It is well known that AMD has been able to overcome Intel's weakness by using small chip technology to reduce the cost of Ryzen processors by 40% thanks to CEO Su Zifeng's small chip plan.
Not only for cpus, but also for AMD's next generation of graphics cards. In fact, the Instinct MI200 series of AMD accelerator cards used small chips last year, and the upcoming RX 7000 graphics card should also use the technology.
Compared with AMD's small chip technology, its rival NVIDIA is still following the traditional GPU design route and has no plan to use small chips. Although the cost is high, the large chip over 800mm2 is also powerful in performance.
In response, NVIDIA Vice President Ian Middot; According to Ian Buck, Gpus are a different kind of performance chip, and in the graphics scene, it's not as simple as presenting a single kernel.
It is understood that in order to not use of small chips around the bottleneck of lithography, NVIDIA chose as opposed to the super chip research and development direction, such as: NVLink interconnection technology is developed, the multiple graphics chips and server, and do this simplifies the programming model, memory sharing allows the CPU to GPU memory to use.