Latest AMD server CPU Roadmap

[Idontcare]

Original (in Japanese): http://pc.watch.impress.co.jp/...008/0513/kaigai439.htm

Translated to English: http://translate.google.com/tr...a&tl=en&hl=EN&ie=UTF-8

Old roadmap versus new roadmap: http://pc.watch.impress.co.jp/...2008/0513/kaigai_1.jpg

Ouch...they are showing Bulldozer pushed out 3 quarters into 2011 now (that's 3 yrs from now!). This can only mean one thing...Bulldozer is being completely redone over from whatever they had on the drawing board last year.

Here's your decoder ring for the now planned K10 server chips: http://pc.watch.impress.co.jp/...2008/0513/kaigai_2.jpg

And this one is the brutal one - an Intel vs AMD timeline of expected product releases: http://pc.watch.impress.co.jp/...2008/0513/kaigai_3.jpg

It's just unbelievable to me that Westmere (32nm Nehalem) is going to be out for >1yr before bulldozer. The next 3 years are going to be hellish quiet in the CPU world.

 

[bryanW1995]

1. to paraphrase viditor: NOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO!!!!

2. to paraphrase nemesis: YEEEEEEEEHAWWWWWWWWWWW!!

for my wallet's reaction, see #1.

 

[Martimus]

Originally posted by: bryanW1995
1. to paraphrase viditor: NOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO!!!!

2. to paraphrase nemesis: YEEEEEEEEHAWWWWWWWWWWW!!

for my wallet's reaction, see #1.

That made me laugh.

I understand why Viditor feels the way he does; he has a boat load of money tied up in AMD. What I don't understand is why Nemesis feels the way he does. I have a hard time reading his posts because they are usually a little incoherent, but I don't understand his hatred of AMD. Does he have a large amount of money tied up in Intel?

 

[myocardia]

Ouch. I wonder what Hector Ruin will have to say about it.

Originally posted by: Idontcare
The next 3 years are going to be hellish quiet in the CPU world.

Do you mean like the last 2½ have been?

 

[aigomorla]

so there scraping montreal for istanbul?

IDC whats the difference between I/F vs G3MX? Im assuming G3MX is a lot better.

6 core istanbul, ummmm, rather get an 8 core neha.

Seems like AMD wont have an octocore out until 2010 while intel should have some at the end of this year to start of next year. :\

Guess AMD is out for me another year.

 

[Extelleron]

Originally posted by: aigomorla
so there scraping montreal for istanbul?

IDC whats the difference between I/F vs G3MX? Im assuming G3MX is a lot better.

6 core istanbul, ummmm, rather get an 8 core neha.

Seems like AMD wont have an octocore out until 2010 while intel should have some at the end of this year to start of next year. :\

Guess AMD is out for me another year.

It seems on the server side, 6-core Istanbul and MCM 12-core Magny Cours are the direction in which AMD is heading; Compared to Montreal, this makes sense. Montreal was just quad w/ 1MB of L2 per core instead of 512KB, which won't affect performance much. Istanbul's 6-cores will benefit server performance more and not be too expensive to produce. A conceputal drawing of Istanbul by Hans de Vries suggested die size would be ~294mm^2, so around the size of 65nm Barcelona.

Of course by the time Magny Cours is out Intel will have Octal-core Nehalem @ 32nm, and with higher IPC, SMT, and higher clocks, could likely beat even the 12-core beast in a lot of applications.

 

[Phynaz]

Do I get to say I told you so?

And here I was about to break down and buy a few shares. Thanks for the info.

 

[CTho9305]

Originally posted by: Martimus
What I don't understand is why Nemesis feels the way he does. I have a hard time reading his posts because they are usually a little incoherent, but I don't understand his hatred of AMD. Does he have a large amount of money tied up in Intel?

He dislikes AMD because AMD has a manufacturing partnership with IBM, and he dislikes IBM because he believes IBM helped Hitler exterminate Jews during the holocaust. No, I'm not making this up, and I can find where he's stated that if you'd like. I'm not sure why Phynaz is so gleeful whenever AMD has trouble.

 

[Idontcare]

I go back to one of my early postings where I said I was bullish on AMD in the mid-term but that the near-term was going to be a crappy ride.

AMD's management don't need 60% gross margins, they just need to pare back the company's size and scope for a couple years (sucks for the employees who fall under the cutbacks, but better some hurt than all hurt), learn to live with single-digit marketshare while they hunker down and do the work necessary to deliver a stellar architecture just like the Athlon K7 did so they can come back with a vengence.

Intel can't do that, they will do a DEC before they reduce their corporate profile and lifestyle to that of a Via. But that is what AMD needs to do for a couple years here.

Bulldozer coming out a full year after Westmere is not good for the near-term. It means the K10.5 Shanghai core is going to have to compete not only with Penryn and Nehalem but also Westmere. Triple ouch.

But in the mid-term, 3-5 years, I am much happier to see AMD taking Westmere and Sandy Bridge seriously enough to be willing to head back to that drawing board and rework a doozy of an architecture. This is painful timeline wise, but it is promising architecture wise as it is a sign that they are clearly at least trying to get ahead of the Intel tic-toc curve again.

 

[Idontcare]

Originally posted by: bryanW1995
1. to paraphrase viditor: NOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO!!!!

2. to paraphrase nemesis: YEEEEEEEEHAWWWWWWWWWWW!!

for my wallet's reaction, see #1.

Viditor is a stock speculator so I imagine he'll do the analysis and look for opportunities to accumulate more low priced AMD stock during the extended drought of positive expectations.

Buying low and selling high is a one-time trick unless you get cyclical volatility, and AMD is the poster child for a sideways cyclical stock if you look at a weekly or monthyl candlestick chart for the past 25 years.

 

[Phynaz]

Originally posted by: CTho9305

Originally posted by: Martimus
What I don't understand is why Nemesis feels the way he does. I have a hard time reading his posts because they are usually a little incoherent, but I don't understand his hatred of AMD. Does he have a large amount of money tied up in Intel?

He dislikes AMD because AMD has a manufacturing partnership with IBM, and he dislikes IBM because he believes IBM helped Hitler exterminate Jews during the holocaust. No, I'm not making this up, and I can find where he's stated that if you'd like. I'm not sure why Phynaz is so gleeful whenever AMD has trouble.

Mostly because baiting Viditor is so much fun.

Remember, I've always said that nobody should take anything I say with more than a grain of salt (maybe not even that), being that I'm nothing but a poster on a message board.

 

[Foxery]

Originally posted by: Idontcare
And this one is the brutal one - an Intel vs AMD timeline of expected product releases: http://pc.watch.impress.co.jp/...2008/0513/kaigai_3.jpg

It's just unbelievable to me that Westmere (32nm Nehalem) is going to be out for >1yr before bulldozer. The next 3 years are going to be hellish quiet in the CPU world.

That's painful, and depressing. I don't suppose outsourcing their fabrication to TSMC will help speed things up if the rumors come true? i.e. They can spend their time and energy on design, and let the specialty company worry about putting it on silicon.

Why is making a 6-core version (Istanbul) of current architecture (Phenom family, 45nm shrink) going to take an entire year??


A general concept that's been in the back of my mind lately is that, if they know they can't compete on performance any time soon, I hope they make some strides in energy efficiency, and carve out some market share that way.

 

[Viditor]

Originally posted by: Idontcare
Original (in Japanese): http://pc.watch.impress.co.jp/...008/0513/kaigai439.htm

Translated to English: http://translate.google.com/tr...a&tl=en&hl=EN&ie=UTF-8

Old roadmap versus new roadmap: http://pc.watch.impress.co.jp/...2008/0513/kaigai_1.jpg

Ouch...they are showing Bulldozer pushed out 3 quarters into 2011 now (that's 3 yrs from now!). This can only mean one thing...Bulldozer is being completely redone over from whatever they had on the drawing board last year.

Here's your decoder ring for the now planned K10 server chips: http://pc.watch.impress.co.jp/...2008/0513/kaigai_2.jpg

And this one is the brutal one - an Intel vs AMD timeline of expected product releases: http://pc.watch.impress.co.jp/...2008/0513/kaigai_3.jpg

It's just unbelievable to me that Westmere (32nm Nehalem) is going to be out for >1yr before bulldozer. The next 3 years are going to be hellish quiet in the CPU world.

Ummm...let's keep this in perspective. This is pure guesswork from Hiroshi Gotou Shigeru (which he rightfully acknowledges in the graphic).

Nothing in the article is based on actual known facts or even inside rumours. Hiroshi is just posting what he thinks is going to happen...

 

[Viditor]

Originally posted by: Idontcare

Originally posted by: bryanW1995
1. to paraphrase viditor: NOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO!!!!

2. to paraphrase nemesis: YEEEEEEEEHAWWWWWWWWWWW!!

for my wallet's reaction, see #1.

Viditor is a stock speculator so I imagine he'll do the analysis and look for opportunities to accumulate more low priced AMD stock during the extended drought of positive expectations.

Buying low and selling high is a one-time trick unless you get cyclical volatility, and AMD is the poster child for a sideways cyclical stock if you look at a weekly or monthyl candlestick chart for the past 25 years.

Well said...
I actually only post pro-AMD comments (and keep in mind that I've posted many comments that go the other way) because there are so few of them posted here. It makes me crazy to only see one side of an argument...

 

[Idontcare]

Originally posted by: Viditor

Originally posted by: Idontcare
Original (in Japanese): http://pc.watch.impress.co.jp/...008/0513/kaigai439.htm

Translated to English: http://translate.google.com/tr...a&tl=en&hl=EN&ie=UTF-8

Old roadmap versus new roadmap: http://pc.watch.impress.co.jp/...2008/0513/kaigai_1.jpg

Ouch...they are showing Bulldozer pushed out 3 quarters into 2011 now (that's 3 yrs from now!). This can only mean one thing...Bulldozer is being completely redone over from whatever they had on the drawing board last year.

Here's your decoder ring for the now planned K10 server chips: http://pc.watch.impress.co.jp/...2008/0513/kaigai_2.jpg

And this one is the brutal one - an Intel vs AMD timeline of expected product releases: http://pc.watch.impress.co.jp/...2008/0513/kaigai_3.jpg

It's just unbelievable to me that Westmere (32nm Nehalem) is going to be out for >1yr before bulldozer. The next 3 years are going to be hellish quiet in the CPU world.

Ummm...let's keep this in perspective. This is pure guesswork from Hiroshi Gotou Shigeru (which he rightfully acknowledges in the graphic).

Nothing in the article is based on actual known facts or even inside rumours. Hiroshi is just posting what he thinks is going to happen...

And you know this because...?

 

[Viditor]

Originally posted by: Idontcare

Originally posted by: Viditor

Originally posted by: Idontcare
Original (in Japanese): http://pc.watch.impress.co.jp/...008/0513/kaigai439.htm

Translated to English: http://translate.google.com/tr...a&tl=en&hl=EN&ie=UTF-8

Old roadmap versus new roadmap: http://pc.watch.impress.co.jp/...2008/0513/kaigai_1.jpg

Ouch...they are showing Bulldozer pushed out 3 quarters into 2011 now (that's 3 yrs from now!). This can only mean one thing...Bulldozer is being completely redone over from whatever they had on the drawing board last year.

Here's your decoder ring for the now planned K10 server chips: http://pc.watch.impress.co.jp/...2008/0513/kaigai_2.jpg

And this one is the brutal one - an Intel vs AMD timeline of expected product releases: http://pc.watch.impress.co.jp/...2008/0513/kaigai_3.jpg

It's just unbelievable to me that Westmere (32nm Nehalem) is going to be out for >1yr before bulldozer. The next 3 years are going to be hellish quiet in the CPU world.

Ummm...let's keep this in perspective. This is pure guesswork from Hiroshi Gotou Shigeru (which he rightfully acknowledges in the graphic).

Nothing in the article is based on actual known facts or even inside rumours. Hiroshi is just posting what he thinks is going to happen...

And you know this because...?

They don't quote any sources at all, and the graphic even says it's an estimate...

 

[Martimus]

Originally posted by: CTho9305

Originally posted by: Martimus
What I don't understand is why Nemesis feels the way he does. I have a hard time reading his posts because they are usually a little incoherent, but I don't understand his hatred of AMD. Does he have a large amount of money tied up in Intel?

He dislikes AMD because AMD has a manufacturing partnership with IBM, and he dislikes IBM because he believes IBM helped Hitler exterminate Jews during the holocaust. No, I'm not making this up, and I can find where he's stated that if you'd like. I'm not sure why Phynaz is so gleeful whenever AMD has trouble.

Wow.

 

[Kuzi]

AMD's revised roadmaps had no mention of Bulldozer at all, so this japanese guy might be right to guess that Bulldozer will not be released until AMD are at 32nm, which should be around 2011.

Originally posted by: Idontcare
Ouch...they are showing Bulldozer pushed out 3 quarters into 2011 now (that's 3 yrs from now!). This can only mean one thing...Bulldozer is being completely redone over from whatever they had on the drawing board last year.

Maybe just delayed because it simply can't be done on 45nm.

 

[Martimus]

Originally posted by: Kuzi
Maybe just delayed because it simply can't be done on 45nm.

That would be my guess. They can't make an economically feasible Bulldozer until they have a smaller process. But I don't have any experience in die shrinks so I really don't know much about their limitations, except what I have read around.

 

[Idontcare]

Originally posted by: Kuzi

Originally posted by: Idontcare
Ouch...they are showing Bulldozer pushed out 3 quarters into 2011 now (that's 3 yrs from now!). This can only mean one thing...Bulldozer is being completely redone over from whatever they had on the drawing board last year.

Maybe just delayed because it simply can't be done on 45nm.

This could very well be true IF bulldozer was being designed for xtor parametrics that were expected to come from HK/MG 45nm AND AMD just decided to not release their rev2 45nm with HK/MG and this is the earliest indication of such (that HK/MG has internally been decided to be pushed out to 32nm now).

 

[Foxery]

Er. Surely the original designs for a new chip are written with the manufacturing process in mind? I can't imagine they spend a year building an engine only to say, "oops, it doesn't fit under the hood."

How early or late in the game do they create the first samples on real silicon?

 

[Idontcare]

Originally posted by: Foxery
Er. Surely the original designs for a new chip are written with the manufacturing process in mind? I can't imagine they spend a year building an engine only to say, "oops, it doesn't fit under the hood."

How early or late in the game do they create the first samples on real silicon?

This was my initial thought too, xtor budgets defined by commercially defined die-size allowances are among the first boundary conditions defined for an IC program.

So given the obvious, that slipping from 45nm to 32nm could not simply be due to the design suddenly exceeding its budgeted xtor count (and corresponding die size restriction) the only other viable explanation I could forsee is that the expected performance of the xtors themselves has changed and thus negated the prospects of producing a commercially viable product clock-speed wise.

Ergo the theory I posited above by suggesting this is some of the first external indications we have that AMD has internally decided to slip HK/MG from 45nm to 32nm.

(BTW I have witnessed similiar before wrt SUN's designs for particular nodes being slipped out a node because it became obvious early enough in the design timeline that the target node's xtor performance was not likely to meet the needs...we (Texas Instruments) slipped HK/MG from 65nm to 45nm for this very reason)

 

[Martimus]

Originally posted by: Foxery
Er. Surely the original designs for a new chip are written with the manufacturing process in mind? I can't imagine they spend a year building an engine only to say, "oops, it doesn't fit under the hood."

Actually, that is exactly what happens at times with engines. It isn't because they designed it to be too big, it is because they overestimated the efficiency of the engine and had to make it bigger to compensate. They design it to one spec, but when they actually test it there are problems which usually involve making it bigger. I would assume the same thing happened here, where the die size just got too big at 45nm during development to feasibly build in order accomplish the performance they were going for.

 

[biostud]

According to this AMD is not going to support DDR3 until 2010