Soldato
Very good points but 1 bad wrong point!
AMD doe not do that anymore?
Nope! Unfortunately AMD still doing this and never learnt from their mistakes. AMD probably sounded a bit like Apple, they desperately wanted to tried to be like Apple to sell "revolutionary" and "dominate" badly and also always "denial" and "defended".
AMD has long catalogue of failures and disasters, the problem began after acquisition of ATI, both AMD and ATI attempted to created APU codenamed Fusion in 2006 ended in massive failure few years afterward but they didn't stopped and refused to gave up so they invested millions really desperately wanted APU project to worked succeed. They tried again with project Desna but it ended in failure once again with little success as billions went down the toilet, shocked they still refused to abandoned risky APU project and continued to invested millions in Llano project then few prototype samples finally worked perfect after 5 years of disasters. AMD was like Roy Taylor mind thinker very confidence in Llano launch believed it would be the most successful product in AMD history and would massively outsold Intel and had shareholders fooled by AMD lies as there was really no demand and customers was really not interested in APUs.
Don't forget others issues like Bulldozer, Barcelona TLB bug, Dresden fab etc.
Apple had it own share of issues too, they had worked test prototype samples of revolutionary sapphire glass in R&D few years ago believed they had technology ready to mass produced sapphire glass. They invested $578 million in GT Advanced Technologies back in 2013 wanted hundreds of millions sapphire glass installed in 2014 products but they experienced massive yield failure with millions failed to met quality standard then Apple abandoned plan and switched back to LCD screen, they realised sapphire glass are very complex and complicated process to mass produced.
Nvidia said HBM vertically stacking is the best approach because it simpler process, they didn't wanted to do HBM interposer stacking because it was very complex and very complicated process, however AMD argued HBM interposer stacking is the best approach so they decided to went with it and regret it struggled very complex and complicated production issue they experienced now.
If Nvidia managed to get all pieces together and worked in pilot production and have no issue with volume production shipped millions of Pascal cards then that would be absolutely major miracle Nvidia had chose the right path. Jen-Hsun Huang would be very proud of his teams of employees, professors, scientists and engineers because they are the brightest and smartest minds worked like Einstein outsmarted AMD engineers' minds worked like Roy Taylor mind gave AMD management, exectives and board of directors poor judgements but this issue still haunted AMD today since Ruiz era board of directors not had the brightest minds gave poor judgements on ATI acquistion, APU Fusion project, Bulldozer etc.
