In a rather unexpected turn of events, it appears that NVIDIA’s upcoming consumer Ampere GPUs, the RTX 3000 series will be based on Samsung’s 8nm node, rather than TSMC’s 7nm. This is important news because the former is essentially a more optimized 10nm node from Samsung which is notably less dense compared to TSMC’s N7 process.
Samsung must have signed a really lucrative deal with NVIDIA as it’ll be the first major chipmaker to make use of the Foundry’s 8nm LPP process. Keep in mind that this node is manufactured without the use of EUVL (Extreme ultraviolet lithography) and features a much lower transistor count than TSMC’s 7nm – it’s basically an extension of Samsung’s 10nm LPP.:
|Samsung 8 nm||TSMC 7nm|
|Fin Pitch||42 nm||30 nm|
|Gate Pitch||64 nm||57 nm|
|M1, Mx||44 nm||40 nm|
Some more info about the RTX 30 series GPUs has surfaced. You can read that here:
NVIDIA RTX 3080 Ti Rumored to Feature 5,248 Cores and 12GB GDDR6X VRAM
This would kind of explain why NVIDIA’s top-end GA102 GPUs have been rumored to feature a TDP of 300W+. More importantly, this means that AMD will retain the process leadership over both NVIDIA and Intel in the consumer market:
NVIDIA RTX 3080, 3080 Ti, 3090 (Titan) w/ 10GB, 12GB & 24GB GDDR6X Memory Based on GA102 Die Likely
NVIDIA RTX 3080 Reportedly 30% Faster than the GeForce RTX 2080 Ti: Early Benchmarks
At the same time though, it shouldn’t prevent NVIDIA from introducing some massive performance upgrades with its RTX 30 series “Ampere” lineup. The company did it with Maxwell and again with Pascal, and I’m quite confident they’ll do with once again with Ampere. Regardless, that means the higher-end GPU is about to heat up, with both NVIDIA and AMD priming their big GPU dies for launch in the coming months.