But the large Vega GPUs are probably on the edge of what is doable on GF 14NM and are pushed to the edge. HBM interfaces on-die are much smaller and efficient compared to GDDR type memory. Look at Polaris or Navi performance/watt when pushed past the optimal voltage-frequency curve?? The APUs are in the ideal parts of the curve,and mobile ones even moreso than the desktop equivalents.
Now think if Intel 10NM cannot clock that high,and Intel to remain competitive is pushing the parts to the edge too?
ATM,GF 14NM and TSMC 7NM are known quantities,but Intel 10NM has had problems for years,and the 10NM we are finally getting is not the original 10NM envisioned,but one with relaxed parameters.
To put this in context,Intel is still only making quad cores on 10NM which all have lower clockspeeds compared to their 14NM equivalents,which clock higher and have double the cores! IMHO,this has all to do with 10NM being a bit crap.
This is why they are using Atom cores for low load functions,as these are a known quantity,will be clocked lower and take up little space compared to the bigger cores. Intel already is making 10NM Atom cores right now. So the larger cores on 10NM have some problems it appears. If they were not why would Intel not be having 16 high performance cores??
AMD currently had 16 high performance cores,and good performance/watt without all this milarky in 2019. Also,what is the likelihood Alder Lake is a monolithic die,which adds to the problems of yields,binning,etc??


LinkBack URL
About LinkBacks
Reply With Quote