NVIDIA RTX 3000 Ampere cards might be built on Samsung’s 8nm process instead of TSMC’s 7nm node
According to tipster @kopite7kimi, NVIDIA will be making use of Samsung’s 8nm fabrication process to build its GeForce RTX 3000 series Ampere GPUs, and not TSMC’s 7nm process as was expected. Samsung’s 8nm node is an optimised derivative of the 10nm process used to fab chips like the Samsung Galaxy S8’s Exynos 8895 SoC. This means that, relative to its 10nm process, transistor density hasn’t been increased substantially.
This is unlikely to have performance implications in the here and now. After all, leaked benchmarks of RTX 3000 series parts (presumably fabbed on 8nm silicon), indicate that NVIDIA is set to retain performance leadership this generation.
Being a generation behind on the fab process hasn’t prevented NVIDIA from delivering superior performance in the past. NVIDIA’s 12nm Turing GeForce RTX 2080 Ti is currently the fastest consumer GPU on the market by a considerable margin.
However, the decision to leverage Samsung’s 8nm process could explain some of the more unusual Ampere rumours. For instance, rumors about extremely high Ampere power draw - with GA102 TDPs in excess of 300W - make more sense now. NVIDIA will have to increase power consumption to feed these relatively less efficient parts.
If they’re being built on Samsung’s 8nm process, NVIDIA’s Ampere parts will likely be less efficient than AMD’s RDNA2 GPUs. However, until we hear more about AMD’s 7nm RDNA2 architecture and about Big Navi, we simply don’t have enough points of reference to make categorical statements.