backtop

Hardware AMD's Next-gen Socket F Revealed
Tuan Nguyen (Blog) - February 23, 2006 5:02 PM
Print E-mail del.icio.us 46 comment(s) - last by NullSubroutine.. on Mar 2 at 6:04 PM

New retention mechanism

1207 pins
AMD's new LGA-1207 is ready for quad-core Opterons

Although at this time, details are scarce, we are able to reveal some images of AMD's next-generation Opteron socket, called Socket F. AMD's Socket F will be a LGA-type socket with 1207 pins, and is an entirely new design from current Opteron sockets. We've read that the new socket will be used for upcoming dual-core Opterons as well as quad-core Opterons. The new socket comes with an updated retention mechanism.

Intel adopted the LGA socket design with Prescott to reduce cost and also to reduce the defect rate on processors.  After all, would you rather scrap a $600 CPU for a bent pin, or a $100 motherboard for a bent land grid?  A few days ago, we got some pictures of the upcoming Intel 771 LGA socket for Xeon motherboards.

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
Great News
By BaronMatrix on 2/23/2006 5:13:35 PM , Rating: 2
I was wondering where 1207 was. Socket F is supposed to support DDR3, isn't it?




RE: Great News
By JackPack on 2/23/2006 5:14:47 PM , Rating: 2
No, 1207 doesn't support DDR3 or FBD as far as I know.


RE: Great News
By BaronMatrix on 2/23/2006 5:26:41 PM , Rating: 2
I heard differently , but can't say. Even with DDR2, they will more than likely take Intel out back and beat the crap out of em again. It would be really cool if they got HT up to 2000/4000 with DDR2-800. The way it looks, they are going to drop the mobile 65nm first and add L3 to the Opteron in 2007. They actually have to for 16 way servers, but it's reasonable that they will want to increase performance before the new architecture and L3 will increase perf.


RE: Great News
By JackPack on 2/23/2006 5:53:20 PM , Rating: 2
DDR3 has such low capacity; 32 devices per channel if I recall correctly. It's not optimal for the server market.


RE: Great News
By BaronMatrix on 2/23/2006 5:29:43 PM , Rating: 2
I actually meant the memory controller, not the socket. I didn't hear about moving the PCIe tunnel, though.


RE: Great News
By cscpianoman on 2/23/2006 5:17:35 PM , Rating: 2
I doubt it. Most memory manufacturers are mostly into DDR2 and only produce DDR3 in minute amounts for the video card industry. Placing those into servers right away would be a very poor move by AMD to their bottom line, especially trying to support two formats at the same time.


RE: Great News
By Furen on 2/23/2006 6:19:26 PM , Rating: 3
DDR3 has nothing to do with GDDR3. GDDR3 is actually more akin to DDR2 with graphics-specific optimizations.


RE: Great News
By cscpianoman on 2/24/2006 9:46:12 AM , Rating: 2
Point taken, but I still hold out on the DDR2 manufacturing. Memory companies won't transition with new expensive equipment until the previous generation has given them all they can.


RE: Great News
By tonjohn on 2/23/2006 8:40:11 PM , Rating: 2
AMD has not made any official statements regarding DDR3 support on any of their nextgen CPUs, however, that does not mean that they don't support it.

I have a strong feeling that these new Opties will have DDR2, FB-DIMM, and DDR3 support (and possibly XDR support).

AMD recently entered into a licensing agreement with Rambus which would allow them to have a some pretty robust memmory controllers.


RE: Great News
By defter on 2/24/2006 1:51:21 AM , Rating: 2
"I have a strong feeling that these new Opties will have DDR2, FB-DIMM, and DDR3 support (and possibly XDR support)."

How they could have DDR3 support, when DDR3 specs aren't finalized yet? Besides, having three different memory controllers on the CPU is just silly.


"AMD recently entered into a licensing agreement with Rambus"

AMD did enter into a licensing agreement with Rambus in 1998. How many RDRAM/XDR supporting chipsets or CPUs AMD has made since then?


RE: Great News
By tonjohn on 2/24/2006 11:32:07 AM , Rating: 3
quote:
How they could have DDR3 support, when DDR3 specs aren't finalized yet?

You make a very valid and understandable point. However, the spec is nearly finalized and it would not be hard for AMD to build in unofficial support for such a thing just like they did with DDR500 support in the Athlon64s and Intel did with HT in the old P4s.
quote:
Besides, having three different memory controllers on the CPU is just silly.

There wouldn't be three different memory controllers, there would be one memory controller that has support for three different standards. Similar to ATi's and nVidia's memory controllers on their respective GPUs.
quote:
AMD did enter into a licensing agreement with Rambus in 1998. How many RDRAM/XDR supporting chipsets or CPUs AMD has made since then?

Here is specifically what I was referring to:
quote:
Advanced Micro Devices and Rambus Tuesday said the former would pay the latter $75 million as a part of a five-year patent license agreement between the microprocessor maker and technology developer. The agreement includes licensing of memory and interface technologies.
http://www.xbitlabs.com/news/cpu/display/200601030...


RE: Great News
By defter on 2/24/2006 5:11:45 PM , Rating: 3
quote:
You make a very valid and understandable point. However, the spec is nearly finalized and it would not be hard for AMD to build in unofficial support for such a thing just like they did with DDR500 support in the Athlon64s and Intel did with HT in the old P4s.


The difference is that support for DDR500 was just a support for different memory dividers. DDR3 chip spec isn't finished and AFAIK spec for DDR3 modules is far from being finished. Thus AMD simply cannot make currently DDR3 controller that is 100% guaranteed to work with DDR3 modules that will be released in 2007/2008.

quote:
Similar to ATi's and nVidia's memory controllers on their respective GPUs.


ATi and nVidia don't have memory controllers that support three completely different standards. Supporting GDDR1/2/3 is quite easy, GDDR2 is backwards compatible and GDDR3 is just a slight modification, all these memory technologies are parallel in nature, have same bus width, etc..

What you are proposing is that AMD will support serial, narrow (16bit?) and fast (clocked at several GHzs) memory technology (FB-DIMM, XDR) and parallel, wide (128bit) and slow (clocked below 1GHz) memory technology (DDR2, DDR3) with the same memory controller. This is just impossible. They would need to implement different memory controllers, reserve different pins to them etc. And besides, why would you use XDR on Socket-F? It definately isn't designed for server use.

quote:
Here is specifically what I was referring to


Here is what I was referring to:
http://www.amd.com/us-en/Corporate/VirtualPressRoo...

"AMD Chooses Direct Rambus Interface Technology for Processor Chipsets

--First Test Chips Are Already Functional at Full Speed--

SUNNYVALE, CA -- October 8, 1998 --AMD announced today it has licensed the Direct Rambus™ high-bandwidth memory interface for use in forthcoming logic chips, and that the technology will be the main memory interface for future personal computer products. Test chips for the company's initial program are functional at full-speed, 800 MHz operation."

How many AMD's RDRAM chipsets made to the shops?


RE: Great News
By themelon on 2/24/2006 4:59:44 AM , Rating: 2
They only support Registered, ECC DDR2 modules. I know this for a fact.


RE: Great News
By Viditor on 2/24/2006 6:57:38 AM , Rating: 2
quote:
AMD recently entered into a licensing agreement with Rambus which would allow them to have a some pretty robust memmory controllers


Rambus has claimed ownership of a portion of FBDIMM IP. While AMD had an agreement with Rambus prior to this, it didn't include the licensing for DDR/DDR2. However, most of the lawyer types have been saying that Rambus' claim on FBD is much stronger, so it's probably a good bet that Socket F will be FBD.


I'd rather bend cpu pin
By bwave on 2/24/2006 12:57:31 AM , Rating: 2
Am I the only one who hates the LGA Socket - from the article - "Intel adopted the LGA socket design with Prescott to reduce cost and also to reduce the defect rate on processors. After all, would you rather scrap a $600 CPU for a bent pin, or a $100 motherboard for a bent land grid?"

Umm, ok first of all, where are you buying a $100 quad-core opteron board from? These boards are going to be in servers. Let's say you go to upgrade the cpu and you bend a pin on the motherboard - then what?? That board is probably discontinued by now - so have to get a different one. You replace the board and you'll screw up Windows and if you're lucky, maybe you can rebuild. Maybe not - maybe have to format - hope this isn't an important server. Versus, you bend pin on new cpu, you just put the old back in and back up and running until you get a replacement cpu. You'll always be able to get a cpu, not a motherboard. Not to mention, how long is it going to take for you to unplug everything and rip out the old board and put in a new one?




RE: I'd rather bend cpu pin
By Furen on 2/24/2006 1:04:58 AM , Rating: 3
Well, the motherboard is pretty much fixed whenever you install a CPU, so having it fall over and the like just doesn't happen. I've lost count of how many CPUs have been ripped from their ZIF socket because the thermal compound hardened, and this almost always leads to lots of bent pins, as does dropping the CPU. People handle CPUs rather roughly and having 1207 pins to bend would be pretty bad. With an LGA setup the socket itself protects the pins to a major extent.


RE: I'd rather bend cpu pin
By Alphafox78 on 2/24/2006 12:46:31 PM , Rating: 2
How does hardened termal compound rip a cpu from its zif socket?! I guess I should check to see if my compound is hardened, I wouldnt want the CPU to get riped out!!


RE: I'd rather bend cpu pin
By Furen on 2/24/2006 3:25:58 PM , Rating: 1
...
The hardened thermal compount rips the CPU out of the socket when you try to remove the heatsink, that's why people recommend "warming up" the CPU and giving the heatsink a little twist when removing it. Do you understand or do I need to draw it for you?


By NullSubroutine on 3/2/2006 6:04:14 PM , Rating: 2
ive pulled my 754 newcastle out of its place several times, still works like a pretty penny.


RE: I'd rather bend cpu pin
By defter on 2/24/2006 1:54:31 AM , Rating: 2
"Umm, ok first of all, where are you buying a $100 quad-core opteron board from?"

Where are you buying a $600 quad-core opteron CPU from?

I don't see any reason why these Socket-F boards should be significantly more expensive than existing Socket-940 boards with the same amount of sockets. Fast Socket-F CPUs will still be several times more expensive than Socket-F boards.


RE: I'd rather bend cpu pin
By Assimilator87 on 2/24/2006 5:17:17 PM , Rating: 2
You guys are missing the actual reason Intel switched to the LGA socket. They wanted to decrease the thermals, although I have no idea how the difference in socket could possibly keep the temperature lower, not to lessen the hazard of bending a pin. Think about it in a capitalistic way. It makes no sense for a company to decrease the risk of a costumer having to buy another of its products. Intel and AMD would love for you to accidentally break a pin on your Extreme Edition or FX that you just blew a grand on.


RE: I'd rather bend cpu pin
By PT2006 on 2/24/2006 5:56:08 PM , Rating: 2
No, I think the article has it dead on. read what this guy wrote from the other article thread

http://www.dailytech.com/article.aspx?newsid=946&c...


socket F
By MDme on 2/23/2006 5:47:02 PM , Rating: 2
i have a feeling socket F stabds for F B-DIMM support. don't ya think?




RE: socket F
By Von Matrices on 2/23/2006 6:12:42 PM , Rating: 4
Of course, it's also revision F of the Opteron.


RE: socket F
By Lifted on 2/23/06, Rating: -1
RE: socket F
By JackPack on 2/23/2006 9:23:59 PM , Rating: 2
Keep those thoughts to yourself.

Most people here are heterosexual.


RE: socket F
By GoatMonkey on 2/24/2006 8:13:58 AM , Rating: 2
Does that make it a "FO"?



RE: socket F
By Griswold on 2/27/2006 5:00:12 PM , Rating: 2
It could also stand for F -word as in F 'in F ast.


RE: socket F
By johnford64 on 3/2/2006 3:55:34 PM , Rating: 2
Or maybe it stands for Fully Buffered DIMM! Why was intels S775 called Socket T. Who cares? Does it effect the actual technology?


More Likely To be Damaged =\
By BillyBatson on 2/24/2006 3:05:28 AM , Rating: 2
quote:
After all, would you rather scrap a $600 CPU for a bent pin, or a $100 motherboard for a bent land grid?


I understand that this reduces the cost for cpu makers. I understand that this increases the cost of mobo's. I also understand that IF a pin or land grid is to be damaged it would be better if it was a land grid on a mobo... HOWEVER in my experience, and from first accounts i have hear, it is far more likely for a Land Grid to be damaged than a pin increasing your odds of damaging something rather than nothing... especially fi you remove and reinstall your cpu often.




RE: More Likely To be Damaged =\
By Questar on 2/24/2006 9:15:29 AM , Rating: 2
Hardly. The whole point is that LGA is LESS likely to be damaged during multiple insertions.


RE: More Likely To be Damaged =\
By TheLiberalTruth on 2/24/2006 5:31:32 PM , Rating: 3
You must realize that as you increase the number of pins, each pin becomes smaller, and as the pins decrease in size, they also decrease in strength, making it MORE likely that the pins will be bent than a land grid. Try taking a pre-Pentium processor (skt 486, 1, 2, 3) and bend one of their pins. The things are like small tree trunks. Then take a modern CPU, ie: skt 478, 754, 939 and bend one of their pins. All it takes is a light touch and it bends.


RE: More Likely To be Damaged =\
By Chuckles on 2/27/2006 10:02:13 AM , Rating: 2
I don't know about the old 80x86's, but I have found it easier to bend the pins on old 68020's, 40's and original Pentium's than it is to bend the pins on an Athlon or 74XX Daughter-cards. I would attribute this due to the height of the pins (bending resistance goes like 1/L^3) and the fact that the pins are much more closely packed on modern processors.
Of course, it's that much more of a pain to bend a bent pin back into proper orientation on a modern processor.


By Shining Arcanine on 2/26/2006 2:00:00 PM , Rating: 2
Currently when you take a heatsink off a motherboard, the CPU tends to come with it. That cannot be very good for the pins' longetivity. With this new land grid design, the CPU is locked into the socket meaning that it will not come with the heatsink, so the pins have a higher longetivity with the new LGA design and it is not the end of the world like everyone claims.


What about AM2?
By Fenixgoon on 2/23/2006 5:11:50 PM , Rating: 2
Is AM2 going to be more of a bride like S754 was? I know Opterons are meant for servers, but a whole new setup might be far more beneficial, performance wise, than AM2, right?




RE: What about AM2?
By JackPack on 2/23/2006 5:23:08 PM , Rating: 2
Bride? Not sure what you mean.

AM2 is supposed to consolidate the consumer desktop socket. Sempron and Athlon will jump to AM2.

Socket F is supposed to support quad-core as well as move the PCIe tunnel to the CPU.


RE: What about AM2?
By Furen on 2/23/2006 6:17:15 PM , Rating: 2
I don't know about a PCIe tunnel (since it hasn't been confirmed) but I do know that the width (or perhaps frequency) of the HT links will be increased and there will be more HT links to allow up to 32-way (socket) glueless designs. I seriously doubt that Athlons will migrate to this socket any time soon because the packaging should be quite a bit more expensive. I'd assume we'll stick to socket AM2 until AMD decides to go to DDR3 two or three years.


RE: What about AM2?
By Fenixgoon on 2/23/2006 8:58:08 PM , Rating: 2
*bridge*.. blasted G key :P


Socket S1
By mlittl3 on 2/24/2006 2:19:04 PM , Rating: 2
Doesnt' anyone find it interesting that the reported pin count for AMD Socket S1 for mobile processors is almost exactly half the pin count of Socket F.




RE: Socket S1
By Furen on 2/24/2006 3:30:59 PM , Rating: 4
638 pins is quite a bit more than "half the pin count of Socket F". I'd guess that not having to have any coherent HT links and a much lower power envelope allows them to drop a lot of the pins. After all, S1 only needs pins for two DDR2 channels, a single non-coherent HT link and power/identification.


RE: Socket S1
By Shining Arcanine on 2/26/2006 2:01:21 PM , Rating: 2
I thought the HT link took care of the DDR2 channels.


AM2 LGA
By rqle on 2/23/2006 6:51:32 PM , Rating: 2
guess ill just wait for AM2 LGA type socket before an upgrade. 939 w/ pci-e with suit me for the time being




RE: AM2 LGA
By Xenoterranos on 2/23/2006 9:20:46 PM , Rating: 2
Loosers! I adopted s754 early on so I wouldn't need to upgrade evAr again!! Ha PWND!!

*sigh*

This better be the only damned socket until Athlons have 16 cores and make ommlets.


RE: AM2 LGA
By Fenixgoon on 2/23/2006 11:38:20 PM , Rating: 2
i just upgraded to 754 for socketA :D

i still love my OC'ed sempron though :)


$100 opteron motherboards?
By cocoviper on 3/1/2006 4:41:03 PM , Rating: 2
quote:
After all, would you rather scrap a $600 CPU for a bent pin, or a $100 motherboard for a bent land grid?


Anyone running an Opteron is not going to be running a $100 motherboard...

1 and 2 CPU capable opteron boards right now easily average in the $300 range. But whateva




RE: $100 opteron motherboards?
By johnford64 on 3/2/2006 4:01:03 PM , Rating: 2
uhhh, yes you can! Unless you have been living under a rock for quite sometime, you can get a Dual Core Opteron 185 an put it in a $60 motherboard if you so choose, Its called Socket 939.


"People Don't Respect Confidentiality in This Industry" -- Sony Computer Entertainment of America President and CEO Jack Tretton

DailyTech Poll
Which web browser do you use on your primary personal machine? 






44 Comments









botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki