The upcoming Geforce GPU architecture will be more incremental than revolutionary as Nvidia appears to be using Intel?s tick-tock strategy.
This incremental comment will all make sense after we give you a few more details to sort things out. We still don?t dare to call the next generation Geforce a Volta. It is becoming apparent that Volta is for Tensor flow performance and Artificial Intelligence/Machine Learning, not gaming. Now Nvidia has two focuses, AI / ML and gaming. Not only does this mean that the outfit has opened new markets it means it is now so old it needs special glasses as it tries to focus on more things.
The dedicated follower of Pascal is now "a Pascal influenced design derived shrink down". With Maxwell architecture, Nvidia chose to optimise the GPU for better memory management. Back then, it became apparent to Nvidia that HBM and HBM 2 would not become mainstream in time for Pascal. Fudzilla already mentioned the fact that Pascal follower will not use HBM 2 memory.
Now, this looks like we were right as Volta-based V100 uses HBM 2 and Pascal fans will still stick with GDDR5X memory. Pascal-based GPUs will use the GDDR5X memory. That is ok; it works even at the highest possible settings.
One more thing we can expect from the Pascal follow up is 12 nm FinFET manufacturing. The 12 nm is likely because the TSMC process is more like an optimisation of the 16nm FinFET node than anything new. The concept that 12 nm is original can be attributed to the marketing skills of TSMC as in the eyes of the population, 12nm sounds much better than 16nm.
TSMCs 12nm FinFET process is an upgrade as the company time to optimise the manufacturing and process.The new Geforce will have a few new tricks. It will be what Intel would call a tock, while Pascal was the real shrink to 16nm FinFET is more of a tick.
Reply With Quote
), Juan J. Guerrero
Originally Posted by SinOfLiberty
.
Bookmarks