AMD Launches Six-Core Istanbul Opteron Processor

[malware]

AMD today announced availability of the world's first six-core server processor with Direct Connect Architecture for two-, four- and eight-socket servers. Six-Core AMD Opteron processors (code-named "Istanbul") extend AMD's commitment to offering server customers superior value at every price point with unmatched platform flexibility.
Across a single platform, AMD can address the need for more cores and greater scalability with the new Six-Core AMD Opteron processor and offer a cost- and power-efficient solution with Quad-Core AMD Opteron processors. Systems based on Six-Core AMD Opteron processors are expected to be available beginning this month from leading OEMs including Cray, Dell, HP, IBM and Sun Microsystems, along with support from motherboard and infrastructure partners. HE, SE and EE versions of the Six-Core AMD Opteron processor are planned for the second half of 2009.

• Six-Core AMD Opteron processors leverage existing platform infrastructure and a low-cost, power-efficient DDR-2 memory architecture which can help lower system acquisition costs;
• HPC, virtualization and database workloads can benefit from increased 4P STREAM memory bandwidth of up to 60 percent enabled by HyperTransport technology HT Assist, which helps reduce processor to processor latency and traffic;
• AMD Virtualization (AMD-V) technology and the AMD-P suite of power management features are available across all performance and power bands, ensuring no-compromise choice;
• The new Six-Core AMD Opteron processor has up to 34 percent more performance-per-watt over the previous generation quad-core processors in the exact same platform. (AMD Opteron processor Model 2435 [SPECpower_ssj 2008 overall 1297 ssj_ops/watt, 501,246 ssj_ops at 287W @ 100% target load] compared to AMD Opteron processor Model 2382 [SPECpower_ssj 2008 overall 970 ssj_ops/watt,376,878 ssj_ops at 272W @ 100% target load])

"Based on close collaboration with our customers, we believe there is a clear value shift changing the economics of the server market," said Patrick Patla, vice president and general manager, Server and Workstation Business, AMD. "The new Six-Core AMD Opteron processor meets the increasing need for a combination of low total cost of ownership, superior performance-per-watt and scalability. Simply put, Six-Core AMD Opteron processors deliver top-line performance that's bottom-line efficient."

View at TechPowerUp Main Site

 

[ShadowFold]

Let's see these go AM3, AMD! You know you want too!

 

[Lionheart]

wow, cant wait to see some benchies.

 

[a111087]

wait, didn't Intel already release 6-core Xeon a few months back?
http://arstechnica.com/hardware/news/2008/09/intel-launches-six-core-xeon-monster-chip.ars

so, calling it "the world’s first six-core server processor" is a bit ... strange. but i guess its like the deal with any other article on techpowerup. we just copy them from other sites that copy it from press kits

 

[HolyCow02]

wait, didn't Intel already release 6-core Xeon a few months back?

I remember seeing something about 6-core xeon's, but not sure if it was that they were released. I though it was just an announcement that they were coming soon. I could be completely wrong though

Anyways, good to see this stuff coming out. Though I feel like the amount of cores per die is gonna get obnoxious, and they are going to abandon creating new architectures. I hope I'm wrong :shadedshu

 

[Disparia]

Need to read the whole thing,

AMD today announced availability of the world’s first six-core server processor with Direct Connect Architecture for two-, four- and eight-socket servers

Dunnington uses an old school style bus system

 

[extrasalty]

I can't wait for some 5 core chips

 

[Valdez]

http://products.amd.com/en-us/OpteronCPUSideBySide.aspx?id=550&id=552&id=553&id=554&id=555

 

[a111087]

http://products.amd.com/en-us/OpteronCPUSideBySide.aspx?id=550&id=552&id=553&id=554&id=555

wow! 75W?

 

[yogurt_21]

hmm I wonder how it performs. a cheaper six core in the server market would be welcome.

 

[Kitkat]

wow! 75W?

yeah thats prety impressive. Im sure thell get it lower too lol

 

[suraswami]

http://products.amd.com/en-us/OpteronCPUSideBySide.aspx?id=550&id=552&id=553&id=554&id=555

thanks.

but 1.3v is bit on the higher side. would be nice to see it 1.1v and around 45w range.

is this native 6 core or sticky bubble gum thing?

 

[Valdez]

wow! 75W?

Yeah that's low, and it got a new stepping, D0.

 

[TheLaughingMan]

Real Men use Real cores

thanks.

but 1.3v is bit on the higher side. would be nice to see it 1.1v and around 45w range.

is this native 6 core or sticky bubble gum thing?

only Intel does the sticky bubble gum thing. All AMD processors were true X core processors. So if they say it is a 6 core, it is 6 cores on a single die.

 

[FordGT90Concept]

It doesn't matter. The difference between MCM and not can only be measured in a lab. AMD shot themselves in the foot with Phenom because they didn't do MCM. Intel dominated that market for over a year because AMD was advertising until they were blue in the face that MCM was inferior. If they did, everyone would die of laughter at AMD (crappy product + do it Intel's way).

A hexacore is a hexacore is a hexacore.

 

[suraswami]

only Intel does the sticky bubble gum thing. All AMD processors were true X core processors. So if they say it is a 6 core, it is 6 cores on a single die.

Thanks I was just confirming it.

 

[TheLaughingMan]

Yes, sorta

It doesn't matter. The difference between MCM and not can only be measured in a lab. AMD shot themselves in the foot with Phenom because they didn't do MCM. Intel dominated that market for over a year because AMD was advertising until they were blue in the face that MCM was inferior.

A hexacore is a hexacore is a hexacore.

Actually, the MCM thing did exactly what was expected, it created a heat issue. This was the primary reason those old Pentium D's had a high TDW, larger cooler, and overheating issues. All of these issues were resolved with the Core 2 Duo....being on one die.

But I will give you one thing, if it works well enough to not make people go, "WTF?" then do it. Granted it would save money to wait until the die shrinks to pull off core increases, which both AMD and Intel do now. That old sandwich trick will never be implemented again.

 

[FordGT90Concept]

Pentium 4/D = Netburst
Core 2 = P6

The Netburst architecture is what made it hot. You see the same thing in Core i7 (loaded stock temp can hit 85C on stock HSF) compared to Core 2 (never goes over 60C on stock HSF). It had absolutely nothing to do with MCM. If you double the number of cores, you usually see a 90% increase in power draw and thermal design requirements.

Core 2 Quad is about 90% hotter than Core 2 Duo because it has twice the number of cores. MCM or not, that's what happens when you add more transistors.


The "sandwich trick" is being used on Core i5 to add the integrated graphics chip. I do believe AMD is doing the same with Bulldozer. MCM, therefore, is likely to become more common than it was in the past.


The thermal threshold is what prevents 100+ cores MCM chips from being released now. You'd need a whole room of cooling equipment dedicated to pulling heat off it and dissipating it. 45nm looks primed for up to 8 cores but we probably won't see much more than that until 32nm.

 

[eidairaman1]

actually its inaccurate, Core 2 Duo had a single Die where Core 2 Quad has 2.

 

[FordGT90Concept]

To clarify:

Pentium 4 = not MCM
Pentium D = MCM

Core 2 Duo = not MCM
Core 2 Quad = MCM

Core i7 = not MCM

 

[[I.R.A]_FBi]

Actually, the MCM thing did exactly what was expected, it created a heat issue. This was the primary reason those old Pentium D's had a high TDW, larger cooler, and overheating issues. All of these issues were resolved with the Core 2 Duo....being on one die.

But I will give you one thing, if it works well enough to not make people go, "WTF?" then do it. Granted it would save money to wait until the die shrinks to pull off core increases, which both AMD and Intel do now. That old sandwich trick will never be implemented again.

....:shadedshu

 

[FordGT90Concept]

Come to think of it, it is actually easier to cool a MCM chip. Like on the Core 2 Quad, the two chips are pretty far apart:
http://xtreview.com/images/core 2 Quad q9100 and Core 2 Extreme QX9300.jpg

The heat is better distributed across the IHS (two smaller hotspots instead of one large one) and then the heatsink. It probably only results in 1-3C gain but, the more MCMs you have, the more it helps.

 

[Wile E]

It doesn't matter. The difference between MCM and not can only be measured in a lab. AMD shot themselves in the foot with Phenom because they didn't do MCM. Intel dominated that market for over a year because AMD was advertising until they were blue in the face that MCM was inferior. If they did, everyone would die of laughter at AMD (crappy product + do it Intel's way).

A hexacore is a hexacore is a hexacore.

Actually, the MCM thing did exactly what was expected, it created a heat issue. This was the primary reason those old Pentium D's had a high TDW, larger cooler, and overheating issues. All of these issues were resolved with the Core 2 Duo....being on one die.

But I will give you one thing, if it works well enough to not make people go, "WTF?" then do it. Granted it would save money to wait until the die shrinks to pull off core increases, which both AMD and Intel do now. That old sandwich trick will never be implemented again.

Pentium 4/D = Netburst
Core 2 = P6

The Netburst architecture is what made it hot. You see the same thing in Core i7 (loaded stock temp can hit 85C on stock HSF) compared to Core 2 (never goes over 60C on stock HSF). It had absolutely nothing to do with MCM. If you double the number of cores, you usually see a 90% increase in power draw and thermal design requirements.

Core 2 Quad is about 90% hotter than Core 2 Duo because it has twice the number of cores. MCM or not, that's what happens when you add more transistors.


The "sandwich trick" is being used on Core i5 to add the integrated graphics chip. I do believe AMD is doing the same with Bulldozer. MCM, therefore, is likely to become more common than it was in the past.


The thermal threshold is what prevents 100+ cores MCM chips from being released now. You'd need a whole room of cooling equipment dedicated to pulling heat off it and dissipating it. 45nm looks primed for up to 8 cores but we probably won't see much more than that until 32nm.

To clarify:

Pentium 4 = not MCM
Pentium D = MCM

Core 2 Duo = not MCM
Core 2 Quad = MCM

Core i7 = not MCM

Come to think of it, it is actually easier to cool a MCM chip. Like on the Core 2 Quad, the two chips are pretty far apart:
http://xtreview.com/images/core 2 Quad q9100 and Core 2 Extreme QX9300.jpg

The heat is better distributed across the IHS (two smaller hotspots instead of one large one) and then the heatsink. It probably only results in 1-3C gain but, the more MCMs you have, the more it helps.

To summarize: The methods of design do not matter. Only the end results and performance matters.

 

[wojo]

There is a new 8 core being worked on in a new chip from china.
http://www.techpowerup.com/70356/China_to_Launch_its_Quad-Core_CPU_in_2009.html
I dont know if it is out yet, also I wounder if it comes with built in spyware.

 

[FordGT90Concept]

To summarize: The methods of design do not matter. Only the end results and performance matters.

Yup, pretty much. MCM or not really doesn't make a major impact on performance at the end of the day.

There is a new 8 core being worked on in a new chip from china.
http://www.techpowerup.com/70356/China_to_Launch_its_Quad-Core_CPU_in_2009.html
I dont know if it is out yet, also I wounder if it comes with built in spyware.

Those cores sound extremely simple almost like an SPE from a Cell processor. You're right on spyware too. Those processors might be part of a Chinese dragnet to control content. I see no other reason why China would see it as a worthy investment.