backtop


Print 91 comment(s) - last by excrucio.. on Jun 1 at 11:12 PM

Even before AMD's 65nm K10 architecture hits store shelves, the company is talking about the 45nm shrink

This summer, AMD will announce its first major architectural change since the introduction of the K8 architecture in 2003.  This new architecture, dubbed K10, will first make an appearance in the server space, with the introduction of the Barcelona-family processors.

K10 features a native quad-core design that incorporates shared-L3 cache, HyperTransport-3 support and backwards functionality with AM2 motherboards.  However, the original K10 desktop and server processors will debut on the 65nm architecture -- a process AMD only started mastering in December 2006 with the launch of the Brisbane desktop CPU family.

In the second half of 2008, AMD will begin to migrate its K10 architecture to the 45nm node.  AMD explicitly mentions that its 45nm process technology utilizes silicon-on-insulator (SOI).  Intel's 45nm process node, slated for introduction later this year, uses conventional CMOS process technology. 

The halo AMD 45nm chip, Deneb FX, shares the same functionality as its 65nm counterpart, Agena.  Both families incorporate native quad-core designs and shared-L3 cache support.  Deneb FX goes one step further, adding support for DDR3 on the integrated memory controller.

However, the bulk of AMD's 45nm quad-core offerings will come with the Deneb (non-FX) family.  AMD suggests Deneb will be the first processor on the new AM3 socket.  Previous AMD documentation indicated that AM2 and AM3 would be forward/backward compatible -- yet AMD engineers claim the AM3 alluded to in 2006 is not the same AM3 referenced in the 2008 launch schedule. 

"At the time AM3 was the likely candidate to become AM2+," claimed one field application engineer familiar with AMD's socket migration. "[AMD] wanted to keep the socket name associated with DDR2 memory and backwards compatibility, but AM3 emphasizes DDR3 support."

After Deneb, and closer to 2009, AMD's guidance states that 45nm Propus and Regor will replace the 65nm Kuma and Rana mid-range productsPropus is very similar to Deneb: 45nm, shared L3 cache, AM3 package.  However, Propus will only feature two cores.  Regor is identical to Propus, but will not include shared-L3 cache support.

AMD's low-end single core Athlon 64 and Sempron appear consolidated with the introduction of the Sargas family.  Sargas is an optical shrink of the 65nm Spica core, with the addition of DDR3 support and AM3 packaging.  AMD's ultra-low end Sparta-family, slated for introduction this year to replace the Manila-family Semprons, has no successor.

AMD product managers are keeping details of their 45nm technology close.  However, this past January AMD and IBM jointly announced plans for high-k, metal gate transistors on future 45nm and 32nm processors. 

This past February, AMD senior vice president of technology development Douglas Grose claimed the company is still anticipating whether or not it will use high-k metal gate technology in later 45nm revisions or if the company will wait until 32nm.

Intel also announced its intention to debut high-k, metal gate technology on its 45nm node, but the company went one step further to confirm this new transistor technique will appear on the Penryn processor.  Intel guidance suggests Penryn will see its first retail availability late this year -- at least a year before Deneb.

Marty Seyer, AMD senior vice president, recently disclosed AMD's 45nm server offering slated for release in 2008.  Seyer stated that Shanghai, the 45nm successor to Barcelona, would feature additional cache and other performance enhancements. 

Seyer or Grose would not comment on what these performance enhancements, though features from AMD's server products typically appear on the desktop components as well.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: Learn to crawl before you can walk
By Mitch101 on 5/2/2007 9:56:40 AM , Rating: 3
From what I have heard on the manufacturing side the shrink doesnt seem to be a problem on any of the fab companies.

Seems the process or materials that works at 65nm is the same at 45nm with little to no redesign changes needed. Call them all lucky as this is how Intel is able to go to 45nm so fast. Almost a free shrink. Overclocks are only slightly more roughly 10% than existing. Im not an engineer so I cant comment on why. Doesnt mean they wont get faster with time but out of the gate expect 10% more than existing chips before exotic cooling is applied and lower temps to start.

However at 32nm the engineering teams need to come up with some new tech but from what I am hearing Vertical chip layering might be more important here than shrinks and AMD partnering with IBM could provide a small advantage for AMD should they be able to hang in there that long. If verticle chips become the next gen then Intel might have to play catch up to IBM/AMD. Not everything will be resolved with shrink it again if layering can be done effectively.

On another front the K10 is faster than Penryn in non SSE4 benchmarks. Go ahead ask me where I heard this? Its an Intel leak not an AMD leak. Intel knows this and that is why they are trowing out SSE4 benchmarks out early but on non SSE4 benchmarks K10 is faster than Penryn. You heard me. As for why AMD is tight lipped is it needs to liquidate the excess Dell screwed them inventory. If AMD pre announces then AMD will end up holding excessive inventory on chips they might have a hard time unloading without losing a ton more money. Where the chips really shine is on the Multi-CPU server side.

So lets say AMD announces and releases benchmarks then their current chips K8 still in the channel stop selling all together. AMD doesnt have the inventory yet to do a full on release and the channel trickles in because no one wants current gen and they only want next gen K10. They will find themselves sitting on a ton of previous gen and barely able to keep up with the channel. Thankyou Dell! scumbags. So AMD chose to wait until they have a ton of chips in stock to flood the channel and hopefully eliminate the majority of K8 stock that Dell screwed them on.

As for AMD aquiring ATI as being a bad move doing this allows AMD to certify motherboards for server production something which would take them a very long time if they didnt own or make a motherboard chipset. NVIDIA was too expensive for AMD. ATI was the right price despite the R600 not being an NVIDIA killer. AMD will have server certified chipsets available immediately. This is more important than graphics because the server market is one of the first markets AMD received most of their profits from. From a gamers point it was a mistake from a Server perspective it was the right choice.

Lets talk R600. Its not a total disaster as they certainly can compete with the mid range line which is where the majority of chips sell and AMD has already said they will be competitivly priced. How can AMD do this well they own chip foundries and while the R600 will be made off site they can certainly make chips inhouse cheaper than NVIDIA giving them a price advantage that NVIDIA cant match because NVIDIA doesnt own chip foundries. As well as ATI's DX10 chipset is DX10 compliant with DRM (Puke) and physics able which has yet to be exploited for use. FRAMERATE DOES NOT MEAN SUPERIOR. If its over 60fps at any resolution the game plays excellent. If one does 80FPS and the next does 60fps it doesnt matter because your monitor refresh rate is still probably 60hz which is most LCD panels. Framerates are becoming a thing of the past and what will make the eventual difference is EYE CANDY and PHYSICS.

Let us not forget ATI makes chips for consoles and this will be an ongoing process for future as well. ATI is not a total bust.

Lastly I will leave you with the last rumor that in the X-Box portable gaming machine might beat the heart of an ATI chip. ;)


RE: Learn to crawl before you can walk
By mikecel79 on 5/2/2007 11:52:16 AM , Rating: 1
quote:
AMD will have server certified chipsets available immediately.

What "server certified chipset" did ATI ever produce? ATI and nvidia may make fine desktop chipsets but they do not belong in any server environment. The only chipsets I would ever trust in a real server environment are made by Intel, Serverworks (Broadcom), or IBM.

Even if ATI were to be building a server chipset it would not be available immediately. It would take months to be certified before any OEM would touch it.


RE: Learn to crawl before you can walk
By Mitch101 on 5/2/2007 12:58:31 PM , Rating: 2
And that process will continue with the addition of ATI providing certified chips as well.

Think about your statement. You trust Intel CPU with an Intel Chipset but you dont trust AMD with an AMD/ATI chipset?

The R690 is already Vista WHQL certified.
AMD/ATI chipsets will be server certified.

There are over 30 motherboard companies already signed up to use the AMD/ATI chipsets.

Its no different than Intel Creating server chips for Intel.

AMD is doing the same and will certify thier chipsets which will cut down on the time for server certification.

As I hear a 2.5ghz K10 Quad is about 33% faster than Intel's 3.0ghz quad core. This was in a benchmark the Intel chip usually wins. Now I wonder why Intel is quickly moving up its 45nm processors and releasing SSE4 benchmarks? Hmmm. 4x4 anyone?

There is a market for something that fast. Expect server demand to be high and chipsets to be certified fast.


RE: Learn to crawl before you can walk
By mikecel79 on 5/2/2007 1:40:12 PM , Rating: 2
quote:
Think about your statement. You trust Intel CPU with an Intel Chipset but you dont trust AMD with an AMD/ATI chipset?


I never said I trusted Intel CPUs on only an Intel chipset. We have many servers that run Intel CPUs on Broadcom, Intel, and IBM chipsets. I stated that ATI has no experience making SERVER chipsets. Why should I trust them to make a reliable server chipset?

quote:
The R690 is already Vista WHQL certified.
AMD/ATI chipsets will be server certified.


Vista WHQL doesn't mean anything in the server space. All it means is that the the drivers function correctly. Vista is no a server OS. There could still be underlying problems with the chipset. High end server use and is very different than a single person using a desktop.

quote:
There are over 30 motherboard companies already signed up to use the AMD/ATI chipsets.


Wow that's great. How many of them are creating motherboards for HP, Dell, IBM or Sun servers utilizing chipsets that were designed by ATI? Zero.

quote:
Its no different than Intel Creating server chips for Intel.


Actually it is. Intel has been doing this for over a decade now. ATI has only been desigining chipsets for a few years. AMD (not ATI) has made server chipsets in the past that have been reliable but ATI has NO experience in the server space. AMD did NOT purchase ATI for it's server chipset experience.

quote:
As I hear a 2.5ghz K10 Quad is about 33% faster than Intel's 3.0ghz quad core. This was in a benchmark the Intel chip usually wins. Now I wonder why Intel is quickly moving up its 45nm processors and releasing SSE4 benchmarks? Hmmm. 4x4 anyone?


How about a link?


By Mitch101 on 5/2/2007 2:05:06 PM , Rating: 2
As if links in the internet were what seperates truth from fiction. Google it there are a few around.

Obviously nothing I say is going to get a positive remark with you or the Intel crew obviously by the downward votes I am seeing on my comments so I can only suggest you wait for it.

This is my last post about this but all I will say is Intel is going to get smacked 6 ways to sunday.

Like the way the Opteron Smacked the P4, the Conroe Smacked the Opteron back, the K10 is going to smack the Xeon/Conroe. Thats not fanboism thats fact but you will know that soon enough.


By ObscureCaucasian on 5/2/2007 7:51:47 PM , Rating: 2
This is probably what he was talking about.... it isn't what I'd call the most reliable information, but it's about all we got.

http://www.fudzilla.com/index.php?option=com_conte...


RE: Learn to crawl before you can walk
By defter on 5/2/2007 1:12:21 PM , Rating: 2
You seem to forget that a large part, if not majority of Opteron servers are using NVidia chipset.

You probably haven't heard about this?:
http://www.nvidia.com/page/nforce_pro.html
http://www.nvidia.com/page/nforce_pro_wtb.html


By mikecel79 on 5/2/2007 2:16:49 PM , Rating: 2
Actually most of the more popular modern Opteron servers are now using the Broadcom (Serverworks) chipsets.

HP DL365 - No information found
HP DL385 G2 - Serverworks HT-2100 and HT-1000
HP DL585 G2 - nVidia NForce Professional 2200 and 2050 chipsets, and AMD-8132 chipset

IBM x3455 - Serverworks HT-2100 and HT-1000
IBM x3655 - Serverworks HT-2100 and HT-1000
IBM X3755 - Serverworks HT-2100 and HT-1000

Dell PowerEdge 6950 - Serverworks HT-2100 and HT-1000
Dell PowerEdge 2970 - Serverworks HT-2100 and HT-1000
Dell PowerEdge SC1435 - Serverworks HT-2100 and HT-1000

Sun Fire X2100 M2 - nVidia (not sure which one)
Sun Fire X4200 - AMD 8000


By raven3x7 on 5/2/2007 4:51:17 PM , Rating: 2
Nvidia is probaby the largst provider of workstation chipsets on the AMD platform


RE: Learn to crawl before you can walk
By ybee on 5/2/2007 12:19:34 PM , Rating: 2
quote:
Seems the process or materials that works at 65nm is the same at 45nm with little to no redesign changes needed.


Wait, Intel will use high-k/metal gate in its 45 nm process. Isnt it suposed to be a big change?

On the other hand they are not swithching to EUV lithograpy yet, so in this respect the transition is going indeed to be relatively easy. Is this waht you mean?



By Mitch101 on 5/2/2007 12:39:25 PM , Rating: 2
Well put as I said I am not an engineer I can only reinterate what someone said about the process of going from 65nm to 45nm as being minimal changes.


By Phynaz on 5/2/2007 12:21:04 PM , Rating: 1
quote:
Seems the process or materials that works at 65nm is the same at 45nm with little to no redesign changes needed. Call them all lucky as this is how Intel is able to go to 45nm so fast.


Intel is doing complete transistor material change at 45nm.


RE: Learn to crawl before you can walk
By whickywhickyjim on 5/2/2007 12:39:52 PM , Rating: 2
quote:
They will find themselves sitting on a ton of previous gen and barely able to keep up with the channel.

That's already happening. Nobody wants the bargain bin shite they're pushing on the market now. If they had only made new 939 chips on the 65nm or even 90nm process, their Q1 sales wouldn't be in the toilet.


By Dactyl on 5/2/2007 1:42:25 PM , Rating: 2
I'm going to upgrade my S939 CPU from 2.4GHz single core to 2.0GHz dual core in the next month. I wouldn't be doing this except for the price drop.

It would be nice if I could upgrade to a dual core K10 in this socket. There's no reason dual channel DDR1 couldn't feed a K10 dual core.


RE: Learn to crawl before you can walk
By FITCamaro on 5/2/07, Rating: -1
By kilkennycat on 5/2/2007 1:39:29 PM , Rating: 2
ATI does not own the Xbox360 chips. They licensed the complete design to Microsoft who handle the complete interface with TSMC. Unlike the original XBox design with nVidia, where nVidia supplied the chips. I believe ATi also gets a royalty fee for each Xbox360 sold. ATi also may get additional design fees for any required participation in a die-shrink. As for the arrangement between nVidia and Sony, I have no details, but since Sony is a perfectly competent hardware manufacturer, I suspect that the arrangement with nVidia was again design fees plus per-shipment royalties. Simplest for all parties involved.


By ObscureCaucasian on 5/2/2007 7:59:37 PM , Rating: 2
ATI has Wii and 360, nVidia has PS3. ATI has number 1 and 2, nVidia has #3. All the chips are manufactured by other parties, but they still get royalties. The gpu in the original Xbox was the only chip that a console manufacturer ever bought (wasn't fabbed by the console manufacturer). I guess nVidia started screwing MS on pricing too because they wouldn't lower their price as technology progressed.

Basically ATI is in much better position console wise.


"The Space Elevator will be built about 50 years after everyone stops laughing" -- Sir Arthur C. Clarke

Related Articles
Intel Unveils "Penryn" Performance
April 17, 2007, 9:11 PM
AMD Talks Details on K10
April 14, 2007, 11:46 AM
Intel Pulls 45nm Xeon Launch Into 2007
February 21, 2007, 1:16 PM
AMD's 45nm Opterons Scheduled for 2008
February 5, 2007, 11:50 AM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki