12 Core AMD Opteron CPU by March 2010

http://www.xbitlabs.com/news/cpu/display/20091215170147_AMD_to_Launch_12_Core_Server_Microprocessors_in_March_Sources.html

Magny-Cours microprocessors feature two six-core or quad-core dies on one piece of substrate
- was it AMD not long ago having a go at Intel for esentially "duck taping" two cores to make one multicore'd chip? and the large "monolithic is better" debate...

Interesting times ahead, wouldnt mind a rig with 12 real cores, wonder how this stacks up agains the up coming xeon version of the i7 980X, and how well these babys scale etc.
 
Solution
69%20Tallest%20Hamburger.jpg
I love Hamburgers :bounce:

B-Unit

Distinguished
Oct 13, 2006
1,837
1
19,810
I love how everyone thinks that its as simple as walk into the CPU Design dept and tell them we want 5Ghz at 65W TDP and they'll give it to you next week. It takes YEARS to develop new architectures, what we have now is it untill the next iteration. All they can do at this point is add cores.
 

smithereen

Distinguished
Oct 4, 2008
1,088
0
19,310


I don't imagine servers are going to find them "friggin useless". In fact, I'm sure the only thing it will be "friggin useless" in is gaming. And let's think, what are AMD's quads "not" good at right now?

Hint: Desktop hexas and server octas and dodecas are a good thing.
 


The Magny-Cours is a server CPU that runs in 2- and 4-socket motherboards. Units like these are what goes into medium-sized servers. Servers typically run very highly-multithreaded applications and more cores yields more performance than more GHz, which is why this particular chip has gotten a lot of press. Power consumption of these units will actually be pretty low as AMD says that the Magny-Cours will have similar thermal ratings as the existing Opteron HE, standard, and SE chips, namely ~70 watts, ~95 watts, and ~135 watts TDP. Even in the worst case, the 12-core Magny-Cours is rated for a similar thermal dissipation as the Core i7s and the Phenom II X4 BEs that most people here run, but it is powering three times as many cores. Or to put in terms more familiar to you, it consumes about as much power as your E8500 overclocked to 4.50 GHz/1.44 V (which uses approximately 133 W at full load.)
 

AMW1011

Distinguished


I'm just going to ask, what do you mean? Do you mean we need quadcores that are faster clock for clock rather than 6-cores that are similar or only slightly faster per core? If so then I agree, there is no need to have more than 4 cores right now and we need more performance per clock than we need more cores because those extra cores will likely be at 0-10% load most of the time in a desktop environment with the exception of some professional apps. Oh well, these chips are already coming and it is far to late to have any hopes of a huge change in tactic.
 


I think he wants a dual-core CPU that has higher IPC and a higher clock speed than his current E8500.
 

znegval

Distinguished
Dec 15, 2009
99
0
18,630


You do realise this processor aims the server market and not the desktop right?
 


Actually, it *is* simple, as long as you threaten to duck-tape them to their desks until they come up with a proper design :D.

Being hungry, sweaty & fulla B.O. has a way to inspire as well as perspire :p...

That's how I like to run my dept. with now 19 stinkin' employees! :D

/jk
 

B-Unit

Distinguished
Oct 13, 2006
1,837
1
19,810



Fix'd ;)
 




...also, this technique can be leveraged on the business side of the house.

Have the Credit and Collections guys tie debtors downwind until they pay!

When Auditors come to check the books, put them in the conference room next to the data center!
 
This is "duct taping" two dies on one substrate, not "duct taping" cores. Did you even read what you quoted??

AMD has been shipping 6 core processors for 6 months, it's no suprise they are putting 2 together.

Maybe these are the ones with the "hyperthreading" like feature? 24 threads?

Still same thing. Its not on one die.

MCM FTW, or so it seems. AMD is technically eating their words now....
 

someguy7

Distinguished
Dec 12, 2007
1,186
0
19,310
It is a little funny. What is really run is all the AMD fanboys that hated on double cheeseburger. But when AMD does it. They defend it with the logic that AMD's double cheeseburger is better. Like the BK vs mcdonalds double cheeseburger kind of adds.

I'm glad AMD is doing it. It makes perfect sense to do it.
 

kg4icg

Distinguished
Mar 29, 2006
506
0
19,010



To be truthful, It's actually spelled ductape, as in ventilation duct's
 

noob2222

Distinguished
Nov 19, 2007
2,722
0
20,860
duck tape is a brand, duct tape is a product.

Think about why AMD would do two dies instead of one? The larger the die, the more errors it will contain (binning process). Keep the dies small, keep the higher binned products available.

The 2nd advantage is your memory controllers are doubled as well and for servers thats not a bad thing to have.

And to everyone saying how AMD is now doing what they said was using duct tape, Intel did the same exact thing with 64 bit processing. What goes around, comes around.
 


...and either one of them can fix almost anything :D

Think about why AMD would do two dies instead of one? The larger the die, the more errors it will contain (binning process). Keep the dies small, keep the higher binned products available.

The 2nd advantage is your memory controllers are doubled as well and for servers thats not a bad thing to have.

And to everyone saying how AMD is now doing what they said was using duct tape, Intel did the same exact thing with 64 bit processing. What goes around, comes around.

The key difference is that AMD's approach doubles the memory controllers, whereas Intel's approach simply added the second die to the same FSB. AMD's approach results with each die in the MCM getting at least as much bandwidth as it does in the single-die monolithic setup. Under lighter loads, the AMD MCMs actually could get more memory bandwidth than a single monolithic die as the active die can do I/O not only through it's own IMC but through its idle neighbor die's IMC over the HT link. Intel's MCM approach resulted in less memory bandwidth available for each die as the two dies not only fought over memory access over the FSB but they also used that same FSB for die-to-die communication. So under heavy loads, each die got less than half of the bandwidth that a monolithic die would get. Intel would have had to run two independent FSBs to the socket for its MCMs to achieve the same kinds of things that AMD is with its MCMs.
 

brendz155

Distinguished
Dec 12, 2009
388
0
18,790
I think we need at least 2 core processor for all of today's app and programs bit 12 core that's over the top you know i only have sempron that's running overclocked to 3300mhz and its a single core and it dont even lag on most games.

I would love to see in the future a processor that runs a 6ghz
probably when im old processor like that will be in PHONES!