Tuesday, February 23rd 2010

AMD Starts Shipping 12-core and 8-core ''Magny Cours'' Opteron Processors

AMD has started shipping its 8-core and 12-core "Magny Cours" Opteron processors for sockets G34 (2P-4P+), and C32 (1P-2P). The processors mark entry of several new technologies for AMD, such as a multi-chip module (MCM) approach towards increasing the processor's resources without having to complicate chip design any further than improving on those of the Shanghai and Istanbul. The new Opteron chips further make use of third-generation HyperTransport interconnect technologies for 6.4 GT/s interconnect speeds between the processor and host, and between processors on multi-socket configurations. It also embraces the Registered DDR3 memory technology. Each processor addresses memory over up to four independent (unganged) memory channels. Technologies such as HT Assist improve inter-silicon bandwidth on the MCMs. The processors further benefit from 12 MB of L3 caches on board, and 512 KB of dedicated L2 caches per processor core.

In the company's blog, the Director of Product Marketing for Server/Workstation products, John Fruehe, writes "Production began last month and our OEM partners have been receiving production parts this month." The new processors come in G34/C32 packages (1974-pin land-grid array). There are two product lines: the 1P/2P capable (cheaper) Opteron 4000 series, and 2P to 4P capable Opteron 6000 series. There are a total of 18 SKUs AMD has planned some of these are listed as followed, with OEM prices in EUR:
  • Opteron 6128 (8 cores) | 1.5 GHz | 12MB L3 cache | 115W TDP - 253.49 Euro
  • Opteron 6134 (8 cores) | 1.7 GHz | 12MB L3 cache | 115W TDP - 489 Euro
  • Opteron 6136 (8 cores) | 2.4 GHz | 12MB L3 cache | 115W TDP - 692 Euro
  • Opteron 6168 (12 cores) | 1.9 GHz | 12MB L3 cache | 115W TDP - 692 Euro
  • Opteron 6172 (12 cores) | 2.1 GHz | 12MB L3 cache | 115W TDP - 917 Euro
  • Opteron 6174 (12 cores) | 2.2 GHz | 12MB L3 cache | 115W TDP - 1,078 Euro
Sources: AMD Blogs, TechConnect Magazine
Add your own comment

125 Comments on AMD Starts Shipping 12-core and 8-core ''Magny Cours'' Opteron Processors

#1
fatguy1992
Where can I buy one of those G34 motherboards?
Posted on Reply
#2
eidairaman1
The Exiled Airman
mdm-adph said:
Well... damn. That was quick.

I've been saying for years that, AMD fan or no, when Intel released their first true-eight core 32nm chip, I'd go Intel, but I think now I'd rather stick with AMD and pick up one of those 12-core beauties when they make it to the desktop market.
imagine the 4 way or 8 way 12Core, Intel is talking about a 48 Core CPU, well AMD will meet the market with 4 of these in 4 way. and then 8x8 way would be 64 cores. Cool my computer has 48 cores in it.
Posted on Reply
#3
TIGR
I wonder if we'll see eight-CPU systems at the consumer level as eidairaman1 suggested. It seems more likely those 48/64 cores will be in a single CPU (or by that time, GPGPU), aside from server-class configurations like the Tyan Thunder n4250QE motherboard plus the M4985 expansion board (eight CPUs total). Yeah, enthusiasts might go with some of those four/eight-CPU rigs but the mainstream doesn't tend to adopt technology until it fits in the same size packages its predecessors did.
Posted on Reply
#4
eidairaman1
The Exiled Airman
TIGR said:
I wonder if we'll see eight-CPU systems at the consumer level as eidairaman1 suggested. It seems more likely those 48/64 cores will be in a single CPU (or by that time, GPGPU), aside from server-class configurations like the Tyan Thunder n4250QE motherboard plus the M4985 expansion board (eight CPUs total). Yeah, enthusiasts might go with some of those four/eight-CPU rigs but the mainstream doesn't tend to adopt technology until it fits in the same size packages its predecessors did.
what i meant to say was a poke at Intels little article about 48core CPus, to me that is already possible to have 48 cores in a machine in a 4 way Motherboard
Posted on Reply
#5
pantherx12
Yeah but not as convenient as just one actual CPU with 48 cores.


There's a company with 100 core CPUs already clocked at 1.8ghz each : /

Once they've sorted out how an OS can actually interact with such a CPU that will be fun and games : ]
Posted on Reply
#6
DanTheBanjoman
Señor Moderator
pantherx12 said:
Yeah but not as convenient as just one actual CPU with 48 cores.


There's a company with 100 core CPUs already clocked at 1.8ghz each : /

Once they've sorted out how an OS can actually interact with such a CPU that will be fun and games : ]
Tilera, but since that's not x86 it's apples and oranges. OSes can handle such amounts just fine.
Posted on Reply
#7
eidairaman1
The Exiled Airman
ya fact of those 1.8 units are Itaniums which are i64 and not x86 compatible (gotta run anything x86 in a virtual environment which then hinders performance and fact of those Itaniums run better in cluster environments)
Posted on Reply
#8
pantherx12
Would work great in a i-phone like device XD

( obviously not the 100 core version, I imagine heat would melt everything ha ha)
Posted on Reply
#9
CDdude55
Crazy 4 TPU!!!
Not buying it.
Posted on Reply
#11
eidairaman1
The Exiled Airman
CDdude55 said:
Not buying it.
no ones forcing ya to so.
Posted on Reply
#12
CDdude55
Crazy 4 TPU!!!
eidairaman1 said:
no ones forcing ya to so.
Yep.

No money for this crap anymore. I'm gonna ride my i7 rig for years then pick up whatever Console is out at the time.

So i'm out.*throws cards down on table*
Posted on Reply
#13
Tartaros
r9 said:
I`m just going to the thread and funny enough I have noticed that no one wondered how it would run Crysis :D:D:D:D:D. So I wonder :D. And I remember a time not so long ago AMD fanboys sayng "AMD would never charge 1000 eur for CPU". Yeah right.
You could buy 4 years ago an opteron 185 for 500 euros or athlon fx 60 (same as opteron but unlocked multiplier) for 1000... amd also sold their cpus at astronomic prices.
Posted on Reply
#14
aj28
[Ion] said:
I'm not really convinced the lower-clocked (sub-2ghz) 8-cores are really all that good. A Phenom II X4 965 would probably be better than the 6128 and 6134
I don't know about this actually. I think people need to look at a product like this and, rather than seeing a 48-core server and wondering "I wonder how fast it can do X-single-operation," they ought to be wondering about how you can split up that workload and accomplish more instances of a single operation (or a variety of different operations) that will run well on just one or two cores. What used to take twenty-four servers towards the beginning of the decade now takes one, with twenty-four completely independent virtualized dual-core systems within!

Mind you, there's plenty of overhead associated with such an operation, but I think the point stands. In my mind, AMD will do well with these... They're no speed demon, but that's not what the industry is looking for right now.
Posted on Reply
#15
Jizzler
fatguy1992 said:
Where can I buy one of those G34 motherboards?
No boards are available right now. That's what I was saying before, haven't seen releases from any of the manufacturers yet.

The H8SGL-F that I posted was displayed at Supermicro's SC09 booth.
Posted on Reply
#16
Melvis
Is the 8 core a 12 core with 4 disabled cores? or two 4 cores sticky taped together?
Posted on Reply
#17
troyrae360
Melvis said:
Is the 8 core a 12 core with 4 disabled cores? or two 4 cores sticky taped together?
I was under the understanding that that the 12 core was 2x 6 core cpu's "sticky taped" I could be wrong though
Posted on Reply
#18
btarunr
Editor & Senior Moderator
Melvis said:
Is the 8 core a 12 core with 4 disabled cores? or two 4 cores sticky taped together?
8 core is two Shanghai-derived dies on MCM. 12 core is two Istanbul-derived ones. The difference between AMD's MCM and Intel's traditional MCM designs is that each die on AMD's MCM has its own memory controllers, and independent HyperTransport links to the system, and to each other. The cores on each die (node) can address memory controlled by the neighbouring die.
Posted on Reply
#19
MadClown
Not enough cores for my liking. Want moar!
Posted on Reply
#20
Athlonite
sticky tape WHO uses sticky tape anymore its all crazy glue or nothin at all. All we need now is a board with 8 pcie x16 slots 128 lanes (gpgpu) and four x4/x1 slots (raid SAS/SSD/10Gbit network) please
Posted on Reply
#21
FordGT90Concept
"I go fast!1!11!1!"
This is getting ridiculous. Most applications aren't very good candidates for multithreading so more per-core performance is still ideal. Someone has to change this trend of gluing more cores on to more core performance. Multiple cores create needless overhead and before long, applications will be slower tomorrow than they are today because overhead exceeds actual work done.
Posted on Reply
#22
aj28
FordGT90Concept said:
This is getting ridiculous. Most applications aren't very good candidates for multithreading so more per-core performance is still ideal. Someone has to change this trend of gluing more cores on to more core performance. Multiple cores create needless overhead and before long, applications will be slower tomorrow than they are today because overhead exceeds actual work done.
Well, like I was saying earlier, I see these primarily as parts for virtualization more than anything else, and I think they will do that job quite well. Plenty of independent cores to work on. Of course there's not much clock speed, but depending on the application that may not be necessary. After all, there's a reason these things start out on servers... I don't see anything beyond the X6 for at least another few years on the desktop front. Not from AMD at least...
Posted on Reply
#23
a_ump
yea but i'm assuming it'd take a hell of a lot more time and money to creat quad cores that run at 4.5ghz stock n oc to 6ghz on air than it would doing the MCM deal.

And with btarunr's statement wouldn't the overhead that was in intel's MCM design be non existant or dam near gone?
btarunr said:
8 core is two Shanghai-derived dies on MCM. 12 core is two Istanbul-derived ones. The difference between AMD's MCM and Intel's traditional MCM designs is that each die on AMD's MCM has its own memory controllers, and independent HyperTransport links to the system, and to each other. The cores on each die (node) can address memory controlled by the neighbouring die.
Posted on Reply
#24
FordGT90Concept
"I go fast!1!11!1!"
Overhead is created by programs to manage multiple threads synchronous or asynchronously. Synchronous creates more overhead than async because all threads have to halt until ordered to move on to the next set of work (like games). The huge wall programmers are going to hit sooner rather than later is the core that is managing all the threads will get overburdened which in turn, creates an uncloggable roadblock. Every core is waiting for that one core to tell it what to do and that one core falls behind leading to huge problem.

Basically, this fad of adding more cores, if it lasts too long, will be bad for developers and consumers. Yes, it's nice to have extra cores to offload work but that doesn't change the fact a 12 GHz CPU can handle more work than a 4 x 3 GHz CPU because of having no overhead.

In order to maintain Moore's Law much longer, semiconductor has to leap ahead of where it is today.


In terms of MCM, it really doesn't matter as demonstrated by Phenom compared to Core 2 Quad.
Posted on Reply
#25
TIGR
FordGT90Concept said:
This is getting ridiculous. Most applications aren't very good candidates for multithreading so more per-core performance is still ideal. Someone has to change this trend of gluing more cores on to more core performance. Multiple cores create needless overhead and before long, applications will be slower tomorrow than they are today because overhead exceeds actual work done.
While multi-core scaling isn't perfect , it's a necessary step to overcome the fact that a single core has its limits. Diminishing performance gains from clock rate increases, exponentially increasing power consumption for each factorial increase in operating frequency, ILP and memory walls, and simple limits to how well a single core can be designed, force us to use multiple cores to keep up the exponential rate of progress that has come to be expected from chip makers.

The human brain (the most powerful computer known) is massively parallel.

Improving per-core performance is still extremely important, and I don't think AMD or Intel are abandoning that in favor of just increasing core count. Look at the per-core difference between C2D and Core lines of CPUs.

Anyway, mainstream multi-core computing is still in its infancy. The main issue seems to be software algorithms and implementation, not some flaw with the concept of multiple CPU cores itself. There will be challenges in the future, such as the jump from multi-core to many-core CPUs, but I see no signs that multi-core computing is a dead end.
Posted on Reply
Add your own comment