Apr 04 2024, 09:29
Nvidia's 3-barrel shotgun keeping competitors at bay while keeping customers:
Why are competitors finding it so hard to eat Nvidia's lunch? We outline three key reasons that give this chip company an unassailable advantage for the near future.
Dominance in the technology industry can be notoriously short. Companies like DEC, IBM, Cisco, Fairchild, Nokia, BlackBerry have all had their moment in the sun only to see it set faster than expected. There are exceptions of course, Intel, Microsoft, and now Apple, that have maintained their competitive edge for long duration. The latest to join the "seemingly unassailable" is Nvidia.
This company made its first millions through the graphical processor unit (GPU) route. As personal computers became more visual in their interface and started offering games as a part of the software ecosystem, the microprocessor companies such as Intel or AMD realised that such visual computing needed specialised components.
Take for example, TVs or mobile screens. When you get into manipulating large images, especially when you get to an HD display that has 2,000,000 pixels on it or now with 4K displays four times that it is tough task. "It takes a tremendous amount of multipliers to be able to scale or rotate images, to be able to apply, say, colour filtering," says Loftus.
Another application is gaming. In the gaming environment, you take a reference picture and are manipulating it in ways to allow the character to be able to walk around, and backflip on a screen. "What's happening there is you have this reference image that is stored in the memory and the GPU is able to bring in millions of bits of information that represents that image, and it's able to do these multiplications to be able to shrink, expand, turn an image that is made up of millions of units," says Loftus.
Nvidia (and other such GPUs but Nvidia is the biggest) have specialised multiplication units to do this. "GPUs had hundreds, and now have thousands and thousands, of dedicated multipliers with specialised memory management units that are able to bring in a large array of data. In parallel, these units push it through hundreds or thousands of multipliers and write that data back out to these very high bandwidth memories," says Loftus.
source:et
Apr 04 2024, 09:42