AMD's Bobcat and Bulldozer

[onethreehill]

http://www.lostcircuits.com/mambo//i...id=91&Itemid=1
http://www.hardocp.com/article/2010/...essors_preview
http://www.pcper.com/article.php?aid=985
http://www.anandtech.com/show/3863/a...hot-chips-2010
http://www.extremetech.com/article2/...2368185,00.asp
http://www.techpowerup.com/129392/AM...hitecture.html
http://techreport.com/articles.x/19514
http://www.hardwaresecrets.com/artic...tecture/1078/1

[96redformula]

When is the actual release date supposed to be?

[Kuroimaho]

When is the actual release date supposed to be?

“Due out in the first half of 2009, AMD's Bulldozer core is the true revolutionary successor to the K8 architecture. While Barcelona and Shanghai are both evolutionary improvements to the current core, Bulldozer is the first ground-up redesign since the K7.”

The current plan is next summer but we know how did it work out last time...

[Opteron146]

@anandtech:

MMX

This should be at least XOP .. .makes no sense otherwise, MMX is dead as dodo.

[Chrono Detector]

We need a release date and specs.

[madcho]

nothing more than was said by dresdenboy yet

edit: only the AM3+ socket was confirmed.

[superrugal]

nothing more than was said by dresdenboy yet

edit: only the AM3+ socket was confirmed.

how about this？

http://www.anandtech.com/show/3863/a...t-chips-2010/5

[radaja]

nice
exciting time ahead

[-Boris-]

Lostcircuits seems to believe that Bulldozer will be a bobcat with an extra core. They mention store, load and two instruction pipes. Guess it's reasonable, but not too interesting.
They're talking about HT in Bobcat too. Seems like something I'll investigate closer. But I think they misunderstood something.

EDIT:
Anandtechs preview is much more in line with my expectations. Seems like Dresdenboy was right when he first talked about two AGUs and two ALUs.
But seems like the performance per core will be significantly lower than sandy bridge, but at the same time seems like they made really small modules. I think the most fair comparison will be one BD module vs. one SB core.

Still wonder about L2 cache. AMD has way more inefficient L2 than Intel at this point.

PCperspective seems to think it's 4 ALUs after all. I Trust Anandtech more.

[JF-AMD]

Lostcircuits seems to believe that Bulldozer will be a bobcat with an extra core.

No, way off. They are totally different architectures.

[madcho]

2 alu + 2 agu / core.

WTFFFFFF

[gallag]

I had a bad feeling about the last few AMD cpu launches and I was right, I have a good feeling about bulldozer though, It looks more and more like it will rock. Also, Being late to the party will not mater if it does rock, Better to wait and get it right imo.

[G.Foyle]

nothing

[ajaidev]

AMD also told us that it will introduce a new AM3+ socket for consumer versions of Bulldozer CPUs. AM2 and AM3 processors will work in the AM3+ socket, but Bulldozer chips will not work in non-AM3+ motherboards.

http://www.extremetech.com/article2/...2368186,00.asp

waaa waa

[Opteron146]

http://www.extremetech.com/article2/...2368186,00.asp

waaa waa

Forget that comment, they just proofed that they dont have any clue.

AM2 CPUs cannot work in AM3, however they should now be able to work in AM3+ ???

Sorry but:



They definitely understood something very, very wrong.

[ajaidev]

Forget that comment, they just proofed that they dont have any clue.

AM2 CPUs cannot work in AM3, however they should now be able to work in AM3+ ???

Sorry but:



They definitely understood something very, very wrong.

lol did not notice that was more concerned over the Bulldozer not working with non- AM3+ parts.

Agree that's not good for a AM3 actual board.

But C32 and G34 will be working.

i don't understand fully this part in anandtech review :



Why a 2 Alu core could be faster than a 3 ?

WTF if there is not 4 alu and 3 agu per core, where is integer performance ?

I'm afraid about this too :



Same clockspeed + deeper pipeline.

AMD says that 3rd ALU was not much used in PhII and it was left so as not to redesign that part. But yes there will be a performance dip due to this in single threaded ALU dependent app's. Lets hope the other enhancements actually work to cover up this missing ALU.

Branch Prediction is said to be aggressive and it has to be with as deep a pipeline bulldozer has otherwise we can have a call problem. On the bright side deeper pipes may enable better overclocking well that depends a lot on the architecture.

[Opteron146]

lol did not notice that was more concerned over the Bulldozer not working with non- AM3+ parts.

Yes I can understand your surprise, I was first shocked, too.

However it does not make sense at all.

Since the introduction of Socket AM3 with 941 pins and CPUs in an AM3 package with 938 pins only (because of socket AM2 backward compatibility), I was waiting for a CPU with a "real" AM3 package with the full 941pins, which wont be AM2 compatible.

I think that is it what AM3+ really is about ... plus some faster memory modes.

Just horrible to see how bad some internet sites are ... now the uninformed people who believe that piece of <censored> misinformation will pop up in lots of forums, and it will be linked everywhere, what a pain in the <...>

[madcho]

http://www.extremetech.com/article2/...2368186,00.asp

waaa waa

Agree that's not good for a AM3 actual board.

But C32 and G34 will be working.

i don't understand fully this part in anandtech review :

The 3rd ALU does have some performance benefits, and AMD canned it to reduce die size, but AMD mentioned that the 4-wide front end, fusion and other enhancements more than make up for this reduction. In other words, while there’s fewer single thread integer execution resources in Bulldozer than Phenom II, single threaded integer performance should still be higher.

Why a 2 Alu core could be faster than a 3 ?

WTF if there is not 4 alu and 3 agu per core, where is integer performance ?

I'm afraid about this too :

Branch Prediction and a Deeper Pipeline

Bulldozer will use a deeper pipeline with less logic per stage compared to current Phenom II/Opteron processors. AMD argues that this will ensure clock speed won’t be a problem with the design and we should expect to see Bulldozer based products at similar if not higher clock speeds than what we have today with Phenom II.

Same clockspeed + deeper pipeline.

[-Boris-]

There's a Bobcat die floor plan on the slides, anyone care about estimating the die size? There's 512 kB of L2 cache consuming 31% of the total chip area. If Bobcat was made on 45nm process and the cache was the same as L2 in Deneb/Thuban cores, it would make Bobcat core be around 13 mm² (core with L2 cache only - gpu engine and memory controller not included).

Were is that floor plan?

Forget that comment, they just proofed that they dont have any clue.

AM2 CPUs cannot work in AM3, however they should now be able to work in AM3+ ???

Sorry but:



They definitely understood something very, very wrong.

The AM2 part is probably a typo, or they are counting Phenom II as "AM2 and AM3".

Agree that's not good for a AM3 actual board.

Why a 2 Alu core could be faster than a 3 ?

What I wonder is how that could save that much die space that they feel it's a good move.

Same clockspeed + deeper pipeline.

Clockspeed is something Anandtech was guessing. We know nothing about that. And we don't know how much deeper the pipeline is.

[Opteron146]

The AM2 part is probably a typo, or they are counting Phenom II as "AM2 and AM3".

Or they dont have a clue ^^
I explained it above what AM3+ probably is about .. only a new package, not a new socket (besides higher RAM clocks).

[-Boris-]

Or they dont have a clue ^^
I explained it above what AM3+ probably is about .. only a new package, not a new socket (besides higher RAM clocks).

You don't need to tell me what AM3+ is about.
I think it's very unlikely that their knowledge about AM2/+ and AM3 is so limited that they honestly believe that an AM2 processor will work in AM3+.

[ajaidev]

Or they dont have a clue ^^
I explained it above what AM3+ probably is about .. only a new package, not a new socket (besides higher RAM clocks).

It must be something like that otherwise the AM3+ should be called AM4. More info is needed on the bulldozer and its socket, i did not see anything about it on anandtech.

Its very sad if bulldozer is not compatible with simple AM3 socket because it will force people to buy new mobos and force them to choose between Sandy bridge and Bulldozer.

[mAJORD]

Agree that's not good for a AM3 actual board.

But C32 and G34 will be working.

i don't understand fully this part in anandtech review :



Why a 2 Alu core could be faster than a 3 ?




WTF if there is not 4 alu and 3 agu per core, where is integer performance ?

I'm afraid about this too :

Core 2 through nehalem (not sure about SB) has 3 ALU , 2 AGu, does it have less performance than K10? At this point, without more details I think it's hard to say if this is really an 'issue' or not. for single threaded performance

[fellix_bg]

WTF if there is not 4 alu and 3 agu per core, where is integer performance ?

I'm afraid about this too :


Core 2 through nehalem (not sure about SB) has 3 ALU , 2 AGu, does it have less performance than K10? At this point, without more details I think it's hard to say if this is really an 'issue' or not. for single threaded performance

K8 and K10 arch could never reach the full potential of the 3-issue pipeline due to fundamental limitations. Bulldozer is simply streamlining the concept and optimizing other parts to gain efficiency -- i.e. more with less, relatively speaking.

[mAJORD]

K8 and K10 arch could never reach the full potential of the 3-issue pipeline due to fundamental limitations. Bulldozer is simply streamlining the concept and optimizing other parts to gain efficiency -- i.e. more with less, relatively speaking.

Sorry, that was not me who said that , I misplaced the quote tag.. Fixed, and I agree