The RISC-V architecture was born at the University of California at Berkeley, in 2010, with an essentially educational purpose, but also with the intention of establishing itself in the medium term as a free hardware alternative both to the x86-64 designs from Intel and AMD and to the CPUs with ARM architecture that have proliferated so much in recent years. As we can deduce from its name, it is a RISC (reduced instruction set computer) type design.
Its most obvious attraction is that since it is a free and open architecture, it is available to any person, institution or company to be used and modified without paying any fee. This is a very powerful point in its favor, and makes it attractive for a very wide range of use scenarios, such as personal computers, embedded systems or supercomputers, among other machines. Yes, also for supercomputers.
NVIDIA also believes in RISC-V
The company led by Jensen Huang has earned the reputation of being a company that above all relies on proprietary technologies. It is their right, of course, but this strategy has caused some users and companies to distrust it and have decided to opt for open rival solutions, such as those proposed by AMD and other companies. Despite all this, something is changing at NVIDIA. And little by little and without making much noise it has been abandoning its proprietary microcontrollers and replacing them with RISC-V chips.
The number of microcontrollers that we can find on an NVIDIA printed circuit board depends on the complexity of the GPU
NVIDIA GPUs, whether or not dedicated to artificial intelligence (AI), coexist with a large number of microcontrollers that, broadly speaking, reside on chips designed to manage behavior of proprietary cores, such as CUDAs. The number of microcontrollers that we can find on an NVIDIA printed circuit board depends on the complexity of the GPU, but typically each of them has between 10 and 40 RISC-V cores customized by NVIDIA itself.
Jensen Huang’s company began replacing its proprietary microcontrollers with microcontrollers with RISC-V cores in 2015, and has since developed at least three such cores: NV-RISCV32, NV-RISCV64, and NV-RVV. Today, practically all of its microcontrollers are RISC-V, so it seems that its proprietary Falcon microcontroller has little by little been disappearing from its product portfolio. Before concluding this article, it is worth noting an important fact.
NVIDIA distributes many millions of GPUs discreet every year. In 2023, in fact, shipped 31 million GPUs for desktops, about the same amount for GPUs for laptops, several million GPUs for data centers and other types of hardware. With this in mind, it is not surprising that by the end of 2024 this company will have integrated approximately 1 billion RISC-V cores into its products this year. That’s where it is.
Image | NVIDIA
rmation | Tom’s Hardware
In Xataka | “It was 100% our fault”: the head of NVIDIA does not ignore the problem when things are wrong
Add Comment