AMD 32 nm CPU Schedule Not In-Sync With ''Bulldozer'' Architecture Development

[btarunr]

It was reported earlier this week, about AMD advancing its development schedule for processors built on the newer 32 nm manufacturing process. A default interpretation of the event back then was the likelihood of AMD advancing the development of its next-generation CPU architecture, codenamed "Bulldozer". The said architecture will be the company's next indigenous x86 CPU design since K8, which was only improved upon with the K10 series.

A new report from TechConnect Magazine suggests that AMD's plans to advance development of 32 nm processors and those Bulldozer aren't in synchrony. This means that the company's first 32 nm processors are not likely to be built on the newer architecture. Bulldozer has already suffered delays. Earlier in 2007, AMD was expected to put a start to its 45 nm CPU lineup with CPUs based on Bulldozer in 2009, although that didn't happen. The first known processor based on Bulldozer, codenamed "Orochi" is expected to be a multi-core CPU with >4 cores, 8 MB L3 cache and a pure DDR3 integrated memory controller. Orochi is set for internal testing in late 2010, following which volume-production is set to commence in 2011.

View at TechPowerUp Main Site

 

[adrianx]

soo...

codenamed "OROCHI"

http://en.wikipedia.org/wiki/Orochi

aa...."8-branched giant snake"

codenamed "Orochi" is expected to be a multi-core CPU with >4 cores

soo will have... are real 8 core .... wellcome to future a coreI7 REAL KILLER (maybe)

 

[ShadowFold]

An actual 8 core, nice.

 

[Famous Hobo]

Sandy bridge will be out by then. Not to mention Intel is going 32nm by the end of this year.

 

[blueskynis]

Devlopment -> Development

 

[adrianx]

simple....marketing...

 

[TheMailMan78]

Devlopment -> Development

blueskynis -> grammar nazi.

 

[WhiteLotus]

This good news, just to raise some cash i expect. It is a shame they are taking there time with something like this.

Although maybe they can concentrate on a super efficient quad for all those normal households with HTPC.

 

[TheMailMan78]

This good news, just to raise some cash i expect. It is a shame they are taking there time with something like this.

Although maybe they can concentrate on a super efficient quad for all those normal households with HTPC.

I would rather them take their time and blow the doors of Intel than come out with something half-ass.

 

[WhiteLotus]

I would rather them take their time and blow the doors of Intel than come out with something half-ass.

My thoughts as well. This news release is to gain some cash.

My thoughts though is why not just skip this - try shrinking the die size even more than 32nm, get ahead of Intel. Intel is going 32nm by the end of the year, then will most likely do another die shrink a year and a bit after that (maybe?). If AMD go smaller than 32nm, make it supper efficient on heat, energy and cooler noise - maybe stick to quads and then double die it to an Octo like Intel did with Duals ->Quads.
If they can get this right then win for AMD... right?

 

[KBD]

bad news, i;m sure many were under the impression that 32nm meant new architecture, myself included. I dont know whats taking them so damn long, this is basically a survival issue for them. Intel went through 3 architectures since 2003, AMD is still on k8. May be it has to with money for R&D? But they just got an investment from an UAE company so they should have some for these purposes and move this along faster. All in all, i'm dissapointed, Intel will be on yet another new architecture by the time Bulldozer comes out.

 

[TheMailMan78]

bad news, i;m sure many were under the impression that 32nm meant new architecture, myself included. I dont know whats taking them so damn long, this is basically a survival issue for them. Intel went through 3 architectures since 2003, AMD is still on k8. May be it has to with money for R&D? But they just got an investment from an UAE company so they should have some for these purposes and move this along faster. All in all, i'm dissapointed, Intel will be on yet another new architecture by the time Bulldozer comes out.

Or you could see it as a great tribute to how well the k8 really is/was.

 

[suraswami]

Money Money Money. It can do things.

But one thing I don't understand, AMD is not a just sprung up company, they have been in this industry for a while. How hard is it to develop something new? Their brains work in 'Tortoise' speed?

They have lost ground on almost everything (except graphics), they can't afford to be slow in churning out things. They better be fast in getting the new arch out fast and this time it better be fully tested and functioning one.

 

[KBD]

Or you could see it as a great tribute to how well the k8 really is/was.

you wont get any argument about that from me. K8 is a great architrecture and was revolutionary for its time but its time has passed, unfortunately. If Intel had continued to build new CPUs based on Netburst or its derivitives K8(K10) would've continued to receive praise from entusiasts. But since Intel didnt, K8 was/is viewed in an entirely different light. AMD did manage to finally catch up to C2Q with Deneb and that is another testament to K8's strenght but since Nehalem is out its league it means that its time to move on to something else.

 

[cray86]

Intel will have 32nm 4-Cores out to the public by Q1 2010

And lets be fair, who really needs an 8-core unless you are doing -

Multimedia / Servers.

I see two problems. As a disclaimer I am not a fanboy, I just call it as I see it. In this market, AMD is going to get even more squeezed by Intel. Intel is hurting, but can survive longer. AMD needs to change their business model (which they are slowly... slowly doing) to focus on middle to lower end CPU's, ignoring the high end market completely. They need to keep ATI moving sense it's the only real money making, competative part of the company they have. The market cap for AMD right now is 1.3Billion. They bought ATI for somewhere in the 5 Billion range. They are in serious trouble. The only reason they stay afloat is because of antitrust issues with Intel. Ironically, that's what's keeping them on life support. NVIDIA is flirting with entering the mobile device x86 market as well. Intel's Atom is already dominating the netbook market, which is growing as well. Here's the question -

For financial viability (they are all businesses afterall) where is the room that AMD can grow into? I would argue

1 - Graphics
2 - Low end Desktop / Laptop CPU's (not netbooks or mobile devices)

Them balling for the server market will be risky, and probably in the end foolish.

 

[alwayssts]

Here's a crazy opinion:

I think the best time for AMD to launch a new arch is when we see Intel incorporate Larabee on-die...Which is exactly what Sandy Bridge is doing, and exactly the time frame.
This is the new frontier, with massive fpu/gpu cores on-die. I think AMD is missing an opportunity with not pairing a 32nm K10 with 28nm GPU next year to compete with 'Dale, but that is beside the point.

It is very important whom gets the on-die incorporation of this function right. I believe AMD will have an easier time having ATi's personel and building on their architecture, but OPENCL and such will largely made this moot when concerning GPGPU applications. Whom sets this function to be best incorporated and scalable, as well as the best cpu IPC to base it on will be important. I believe Sandy Bridge will not be much more than Nehalem+on-die larabee, as Intel has to devote a large amount of time/resources to building their new/different architecture. It's likely AMD's gpu cores or function will mimic the arrays of their gpus, which allows them more time to concentrate on cpu ipc. If they finally use Z-RAM or such, the cores could have a massive cache. This matched with up to eight cores, a better ipc, and/or some type of HT-like technology, and they could potentially leap frog the sleeping giant on the cpu front.

I can hardly blame AMD for this decision. The bright side is that people with AM2+/AM3 will have a platform that will last into 2011 with mutiple chipset and cpu upgrade options. X58/P55 on the otherhand will likely last through the 45/32nm current arch, and will require an upgrade to Sandy Bridge proprietary mobos. While the current (and 32nm version) of Intel's current arch will undoubtabley destroy K10, AMD is hunkering down for the next big battle which is more important than a short-term competative product. They know they can't compete now, but want to have something truely competative against Sandy Bridge, and for that I am hopeful. At least they are creating 32nm K10's, instead of just waiting until 2011 to launch the arch with the process, and that is what the 'moving up 32nm' announcement was about.

It all may look weak now, but how they finally pull of this transition could be a bigger deal than K8 vs. Netburst, as there are multiple fronts of consideration of what performance is (games, threading, cpu ipc, gpgpu) rather than just the how the CPU itsself compares to it's competitor.

Can I get an Amen, my brother?

 

[TheMailMan78]

Here's a crazy opinion:

I think the best time for AMD to launch a new arch is when we see Intel incorporate Larabee on-die...Which is exactly what Sandy Bridge is doing, and exactly the time frame.
This is the new frontier, with massive fpu/gpu cores on-die. I think AMD is missing an opportunity with not pairing a 32nm K10 with 28nm GPU next year to compete with 'Dale, but that is beside the point.

It is very important whom gets the on-die incorporation of this function right. I believe AMD will have an easier time having ATi's personel and building on their architecture, but OPENCL and such will largely made this moot when concerning GPGPU applications. Whom sets this function to be best incorporated and scalable, as well as the best cpu IPC to base it on will be important. I believe Sandy Bridge will not be much more than Nehalem+on-die larabee, as Intel has to devote a large amount of time/resources to building their new/different architecture. It's likely AMD's gpu cores or function will mimic the arrays of their gpus, which allows them more time to concentrate on cpu ipc. If they finally use Z-RAM or such, the cores could have a massive cache. This matched with up to eight cores, a better ipc, and/or some type of HT-like technology, and they could potentially leap frog the sleeping giant on the cpu front.

I can hardly blame AMD for this decision. The bright side is that people with AM2+/AM3 will have a platform that will last into 2011 with mutiple chipset and cpu upgrade options. X58/P55 on the otherhand will likely last through the 45/32nm current arch, and will require an upgrade to Sandy Bridge proprietary mobos. While the current (and 32nm version) of Intel's current arch will undoubtabley destroy K10, AMD is hunkering down for the next big battle which is more important than a short-term competative product. They know they can't compete now, but want to have something truely competative against Sandy Bridge, and for that I am hopeful. At least they are creating 32nm K10's, instead of just waiting until 2011 to launch the arch with the process, and that is what the 'moving up 32nm' announcement was about.

It all may look weak now, but how they finally pull of this transition could be a bigger deal than K8 vs. Netburst, as there are multiple fronts of consideration of what performance is (games, threading, cpu ipc, gpgpu) rather than just the how the CPU itsself compares to it's competitor.

Can I get an Amen, my brother?

Shut up.

 

[FordGT90Concept]

Sandy bridge will be out by then. Not to mention Intel is going 32nm by the end of this year.

Yup, Intel will be handling 16 threads (maybe even more) by then. I'm most interested in how many instructions per cycle it can handle being that it is a departure from K8/K10. Anything is really possible in regards to it.

 

[alwayssts]

Yup, Intel will be handling 16 threads (maybe even more) by then. I'm most interested in how many instructions per cycle it can handle being that it is a departure from K8/K10. Anything is really possible in regards to it.

I can't see it being more or less than 4-issue. With the way instructions are broken up versus conroe/Nehalem on K10 though (and it being 3-issue) it should be superior to what's currently around. That being said, things may change for SB.

 

[mdm-adph]

blueskynis -> grammar nazi.

Grammar != Spelling.

I am a grammar nazi.

 

[FordGT90Concept]

I can't see it being more or less than 4-issue. With the way instructions are broken up versus conroe/Nehalem on K10 though (and it being 3-issue) it should be superior to what's currently around. That being said, things may change for SB.

Sandy Bridge is to Nehalem as Penryn is to Conroe: maybe a 10% performance increase, tops.

AMD has always taken the wider-is-better approach so I wouldn't at all be surprised if it is >4 and consequently has a much lower clockspeed. Still, 8 MiB L3 cache is very large for an AMD processor. Hmm, there's a lot of potential there but it is way too soon to make any judgements.

 

[KBD]

While the current (and 32nm version) of Intel's current arch will undoubtabley destroy K10, AMD is hunkering down for the next big battle which is more important than a short-term competative product. They know they can't compete now, but want to have something truely competative against Sandy Bridge, and for that I am hopeful. At least they are creating 32nm K10's, instead of just waiting until 2011 to launch the arch with the process, and that is what the 'moving up 32nm' announcement was about.

Thats one way of looking at it. I guess in the end AMD is getting something that wasnt on the roadmap earlier - K10 on 32nm. Hopefully the arch can be tweaked further and not be a simple dieshrink, basically i'm hoping for another Deneb or better. And even though it probably wont be enough to take i7 head on it will at least close the gap a bit more.

 

[TheMailMan78]

Thats one way of looking at it. I guess in the end AMD is getting something that wasnt on the roadmap earlier - K10 on 32nm. Hopefully the arch can be tweaked further and not be a simple dieshrink, basically i'm hoping for another Deneb or better. And even though it probably wont be enough to take i7 head on it will at least close the gap a bit more.

I second that. Also I was messing with ya alwayssts

 

[a111087]

2011?! wow, hopefully it will not be too little, too late again... it better be something really special