Preliminary Bulldozer and Llano Pricing Revealed

[2good4you]

I have a hard time believing that because AMD is pricing its fastest BD (at launch)at 320 so there is no way they would sell it that cheap if it was faster than intels chips.

I don't believe the prices. At least, the performance we think it'll deliver does not match the price (or AMD is very friendly to us deliver new super cpu's for good price). Because of what I wrote above, it must really shine in singelthreaded workloads, and also be great in multithreaded workloads. We have so many powerful news in the BD design, and what we can see, all the weaknesses from K7/K8/K10 is gone. The first thing I see is branch fusion and the massive front end. AMD had always earlier very powerful execution engines, but bad utilized. Also, we don't have to forget:

according to AMD, Bulldozer is designed to be nothing less than "the highest performing single and multi-threaded compute core in history".

back in the A64 days when AMD was faster than intel they put out 800-1000 FX chips becasue they had the performance lead.

also a note

the sandy extremes are only 6 core chips

I hope you are right because that will drop intels prices big time.

The 6-core Intel's have SMT and theoretically gets more performance than 8 core BD. But we'll wait and see.

Many things we don't know yet. One thing is for sure, and it is that BD will be quite cheap to produce relative to it's performance thanks to it's high-density-design I think.

If AMD Bulldozer still can't beat Intel's current 6 core offerings that will be indeed embarrassing. Why can't they use hyperthreading like Intel does?

SMT is a great feature, and is proven in heavy parallell server workloads. One great example is the IBM Power 7 with 32 cores and 128 threads. Anyway SMT/CMP becomes more and more important in personal computing also, and the design of AMD's multithreading is a great combo of SMT and CMP and creates the best solution of both worlds.

Intel's SMT (HT) uses about 5.5% extra transistors in the cpu and delivers at most 25-30%. Also it can perform worse sometimes because of the nature of SMT. A CMP or AMD's design will never perform worse because of a branch misprediction. AMD's design will use 12% more transistors and perform 80%, which means 160% from one core/module. If Bulldozer can perform this good with the shared front-end, then we have a new better chip-design for parallelism/thoughtput.