Nvidia RTX 20 Turing GPU expectations

I definitely agree. I find explaining Su's strategy is easiest to understand what is going on at AMD.

First, Su took the money used to stay competitive on GPUs (even though losing) and moved that to CPU manufacturing and development of Zen. She practically chose to abandon the GPU high end side to focus solely on correcting the ship on the CPU core business at AMD. That is fine, and she seems to have succeeded in spades.

Second, because Raja did not like the budget being slashed, he convinced Su to spin off RTG, that way he had more autonomy. What he did not like or realize was Su would pull 2/3ds of the engineers to work on an embedded, low power solution for Sony (Navi), leaving Raja with 1/3d the engineers for Vega (which was delayed, in part, for Apple, as that GPU was made with Apple to a degree, something that came out after its release). So, Su is getting other companies to help pay for development to keep that side afloat while going full bulldog attack mode on the CPU side.

Third, after refinancing debt, AMD has sunk significant amounts into R&D and die shrinks. It sees staying on the bleeding edge node as essential for it to keep and grow market share. Considering AMD's projections have been conservative, and Intel is saying its job is to hold AMD to 15-20%, Intel recognizes its own problems and recognizes AMDs aggressive moves. The 14nm capacity shortage was not factored into AMD's 4.5-5% server market share, most likely, when Su gave those numbers during Q2 earnings. If not factored in, AMD could exceed projections for the year, and with 7nm chips being drop in for those servers, AMD could hold onto that market share for a period of years (up through 1-2yrs past Zen 3 in 2020, the last supported year for the socket, etc.). HPE suggesting AMD servers to consumers with the Xeon shortage will only help grow this, especially as the 12-18 month test beds for Zen based chips finish up at server farms and more companies are publicly announcing Epyc deployments. That is only compounded by the security flaws coming out, many effecting Intel and not AMD, although some effect both. So, the investment to hit 7nm, meant to go against Intel's 10nm, which on servers is now coming a year after AMDs 7nm offerings, is significant for AMD to do a market grab while having a short lived process lead (whether the gutted 10nm Intel process will outperform 7nm DUV or 7nm EUV has yet to be seen, but many have started to say Intel is behind on 10nm node vs 7nm from other fabs, and that their 7nm, which is 2021 or later, which coincides with Intel's adoption of EUV after a statement last week, is the earliest that one could look at Intel getting a process lead again).

So, the question becomes, with what is known from them using Sony to develop Navi, and with the windfall from the server market grab, plus the design of super-SIMD and abandoning of GCN, when they can execute what as far as architecture, and whether they can pull themselves back from the mindshare drain from them abandoning the GPU market, where they were still losing, but keeping head above water, in favor of the CPU market. That I do not know. But, this is why I put a dark couple years on the GPU market, overall.

 

Back on topic, though, GN discussing the changes and Hardware Unboxed: