Printable View
From everyone's fav site:
http://www.semiaccurate.com/2010/09/...ed-orochi-die/
Odd thing about this shot... look at the top dies, and then look at the bottom ones. If we use the L3 as a bit of a ruler, it looks like the top cores are wider and have less L2...
That's Orochi, the future upgrade to Llano, with integrated GPU.
Bulldozer is the architecture, not the CPU.
this is not technically a bulldozer die shot (it's an architecture as florinmocanu said) as this is not zambezi, but orochi, the new fusion product with some bulldozer core... but anyway, it's a nice picture, bringing some infos...
Since when is Orochi a Fusion Part ??
Orochi is the internal Die/mask name - the Marketing names are Zambezi and Valencia.
Because this was a GF event, it is no wonder, that they used the internal Die/mask name.
However, the visuals are indeed different between top and button. Looks strange but I can see structural similarities between top and bottom parts of the left site, probably just a visual distortion.
Orochi is not fusion.
That's what I thought at first as well, but look at the L3 cache and use it a yardstick. The top dies go noticeably farther on the 'yardstick', How odd...Quote:
Originally Posted by Opteron146
There's another thing. There are small little white-ish rectangular sections that stand out at the top of the top dies, and the bottom of the bottom dies. Three in total for each core, except on the bottom cores the die ends after the 3rd rectangle and turns into cache. On the top dies the core continues for awhile after the third structure.
Some get hans in here!
This is what Charlie says.
Zambezi is the code-name for the normal Desktop BD, Orochi is the code-name for the CPU+GPU Fusion product.Quote:
Orochi is the second gen Fusion part, an 8-core 4-module Bulldozer based chip. Expect it in late late 2011. This will be the part that gets AMD back into the server game.S|A
why does lower part differ from upper one?
http://www.semiaccurate.com/static/u...i_Die_Shot.JPG
can we have photo in thread, dont like to give them hits (petty i now)
Charlie is wrong, or at the very least misspoken.Quote:
Originally Posted by Florinmocanu
Also, just use your eyes. There are four modules, one on each corner, and two cores per module. As Charlie said, this is a 4 core, eight module part. So where exactly is the supposed GPU?
Again, use the L3 as a yardstick. The upper cores obviously extend further up our virtual 'yardstick' than the lower ones do. Besides that, as I said before:Quote:
Originally Posted by MAS
Just look for awhile, you'll see it. Again, look at how far up the L3 each module goes.Quote:
Originally Posted by hurleybird
He's wrong as can be clearly seen from the die shot. There's no GPU on that die, just 8 cores/4 modules.Quote:
Originally Posted by Florinmocanu
It's a visual distortion caused by the picture being taken from an angle.Quote:
Originally Posted by MAS
orochi is 4 module 8 core part without gpu...Quote:
Originally Posted by Florinmocanu
Anyone can see that the upper cores are wider than the lower, and that's not just a picture stretch/visual distortion. You can actually see more parts in the upper cores.
and it looks like theres alot less L2 for the top cores/modules aswellQuote:
Originally Posted by Mats
Hmm, you're right actually. Quite strange.Quote:
Originally Posted by Mats
lower are K10 cores, upper are bulldozer cores :)
Perspective corrected shot:
http://a.imageshack.us/img210/148/orochidieshot1.jpg
The upper two "modules" are definetly larger, but their respective L2 arrays are opposite of that. :shrug:
And they definitely have many visibly similar structures. A couple modules being larger I could write off, but the L2 also being smaller really has me scratching my head.
The upper ones are about 19% larger, now if they were 12% it could mean the upper ones are modules, and the lower ones being cores.
I fail to see the purpose for that ,tho.:shrug:
Where is Hans de Vries?:D
Judging from a quick use of the Photoshop measure tool, the top modules have ~14.5% greater area than the bottom ones. The bottom modules have ~ 26.5% more cache area.
Mats, I think it would be the other way around. Top ones would be full cores, while bottom ones would be modules with shared components to save on die size.
Another though... could be power related. Supposedly AMD has built in the the ability to fully shut off individual modules, however since one module always needs to be on you could save die space by taking that functionality out of one module. In this picture however, half the modules are sized differently...
bulldozer and bobcat on single die :D
Semi accurate does not open for me :(
Here is a roadmap for reference "First bulldozer based cpu :D"
http://www.xbitlabs.com/images/news/..._2008_2011.png
Here is another site with same/similar news
http://www.techeye.net/chips/amd-sho...ks-more-fusion
Right Oliverda, we need Hans !!!Quote:
Originally Posted by Oliverda
^^
I measured again and got 15% difference this time.
It can't be the other way around, the difference is too small for that.
The Orochi disappeared from the roadmaps on the last year's analyst day...Quote:
Originally Posted by ajaidev
http://a.imageshack.us/img841/8342/57270432.jpgQuote:
Originally Posted by fellix_bg
p.s.: Sadly, the core structures are blurred and no crucial detail could be seen.
WHo says it's a 4 module/8 core part? I'd say it's two modules/four cores (one on the left, one on the right).
So Orochi is 4 core if so Id take it each module is per a side(large upper and smaller lower linked = module)
distance between smaller core and shared resources into bigger core is very longQuote:
Originally Posted by slaveondope
I thought so too, but I can't see why they'd put one integer core that far away from the rest of the module.Quote:
Originally Posted by G.Foyle
No. You aren't going to have L3 (or L2 in your theory I suppose) sitting between the cores in a module, they would be more tightly linked than that. All the information we have already points to orochi being 4 modules anyway.
Does anyone else get the feeling that AMD is playing with us/Intel? I mean, showing such a provocative die shot with the important fine details blurred out kinda seems like they are intentionally trying to confuse the hell out us!
Amm am i not seeing 4 L2's "unconnected" is it not true that the cores of a module are suppose to share the L2 not to mention other elements? Why put one part on the module on the left and another on the right?
im getting the feeling this is just showing us what 2x K10 cores would look like with a little more L2 @32nm vs 2 BD modules with less L2
I think the 2 module theory can safely die now. Bobcat + bulldozer theory is interesting, but the modules seem to have too many similar/identical structures.
Perhaps we're actually looking at a six core part. The road-map that was posted earlier:
http://www.xbitlabs.com/images/news/..._2008_2011.png
Shows Orochi not as an eight core part, but a >4 core part. My best guess here is that only the top two are actual "modules". The bottom 2 are single cores, with lots of cache, designed to get lots single threaded performance. Heterogeneous computing.
Hell, it's just as likely something else, but that's the only thing I can think of that makes sense right now. If true, it's hard to believe AMD would let something like that out of the bag.
Could be that this is a Fusion part... but it would look strange for a Fusion part too.
I'm becoming more convinced of my six-core theory. The only part about it that doesn't make sense is why AMD would show their hand like this.
JF already said that they would do some photoshopping to hideQuote:
Originally Posted by fellix_bg
the details. The lower modules look more as what I was expecting.
Regards, Hans
The cores seem to be too big for single bobcat cores but single core bulldozer is a possible assumption with greater L2 cache and much better single/dual thread performance.Quote:
Originally Posted by hurleybird
So in this theory we would have 2+2+1+1 cores = 6 cores
Yeah, the high resolution picture is obviously shopped to hide the details in the core. Are you saying that you think the different sized caches/modules are also shopped? That seems less likely to me, but if that's it I'm sure we could get JF-AMD to confirm it.Quote:
Originally Posted by Hans de Vries
This is the 8core/4module part and the image is heavily photoshoped (that's why the 2 module above look different/bigger).It's just AMD being cautious.
Yep, they've blurred out the core/module areas of the die. See the larger picture few posts above.Quote:
Originally Posted by Hans de Vries
So can you guess what die size would be from that pic?Quote:
Originally Posted by Hans de Vries
Also doesn't it look like top two modules cache seem a little compressed compared to the others?
That is an 8-core Orochi.
And some blurring.
And some photoshopping.
Actual die shots are released at launch, not before. Don't start doing math on this, if you think the "single threaded client performance from a server benchmark" numbers were way off, anything done off of this die shot will be even more incorrect.
So, this isn't an actual die shot? just a mock up? the modules aren't different sized?
Forgive me, but it does seem a bit funny to create a fake die shot, blur out the 'sensitive' parts, and then present it to the world as the real thing at a GF event...
http://i51.tinypic.com/2dujinb.png
It's a fake by AMD. :)
I think that the yellow highlighted parts are the visible corners of the real cores. You can see smaller parts on the top because they put larger fake cores to there.:)
Ahhhhhhhhh:shocked: Quit screwing with our heads man:rofl: :toast:Quote:
Originally Posted by JF-AMD
So are the photos photoshopped? What core is this actually? I did some math based on this photo and figured out that AMD killed my dog.Quote:
Originally Posted by JF-AMD
WHY WON'T YOU JUST GIVE A STRAIGHT ANSWER??? :confused::confused:
I can see the reason why AMD won't show us the die for now, but why showing a fake one instead?
To confuse competition i guess.
It's the real die guys. Just photoshopped - heavily - to obscure the internal structures so you can't figure out what they are if you were say, an enormously sized competitor trying to model your processor in simulation to see how it stacks up months in advance. ^_^
It's the real one that has been altered quite a bit. The structures inside are blurred and the modules are "resized" in order to mask the ratios inside the core.
To keep us talking...
The big joke is probably on the Intel engineers who have been pouting over these photos :P
Do you know the word, "confidential"?:)Quote:
Originally Posted by MrMojoZ
At least we know that the L3 cache is divided into four parts. :D
They learned from Nvidia, I guess? :rolleyes:Quote:
Originally Posted by Mats
http://a.imageshack.us/img266/6173/01192453925l1.jpg
covering the real cores with something else then bluring them makes a lot of sense... can't wait to see the real die shot and some samples :D
It is a real die shot, not a fake. But, to keep the competition from knowing what is really under the covers you change it. This is a pretty common tactic in the industry.
Do we? It can be shopped can it?Quote:
Originally Posted by Oliverda
Well, that makes things less cut and dry. Did you change the *shapes* of the modules and L2 cache or not? that's what everyone wants to know. I mean, if you just blurred and obscured the sensitive stuff it's obviously still legit, but it seems that if you would go as far as changing up the shapes and sized of cache and cores that would make it 'fake'.Quote:
Originally Posted by JF-AMD
Now I remember when Intel revealed Nehalem for the first time, the die-shot they presented was actually a shot from one of the underlying interconnect layers of the chip, not the top level with all the visible logic and cheesy details. I guess it makes sense. ;)
Hans made back then an overlay diagram of that die-shot and he missed by a mile the L2 cache locations/sizes and lots of other stuff.
:rofl: :clap:Quote:
Originally Posted by fellix_bg
So only the cores internals are blured out other than that the die shot is real as in the difference in the size of the above and lower cores are true?Quote:
Originally Posted by JF-AMD
If the cores are "changed", then it is fake, not real because the real die is different.Quote:
Originally Posted by JF-AMD
If you mean only blurring or scaling by "change", then ok.
However, because the top and bottom modules are of different size, it seems you mean the copying/pasting some parts.
OMFG it's huge!111 It's like 5" x 6" Where's Hans? We have to calculate IPC/mm^3/£/mm^2 now!
No wonder they couldn't use the AM3 socket... now lets keep on going like this for 30 pages until MM stops us blablaah..
Quote:
Originally Posted by fellix_bg
We do. The center of the die has not been "shopped".;)Quote:
Originally Posted by Mechanical Man
Quote:
Originally Posted by Mats
dont forget the tdp :evilvoicelaugh:
Is it not? How can u know?Quote:
Originally Posted by Oliverda
Heh, I knew I'd forgotten something...Quote:
Originally Posted by Sn0wm@n
Do you mean TDP or TDP/MB L2 cache?
Who am I?
''From that die shot I can tell that the IPC/TDP/mm2/frequency/L2/L3/socket pins is definitely much lower than Phenom II, plus the '32nm Orochi Die' description is clear evidence that Bulldozer is DELAYED by 3 quarters and will launch in Q3 2012. It's all there.''
:clap::rofl:Quote:
Originally Posted by Dimitriman
Quote:
Originally Posted by Mats
tdp as in power draw ...
by the amount of post bulldozer would perform much like an old 80's cpu ....Quote:
Originally Posted by informal
how sad,he's not even here,yet he still controls AMD threads:clap:
this thread is strong with the fanboi-force...
Quote:
Originally Posted by 570091D
:lsfight: :D
Fixed. :) :up: n (TBH -- I don't really see the fanboy crappola in this thread like we saw in the other BD thread).Quote:
Originally Posted by 570091D
this is funnyQuote:
Originally Posted by Dimitriman
i'd lol
AMD vs. Nvidia. Or internal fight between AMD vs. ATI.Quote:
Originally Posted by Sn0wm@n
Note :
This is:
http://i624.photobucket.com/albums/t...uoihanoi-1.jpg
Isn't it possible that this is only a part of a full die?
A full die shot has on all the sides a blue ribbon that states the division between the processors on a wafer for the cutting (I think). This die shot has only two sides with a ribben (top and right) seems to me that this is not the full die?
Compare that to a full die shot of istanbul.
Still doesn't explain the reasen why the different pieces are not the same size.
Quote:
Originally Posted by Mechanical Man
I have two good eyes.;)
Looks like AMD photochop goons messed up not only the CPU cores, but a part of IMC/NB area:
http://a.imageshack.us/img521/2912/65142382.jpg
The structures, highlighted in the green rectangle, have been clone-stamped over and over again with patterns from the IMC/NB to the right side, to fill out (completely conceal) something equally important to the cores.
I can only speculate, that this is an L3 directory cache and routing logic, if we are to believe the four L3 partitions are private to each "module".
Nice find, interesting that they conceal such a large part of the die. (edit: IT'S THE NINTH CORE! :O)Quote:
Originally Posted by fellix_bg
I wonder why there is so much "wasted" die space, so much brown areas that don't seem to do much at all. Is it all decaps?
:rofl::rofl::rofl::up:Quote:
Originally Posted by fellix_bg
Wait wait...no hexa core for bulldozzer?
A.) that is a full 8-core dieQuote:
Originally Posted by meyns
B. Please read the thread for all of your other answers.
No native hexacore, but AMD might disable one of the modules you see in the dieshot to make one.Quote:
Originally Posted by SubZero.it
or make a 3 module hexacore .... anyway anything is possible but one thing is for sure ... 32nm and 22nm are going to be epic on the amd side
Or two different cores in two different modules...Quote:
Originally Posted by -Boris-
A native hexacore is very unlikely, it would need a total redesign of the ship to put three modules in line. Probably better to just disable 4-module chips.Quote:
Originally Posted by Sn0wm@n
I doubt that is possible, the cores share to much parts within a module. You would only turn of the integers, all other parts still has to be activated.Quote:
Originally Posted by Mechanical Man
Check this two roadmaps :
http://www.xbitlabs.com/images/news/..._2008_2011.png
http://img13.abload.de/img/amd_deskt...map_bil0sg.jpg
I think AMD choosen to Show us the die of Orochi. A BD core seem to be canceled, i would say BD1.
It's clear the second roadmap is less old, with thuban and AM3r2 for BD.
So i think Orochi was BD1, Zambezi is BD2, the real BD incoming.
And lynx maintream have a GPU on die, it's not for only low end. !!!
Good heavens!Quote:
Originally Posted by hurleybird
If you stare hard and long enough, you will discover a McDonald's BigMac nestled next to the L1D structure.
No. IIIRC Orochi(the name of the chip family) is Bulldozer(code name for the whole next gen core design) that was always slated for 2011.The "BD 1" was a 45nm version that was never on the 2011 roadmap. Zambezi is the code name for the desktop parts in the Orochi group.Quote:
Originally Posted by madcho
The new design is fully modular(M-Space), in theory they should be able to swap diffrent core types/GPU cores round in almost any combination. Think the first few with BD cpu's will only have BD cores in and they will save the mixing and matching till after that's out the door. Looks like they have 4 modules slots to play with at 28nm. If the GPU modules turn out good, they could put 2 of those into those slots and have an amazing all in 1 chip for gamers. Extreme gamers will still be using the tri-xfire/sli setups tho :)
If orochi was "BD1", why are they showing it on the 32nm node, when BD1 was 45nm?Quote:
Originally Posted by madcho
Surely this is nothing more than a 32nm DB, with photoshopping and pixelating to stop those evil and dastardly Intel spies?
cant wait till amd shows us some numbers :D