When you design any SoC or chip, you have some targets with frequency, performance and power that needs to be met in order to mass produce the chip for reasonable costs and volumes. This sensitive math is an oversimplification of several years of work.
The Vega chip actually looked much more competitive in the development phase compared to the product launch. One of the reasons why was that some key power optimization engineers were pulled from the Vega team to the server Naples team.
Power gurus went to EPYC
The server, chip now known as EPYC had more priority for long-term performance and the survival of AMD and we definitely agree with this assessment. But since AMD doesn’t have a lot of talented engineers sitting around, moving some people from one group to another could affect the end result.
AMD knew that gaining any market share in server would look good with investors and stock holders and had to make a very promising chip. Even today, almost a year after the EPYC launch, AMD is having limited success, but it did manage to win some key customers and continues to grow.
Taking key talents from GPU and moving them to the server business meant that EPYC has some impressive power to performance ratios but at the same time Vega didn’t. Our well informed sources didn’t want to reveal the names but there are not that many power optimization magicians working in these chip companies, it comes down to a few people. A great leader with a decent team can do a great job.
We cannot blame it all only on that decision but at the shipping clocks that were safe to mass produce for Vega 64, the chip has a higher TDP and power than planned and predicted. At least now it makes a bit more sense.