AMD's Bulldozer goes on celebrity diet

AMD’s Bulldozer appears to be taking its turn in the spotlight seriously and has been shedding those love handles.

When the AMD press officer started flogging Bulldozer the beast was a chunky eight core / four module CPU with a hefty two billion transistors.

Well, it seems in just a few days of crash dieting, and perhaps the odd day or so in the gym, Bulldozer has managed to slim down to a tight 1.2 billion transistor CPU.

AMD insists that it has not had to resort to any crash diet pills or anything that is not good for Bulldozer. In fact the chip’s two billion transistors did not seem to be doing much other than an unsightly overhang in the middle range. A chip that fat should have been on a par with Llano, but when it came to walking up a flight of stairs, or even a jog against anything from Chipzilla the Bulldozer was puffing under the weight.

But as Extreme Tech  points out, if two billion transistors seemed awfully fat, 1.2 billion is pretty darn thin and that can’t be good.

“The only way to make AMD’s figures even provisionally fit is to assume the company shrank Bulldozer’s core from 91 million to 60 million transistors,” said Mr Tech who sees that as cheating on the diet.

AMD, for clarification, insisted that it was simply correcting a mistake and that the Bulldozer was never really that fat in the first place.

Of course there is talk that AMD is trying to gloss over Bulldozer’s problems by giving it a morale boost and claiming that it has been dieting when all it had really been doing is spinning the number of functional transistors as opposed to the actual number laid down on the chip.

There is no proof that it would do such a thing, and it is much better to think that the chip has been exercising away those access transistors.

Currently the word on the street is that Bulldozer’s 1.2 billion transistor count is somewhat lower than it should be, but closer to what everyone expected when the chipmaker moved from 45nm to 32nm.

It means that Bulldozer did deliver the transistor savings and die area reductions that were expected and most of its problems are caused by a flabby cache performance. There are few exercises that anyone can do to remove that.