Friday, January 30th 2009

Intel to Unveil Octo-Core Xeon Next Month

Sitting tight on the desktop and server performance thrones with the Core i7 and Xeon Dunnington series procesors, Intel is looking to carry on with its product launch-cycle with the introduction of octo-core (8 cores) enterprise processors later this year. The company is expected to detail the industry about this upcoming processor series as early as next month at the Solid State Circuits conference in San Fransisco between February 8 and 12.

The processor in question will be based on the Intel Nehalem architecture and will consist of eight x86 processing cores, a massive transistor-count of 2.3 billion, and will be built on the company's current 45nm manufacturing process. It will hold 24 MB of L3 cache, a quad-channel memory interface and QuickPath Interconnect system interface. Furthermore, it is aimed at quad-socket server platforms. Intel will keep this only upto a presentation level at the conference and not a launch. It is expected to start off with quad-core Xeon processors based on the new architecture later in this quarter.
Source: TechConnect Magazine
Add your own comment

25 Comments on Intel to Unveil Octo-Core Xeon Next Month

#1
mtosev
waw. what a stupid name "Octo". no comment.
Posted on Reply
#2
Wile E
Power User
mtosevwaw. what a stupid name "Octo". no comment.
Octo means eight. You know, like octogon or octopus.
Posted on Reply
#3
laszlo
this is bad news for amd;this octo cpu will outperform all in perf and i don't think amd will survive more that 2 year from now on...
Posted on Reply
#4
Deleted member 3
Quad socket > FB-DIMM2, they better fixed performance issues FB-DIMMs have or we still have the same problem as with the current Xeons, 8 cores basically being held back by lack of memory bandwidth in many cases. At least they have 4 channels per socket and NUMA, it would be nasty if 32 cores (with htt?) had to share that already limited bandwith :)

Not that it matters much to us, we won't buy quad socket stuff. And the dual socket boards use DDR3 so should actually have a lot of bandwidth due to NUMA.
Posted on Reply
#5
dani31
8 cores x 4 sockets = 32 cores w/ hyperthreading = 64 logical cores in one system. Geez.
Posted on Reply
#6
thebeephaha
If they make an octo core xeon that will work with an x58 chipset on a consumer board like they currently make quad xeons for 775 I will be so happy.
Posted on Reply
#7
Flyordie
Yes, but it will cost Intel ALOT to make...
Intel's CPUs are huge in comparison to AMDs.
AMD's literally cost less cause they use less silicon. So don't expect a simple price tag on the Intel Octo cores. (Especially since they can barely fit it onto their "socket". That little green PCB that they solder the die on) ... it can barely fit on that. AMD can fit up to 16 cores before it becomes an issue @ 45nm.
So I hope Intel comes up with a solution for their problem before AMD releases theirs.
Posted on Reply
#8
Deleted member 3
thebeephahaIf they make an octo core xeon that will work with an x58 chipset on a consumer board like they currently make quad xeons for 775 I will be so happy.
Slim chance, the 6-core ones don't even exist in 771 flavor. So the 8 cores ones will probably remain with the quad socket boards. And since it uses a different socket it won't work on dual or single socket boards. I'm not sure if Xeon DP's will work on single socket boards though, socket seems the same. Perhaps we're going back to the Pentium 3 days.
Posted on Reply
#9
Wile E
Power User
FlyordieYes, but it will cost Intel ALOT to make...
Intel's CPUs are huge in comparison to AMDs.
AMD's literally cost less cause they use less silicon. So don't expect a simple price tag on the Intel Octo cores. (Especially since they can barely fit it onto their "socket". That little green PCB that they solder the die on) ... it can barely fit on that. AMD can fit up to 16 cores before it becomes an issue @ 45nm.
So I hope Intel comes up with a solution for their problem before AMD releases theirs.
That's not true at all. Core i7's die size is 263mm^2 and Phenom II is 258mm^2, so if Intel can't fit 8 cores onto their chip, then AMD won't be able to either.
Posted on Reply
#10
Deleted member 3
FlyordieSo don't expect a simple price tag on the Intel Octo cores.
Completely unrelated to the die size though, these are MP's, they are extremely expensive no matter how small the die or how cheap to produce. These are for large servers, you won't normally even see them in SMB environments.
Posted on Reply
#11
Flyordie
Wile EThat's not true at all. Core i7's die size is 263mm^2 and Phenom II is 258mm^2, so if Intel can't fit 8 cores onto their chip, then AMD won't be able to either.
Yeah, expect that to change. ;-) Only reason PII has a bigger die... memory controller...
Remember, AMD's PII has both DDR2 and DDR3 controllers and Intel's does not. Memory Controllers take up alot of space still. From Nordic Hardware-
"It's probably not a bad idea for AMD to take advantage of their relatively small core size. At 45nm it can cram two cores (with 256 kB L1 total) into 30mm2. A single Nehalem core with 256 kB L2 also occupies ~30 mm2. "

So... I wouldn't say that isn't true at all.
Posted on Reply
#12
Wile E
Power User
FlyordieYeah, expect that to change. ;-) Only reason PII has a bigger die... memory controller...
Remember, AMD's PII has both DDR2 and DDR3 controllers and Intel's does not. Memory Controllers take up alot of space still. From Nordic Hardware-
"It's probably not a bad idea for AMD to take advantage of their relatively small core size. At 45nm it can cram two cores (with 256 kB L1 total) into 30mm2. A single Nehalem core with 256 kB L2 also occupies ~30 mm2. "

So... I wouldn't say that isn't true at all.
The PII 920-940 is DDR2 only. No DDR3 support, afaik. The cpus that offer DDR3 end in 5. Like a few members here have 945 ES chips, which are both DDR2 and 3.
Posted on Reply
#13
hat
Enthusiast
Imagine a supercomputer with like 100,000 of these processors...
Posted on Reply
#14
phanbuey
hatImagine a supercomputer with like 100,000 of these processors...
and a $10,000 a month powerbill :p.
Posted on Reply
#15
btarunr
Editor & Senior Moderator
phanbueyand a $10,000 a month powerbill :p.
That would make it incredibly energy-efficient. $0.10 / month / CPU.
Posted on Reply
#16
rizla1
FlyordieYeah, expect that to change. ;-) Only reason PII has a bigger die... memory controller...
Remember, AMD's PII has both DDR2 and DDR3 controllers and Intel's does not. Memory Controllers take up alot of space still. From Nordic Hardware-
"It's probably not a bad idea for AMD to take advantage of their relatively small core size. At 45nm it can cram two cores (with 256 kB L1 total) into 30mm2. A single Nehalem core with 256 kB L2 also occupies ~30 mm2. "

So... I wouldn't say that isn't true at all.
i7 has a built in mem controller too
Posted on Reply
#17
Disparia
DanTheBanjomanCompletely unrelated to the die size though, these are MP's, they are extremely expensive no matter how small the die or how cheap to produce. These are for large servers, you won't normally even see them in SMB environments.
Oh yeah, I could run the entire company I work for off a decked out 4P/32C/64T Xeon box, and even then it would be a bit overkill.

Only reason we're looking at 3 dual-quad boxes in the near future is for VMware High Availability. Each one with 32GB, 3TB storage, and two mid-range QC Xeons will end up around $4K. That probably wouldn't even buy TWO of these new OC Xeons, to help anyone with perspective on the price difference :D
Posted on Reply
#18
PCpraiser100
laszlothis is bad news for amd;this octo cpu will outperform all in perf and i don't think amd will survive more that 2 year from now on...
Yeah, that is a shame. But don't forget, this is for workstation and business only unless if you dare to pay thousands. All AMD has to do is just get ready to compete with Intel at Price, since the need for these processors are so far insane.
Posted on Reply
#19
lemonadesoda
DanTheBanjomanQuad socket > FB-DIMM2, they better fixed performance issues FB-DIMMs.
This is Beckton on FB-DDR3. But I also heard about the move to ditch FB-DIMM and go with Registered ECC DDR3.

Let's hope it is cooler and faster.
Posted on Reply
#20
PCpraiser100
lemonadesodaThis is Beckton on FB-DDR3. But I also heard about the move to ditch FB-DIMM and go with Registered ECC DDR3.

Let's hope it is cooler and faster.
It is, lower timings.
Posted on Reply
#21
lemonadesoda
Higher clocks, higher speed, higher bandwidth, lower power, but typically HIGHER timings in cycle count. What that means in terms of latecy? We will need to see. DDR2 had higher latency that DDR1 on launch. Probably true on DDR3 too. That's why L3 cache is so important.
Posted on Reply
#22
Easy Rhino
Linux Advocate
im assuming something like this will work best with a ddr 4 standard. clearly the bus speed on the boards needs to speed up first.
Posted on Reply
#23
trickson
OH, I have such a headache
NOW here is some real INNOVATION ! Not just MHz speed no real innovation !
Posted on Reply
#24
FordGT90Concept
"I go fast!1!11!1!"
Easy Rhinoim assuming something like this will work best with a ddr 4 standard. clearly the bus speed on the boards needs to speed up first.
Intel overcomes that with FB-DIMM. FB-DIMM has a memory controller right on the sticks that takes, and fulfills requests given to it. It adds some latency but it vastly decreases the complexity of the processor memory controllers communicating with the memory allowing for much larger pools of memory modules.

In short, FB-DIMM can do more with less. It won't be going anywhere any time soon. I can't see plain DIMMs being used when there's 8 or more slots. There's too many physical connections causing motherboards to get extremely expensive. FB-DIMM = more expensive memory, cheaper motherboard.


L3's have made a come back because the gap between DDR3 performance and L2 is so vast.
Posted on Reply
Add your own comment
Dec 18th, 2024 15:35 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts