vacancies advertise contact news tip The Vault
EPIC Competition in the run up to 2015 - win something everyday! [x]
facebook rss twitter

AMD revises Bulldozer down from 2.0 to 1.2 billion transistors

by Alistair Lowe on 6 December 2011, 09:28

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qabacw

Add to My Vault: x

What does this mean?

When AMD's Bulldozer four-module, eight-core architecture was first announced, the provided figures suggested that the FX-8150 was constructed of two billion transistors. AMD has since revised this figure down to a 'meagre' 1.2 billion transistors.

Well, for starters it means that the Bulldozer architecture isn't a lost cause when it comes to design efficiency. With the previous two-billion figure, the FX-8150 would have had almost double the transistor count of Intel's hex-core Core i7 990X, even though it had struggled to hold its own when benchmarked against the Intel competition. A figure of 1.2bn suggests that AMD did indeed manage to cut down on core logic and successfully share processing components in the new design, to good effect, as promised.

On the other hand this raises several questions. Had AMD been using a wasteful, heavily-automated system for layout of its architecture? If so are two billion transistors actually present on the die but simply not all in use? Given the die area this seems unlikely but possible, or perhaps AMD had initially intended to have more on-die cache? Was this related to manufacturing troubles at GlobalFoundries? Does all this mean that there's a good level of wiggle room for future optimisation of the architecture?

So many questions, so few answers, as AMD's official response to the matter remains that the firm was simply correcting a mistake and that the previous 2.0 figure had been a rough early estimate. 800m transistors: that's some mistake.



HEXUS Forums :: 10 Comments

Login with Forum Account

Don't have an account? Register today!
1.2B logic to 2B on-die seems unlikely. 2B on-die would mean a similar transistor density to Llano, but the GPU on Llano has a lot to do with that. I think I read somewhere it has something to do with dual-die Opteron i.e. it has 2B transistors in use, maybe due to disabled memory controllers or something? Please don't quote that though, I'm unsure of the validity of the source.

What does it mean?
Well, probably nothing really...
Hmmm....that's some variance they work within! :p
considering the levels of precision involved in making a processor, you would think a 40% error would be something they should have noticed a long time ago
This news was mentioned a few weeks ago by AMD but for some reason the media has picked it up only now??

It could be like with Sandy Bridge which has two figures;1.16 billion transistors and 995 million transistors. The 1.2 billion number is probably the number of active transistor units;AFAIK certain units are made up of multiple transistors and certain transistors serve no electrical purpose,ie, used to separate certain parts of the chip or serve as redundant units.

If you look at the die it seems there is a reasonable amount of free space:

http://pc.watch.impress.co.jp/img/pcw/docs/430/044/5.jpg [pc.watch.impress.co.jp]

Some people have speculated that it separates key components and hence minimises heat hotspots.

However,TBH,what is more important though is performance for a given price.
I've read elsewhere that AMD were not aware of the mistake, it was the PR companies fault and as a result that PR company no longer works for AMD.

Also in the news from another website is that AMD are apparently considering leaving the desktop CPU market as they can't compete or make money from their operations - they've had enough of running in the RED on their banks statements.