Tom's Hardware Forums » CPU & Components » CPUs » Rev G die pic might have given away AMD's Emergency Edition
 

Rev G die pic might have given away AMD's Emergency Edition




Word :   Username :  
 
 Page : 1 2 3
Next
Author
 Thread : Rev G die pic might have given away AMD's Emergency Edition
 
Profile: newbie
More Information

Last message on previous page:
What an argument. All reasoning. No facts. No certainty.

Quote :

right?


Quote :

doesn't explicitly say


Quote :

...(rumor)... right?


Quote :

So Ruiz...



I can infer a lot of things from what I didn't learn from not reading to.

Related Product

Register or log in to remove.

Profile: member
More Information

Uh.. the "right?"s were indications that those were questions to you. 8O

First "right" = Has anyone from AMD officially confirmed bulldozer plans?

Second "right" = Well, not much meaning here, just argument. :D

Third "right" = You seem to be an expert on bulldozer. It was a question to you: Isn't it just a mobile architecture?

Profile: member
More Information

And another thing: Your link that says K8L is for mid-2007 is a news report, quoting someone from Technology Analyst Day. Dated June 1.

My link was to an interview with the top person at AMD. Dated June 5.

And, Ruiz talked about "the architecture we're going to introduce next".

I am not saying I am absolutely sure K8L will come only in 2008. I hope it comes earlier.

Profile: old hand
More Information

Quote :

Joset - I want to respond in more detail after I read your post and digest it, but on point on the dieshot compare you put up....

The right hand die is accurate, the die area is 199 mm^2, the AM2 revision I think I read pushed that to 220 mm^2 with the new IMC, at least that is what the Rev F is showing.

However we can back caclulate and get a better estimate of a lower limit die size of the 65 nm K8L with a little effort.

By just eyeballing the die on the right, one core and 1024 K L2 cache, the cache takes up about 1/2 the die, so at 90 nm the logic area for the Rev F core so 1 core is about 110 mm^2, of that about 55 mm^2 is logic. These are just eyeballing easy calcuation numbers.

Looking at the K8L quad core, the L2 cache is roughly 1/3 of the die. At a 65 nm shrink one core would be ~ 27 mm^2 for the logic alone, add 33% due to 512 K of L2 cache is 9 mm^2, so one core of the K8L layout is roughly 36 mm^2. Multiply this by 4 and you get 144 mm^2 for 4 cores, then again eyeballing maybe another 10% so a K8L is looking to fall around 158-160 mm^2 -- this is respectable and a healthy die size. I believe the picture quotes 150 mm^2.

The problem is that of the die, and again this is an estimate then of the total die 67% is logic as opposed to 50%. This will make the K8L slightly harder to yield as defects in SRAM are "repairable", but a defect on logical units is not. That is the point I was driving to.



Yes, I arrived at similar numbers, roughly.
If this image from www.chip-architect.com is any worth, the relative sizes would appear like this:

http://images3.pictiger.com/thumbs/bd/3a255019f8a62b68c2399cca783d32bd.th.jpg

(See my previous post, on this matter).
Even with halved L2 cache per core (despite a higher logic area) and a [supposed] shrink node and the added on-die 2MB L3, this seems quite a small die for a quad-core (compared to a 65nm, dual-core 4MB L2 Conroe/Woodcrest).

I agree on your yield/defect remark, although the increase in logic area between the two node cores would amount to roughly 17% (significant, still), although such a [supposed] small area would also be less prone to external origin causes, like cosmic-rays, for instance.

Still, I wouldn't wonder away from a speculative mockup or CAD simulation preview. My opinion.


Cheers!

Profile: addict
More Information

Quote :



Intel was a sorry investment for the last two years and is looking to hold onto a mere duopoly. Intel's management is having to get used to not being able to bully their customers. Intel has a long way to go at the top. They have failed acquisitions, failed architectures, and failed business models behind them. That's why I will never go long Intel.




Try not to get too emotionally attached to AMD. I realize you are only 20, but have no doubt read about how AMD has been a dissapointment to investors for many years.

Try to keep an open mind about intels future prospects and who knows, you could end up making a ton on intel on the long side as well.

Profile: Eternal Poster
More Information

Quote :

Try to keep an open mind about intels future prospects and who knows, you could end up making a ton on intel on the long side as well.



Keeping an open mind about any person, place or thing is good advice - even Intel! I may take that advice and buy another Antec PS some day. Like after Alzheimers kicks in and I can't remember that the POS cooked my mobo and barbequed my CPU.

Zzzz...

What were we talking about?

Profile: Forum Resident
More Information

Quote :

Take a closer look. Rev G is not K-8. It's not K-8l. Looks to be badassed. AMD has 65nm working. Check the analyst day slides. They have the most efficient fab in the world to. At the presentation, they had 65nm machines running outside. AMD is getting the most out of the vanilla k-8. It's called maximizing ROI (Return on Investment).



I'd like to counterpoint this post... This is why I don't think things are looking to rosy for AMD, and that Intel is actually a good investment right now. And I felt I wanted to touch on some previous comments that were made.

Quote :

The earliest partner roadmap from May claimed we would see 65nm Brisbane Athlon 64 processors in Q1'07. The most recent update says we will actually see some limited quantity shipments of 65nm Athlon 64 chips in December 2006. ...AMD has no single core 65nm AM2 Athlon processors on its roadmap at this time. ~Source



Quote :

Interestingly enough, despite AMD's recent announcement of its new 4x4 platform none of the motherboard manufacturers we spoke to had heard of it, much less had a design ready to go. We definitely got the impression that 4x4 was a last minute effort to compete with Conroe on the high end.

We've also encountered some frustration from motherboard manufacturers with AMD because they claim that AMD's availability of Socket-AM2 CPUs simply isn't as great as AMD had originally promised. ~Source



To reiterate, at least some motherboard manufacturers are already unhappy about the lack of the supply of AM2s. Secondly AMD themselves stated the 65nm Brisbane processors would be in short supply until somewhere around the first half of '07. This info is from Dailytech's own articles on Brisbane and Computex.

AMD's own "4x4" response to Conroe was a spur of the moment thing to steal back some publicity, and what was funny was none of the mainboard manufacturers had any prior knowledge of "4x4" before AMD's announcement of it. This is why I think Brisbane itself is the surprise hinted at the end in the Anandtech article. Specifically, I believe (And it's confirmed) AMD simply pushed up the dates on it.

Furthermore, I would expect AMD to have working 65nm engineering samples floating around... as Intel already has working samples of quad-core Kentsfields distributed to those motherboard manufacturers, who themselves stated they have Kentsfield samples already up and running in 965/975 chipset motherboards.

As for Brisbane: I am sure AMD knows they cannot simply ramp up the clock speed to match "Core"s performance, because once they do, Intel can reciprocate and AMD will be back to square 1, while Intel gets more publicity out of it for how powerful their new uArch is. Intel deciding to hold back on the 3.2ghz EE processor seems to support this theory, when considering that B0 stepping Core's can reach 3.8ghz on either stock voltage OR air cooling.

More evidence for my point of view is from Anandtech's recent technical article on server processors http://www.anandtech.com/IT/showdoc.aspx?i=2772, which I quoted a snippet from the concluding page below:

Quote :

The Intel Xeon 5160, a.k.a. Woodcrest, will simply be the most powerful server CPU this year (though it's not yet available for purchase of course). As our extrapolated calculations show, even a 2.6 GHz Woodcrest will outperform the current Opteron 285 with a 5 to 55% margin, nothing short of impressive. The new Xeon is however not invincible: the Opteron can still give some serious resistance when running some instruction mixes with lots of rotates, add-carry or load effective address instructions. RSA, AES and other benchmarks clearly show this. Intel will still have to convince some software vendors to port to SSE if it wants Woodcrest to be the completely superior CPU. The advantage in MySQL is also rather small, a result of the relatively high latency of the FB-DIMMs. But we are nitpicking: Intel's newest Xeon has taken back the performance/Watt crown. In one word: Woodcrest rocks!

...The result is that AMD will not be able to regain the performance crown in the dual and quad-core market until the K8L arrives.



Anand frequently made extrapolations throughout the article about how well a ramped up Opteron would perform to the 3ghz Woodcrest, and it was not pretty. If I followed that correctly, exchanging the higher latency FB-DIMMS for DDR2 DIMMs should only help the performance numbers with Core 2 Duo.



So, when does the K8L arrive that they mention? Well, which flavor of K8L... dual-core or quad-core? ;)

I got tired of hunting articles for mentions to dates, so I used articles dated from May/June to compile this chart from, also all from Dailytech with the exception of a Dec '05 dated Intel Roadmap update on THG. Any suggested updates, fixes to errors, or things to add to this chart are welcome, or if you know of a better one somewhere else that's fine too, just give me a linky please! Links to support this data will be given upon request (Assuming I can find them again!). AMD shoving half of their long-term plans forward half a year for some preventive damage control didn't help though.


http://img466.imageshack.us/img466/1530/cpuroadmap5qq.png


Also, I found an old article on THG that I can't quite believe, can anyone corroborate this?

Quote :

Fab D1D in Hillsboro, Oregon, will be 45-nanometerized first in order to be ready for mass manufacturing of final 45 nm products in the second half of 2007. This generation is still going to be based on the processor design that Merom is going to introduce in mid 2006, but due to several modifications in addition to the die shrink it will be called Penryn. ~ Tom's Hardware Guide



If Intel really is still planning to ramp to 45nm while AMD is just finishing their 65nm ramp in '07, it isn't going to be pretty... It already appears that Intel will be touting 8-core chips when AMD releases their quad-core designs and much looked forward to quad-K8L...

Okay, since I got into processors I guess I have to throw this one in too. One thing I (amazingly) have not mentioned in this huge post is Intel's Nehalem, which I had never previously heard about until I became attracted like a moth towards a 300watt light bulb to the stickied Conroe information thread. The only legitimate article I found on Nahelem is http://www.pcworld.com/news/article/0,aid,105951,00.asp. In this article they also go on to mention

Quote :

In 2003, the company plans to release an updated version of the Pentium 4, code-named Prescott, which will be produced using a 90-nanometer process and incorporate several improvements over existing Pentium 4 models, Burns said.

so it would seem to be Nehalem has already come and gone under a different name. :lol:

To use a less than reliable source: http://www.theinquirer.net/?article=31316

Quote :

The new chips this year will use 65 nanometre but he said Intel will ramp to 45 nanometre swiftly using the Nehalem technology. Yes, Intel re-uses codenames.

So there is no Nehalem, at least outside of some kind of process technology, not a CPU uArch/design....

Edited to add in the sources to the first two quotes that I should have done in the first place and a typo or three.


Wow, so that's what not having a life is like. Wow.

Profile: Eternal Poster
More Information

Wow, so thats what pointless posts look like, wow. STFU.

Profile: Forum Resident
More Information

Quote :

It is hard to say, Z-RAM is a great technology, but I do not think AMD would be able to fold it into the process this quickly. Who knows, stranger things have happened.



It's been said that ZRAM would have to be at least 8MB in size for the performance to offset the latency penalty. I wonder if AMD could ever manufacture something similar to Xenos's "smart cache" linked up with the HTT bus? They stated that K8L will be a modular design and all the HTT links would surely make the technology viable. What do you think Jack?

Who said ZRam Has latency issues? I heard it was as fast as SRAm but half the size.

Profile: old hand
More Information

Quote :

Wow, so thats what pointless posts look like, wow. STFU.



Second that.

Profile: Forum Resident
More Information

Quote :

Wow, so thats what pointless posts look like, wow. STFU.



Ummmm, aren't you extinct? I was just amazed that he wrote all of that considering that I doubt Intel is paying him anything. Most of you just sound like butt boys of Otellini. I like both companies but AMD was ALWAYS more affordable and JUST AS FAST. In some cases faster.

You're just mad cause you can't play that "OWNED" crap with me.

Profile: Forum Resident
More Information

Quote :

It is hard to say, Z-RAM is a great technology, but I do not think AMD would be able to fold it into the process this quickly. Who knows, stranger things have happened.



It's been said that ZRAM would have to be at least 8MB in size for the performance to offset the latency penalty. I wonder if AMD could ever manufacture something similar to Xenos's "smart cache" linked up with the HTT bus? They stated that K8L will be a modular design and all the HTT links would surely make the technology viable. What do you think Jack?

Who said ZRam Has latency issues? I heard it was as fast as SRAm but half the size.


And before the p-nut gallery speaks here's a link.
Non-Missing Link

Profile: Eternal Poster
More Information

Quote :

I like both companies but AMD was ALWAYS more affordable and JUST AS FAST. In some cases faster.



They won't for much longer. :wink:

Quote :

I was just amazed that he wrote all of that considering that I doubt Intel is paying him anything. Most of you just sound like butt boys of Otellini.



Sure fanboy.

Quote :

You're just mad cause you can't play that "OWNED" crap with me.



Man that was funny seeing you get owned on anandtech.

Profile: Ancient Poster
More Information

Quote :

Wow, so thats what pointless posts look like, wow. STFU.



Word.

Profile: Forum Resident
More Information