The writing is on the wall. AMD now only has a
4.9% market share in servers.
There is demand for low power servers, a new market segment:
http://www.mercurynews.com/business/ci_19240331
IMO, both Intel and AMD dropped the ball by not making Atom and Bobcat for servers and network appliances from day 1. While no single processor will ever be enough for some users, there are many users out there for which it is just wasteful, and both Intel and AMD both treated them like crazies, claiming performance per Watt was so superior in the big servers (forget that they take up several times the space, cost a ton, and you'd need to add expensive high speed interfaces for highly scalable applications; and then you have glorified network appliances, like most basic firewalls, DNS servers, etc., where sub-1U low-power boxes would be nice ways to use rack space).
While it's becoming a new market segment now, that many weaker nodes with their own memory was the best way to scale has been known since the 70s. We had just been getting such speed advancements in big fast CPUs, that nobody halfway mainstream cared until the single-thread performance started dropping off.
Clearly there have been a bunch of issues (manufacturing) resulting in the delays but I'm hoping AMD (and Dirk Meyer) will get the last laught by the time Piledriver and laters CPU's release.
Maybe, but if anyone inside AMD thought that they would be able to do a good job of manufacturing a CPU by its release date, and designed it based on hitting very high speeds early on, that would have been a colossal failure all by itself. Designing it to scale up over time, with minimal changes to do so, OK, but hitting >8GHz at release tells me they were trying to make a speed demon (intentionally weak per clock, to hit higher clocks), like the P4...even today, practically nobody wants one of those.
My question is if AMD cannot do a better design/execution than bulldozer in the CPU segment, which has been their area of "expertise" for the life of the company, do they have the resources/engineering talent/production facilities to get into the tablet/smartphone business and be competitive?? And AMD may be ahead in the low power fusion segment now, but you can bet Intel will not ignore this segment either.
Bobcat/Zacate and Llano went pretty smooth, don't you think? Both were late, but mostly due to genuine production problems, and they got each out in time for it to be successful (FI, Llano needed only to be out by back to school, and widely available by holiday season, where Bobcat needed to out yesterday).
Then, ironically, it appears to be the aspects of BD which were taking such a pounding before its release that are working quite well (CMT int, shared SMT FPU), and the it was the more traditional single-thread performance that really brought it down. They can't realistically can everything BD, but focusing on products of substantially less complexity would probably be a good move.
Like others, I am suspicious that much of what we're reading is code for, "developing markets pay better, and represent less risk."
Facebook
Twitter