to NVLink limits the practical utilize of external equipment and limits trouble -free interoperability provided by open standards (although Nvlink does not seem to be an open standard). Analysts interrogated by Digitimes say that the dominance of NVIDIA in AI equipment and its vertical control over software, prices and supplies are not motivated for true openness. That is why the Fusion Nvlink program more resembles a calculated effort to maintain good will among selected partners, expand the NVlink ecosystem in a controlled way, enables customers to look for the sovereignty of building their own processors or special and special accelerators.
At the end of the day, the biggest competitive risk for NVIDIA is not a single, very successful product on the part of rivals, but a competitive ecosystem in which many competitive processors adapted to specific loads. While NVIDIA cannot stop the appearance of such an ecosystem, it can either delay the moment when it develops, or somehow prepare for its appearance with the assist of selected partners, while maintaining its record margins.
Follow Tom’s equipment in Google News To get current messages, analysis and reviews in your channels. Click the Fight button.
