nehalem quad/octo

[ilkhan]

Ive heard nehalems will have up to 8 cores per socket.
Is that a native quad with 2 dies (like current quads do with 2x duals) or 8 cores on one die (native octo)? I dont care one way or the other, Im going to grab a quad on bloomfields release day I think, but Im curious.

Also, bloomfield is the nehalem desktop core from what Ive heard. Is that the quad (would follow the x-field scheme) or the octo?

 

[Mystofolyse]

They will be natively quad and octo and are both associated with the Bloomfield codename.

http://en.wikipedia.org/wiki/Nehalem_(CPU_architecture)

Nehalem has the most significant new architectural changes as of today since the Pentium Pro back in 1995. Nehalem is very scalable[citation needed] with different components for different tasks. According to Intel it will have 1 through 8+ cores, integrated memory controllers for DDR3 with 1 to 4 memory channels and be manufactured using the same 45nm manufacturing process as its predecessor, Penryn. Intel announced at the Intel Developer Forum in September 2007 that some versions of Nehalem will also have an integrated graphics subsystem off-die, but in the same CPU package. Nehalem processors will also utilize the new point-to-point processor interconnect, the Intel QuickPath Interconnect, replacing the legacy front side bus.

Nehalem will be launched in both native quad core and native octo core. [4] The definition of "native" (monolithic) is the configuration with all the cores situated on a single silicon die. The quad core variant is already known to have 731 million transistors, and a currently unknown amount of L2 cache. For Nehalem processors with additional features, such as GPUs, there will be two separate dies; one with the cores, and the other one will have the additional features.

 

[ilkhan]

interesting. Since we have the transistor count I wonder what the die is is going to be for the octo version (16 friggin threads!)

are there any known plans to run a dual die version? That would go from 4 threads per socket straight to 32 threads in a single socket.

And finally DDR3 controller only or will there be a DDR2 version?

 

[Brahmzy]

When's ETA?

 

[ilkhan]

When's ETA?

2H-'08 is the best we know at this point, AFAIK.

 

[Eulogy]

2H-'08 is the best we know at this point, AFAIK.

I'm GUESSING that, with Conroe and Yorkfield and examples, that it'll be more towards Q4 '08.

 

[ilkhan]

I'm GUESSING that, with Conroe and Yorkfield and examples, that it'll be more towards Q4 '08.

:shrug:
They already demoed the chips in action. I dont know what exactly is holding up the launch at this point. 45nm production? Additional testing? I just dont know. But the silicon should be good to go.

 

[dexvx]

:shrug:
They already demoed the chips in action. I dont know what exactly is holding up the launch at this point. 45nm production? Additional testing? I just dont know. But the silicon should be good to go.

Just because it can launch Windows and run Cinebench doesn't mean its a stable product.

 

[Eulogy]

:shrug:
They already demoed the chips in action. I dont know what exactly is holding up the launch at this point. 45nm production? Additional testing? I just dont know. But the silicon should be good to go.

Same thing with Conroe. They had good working silicon what, 8 months in advance?

 

[Jasonx82]

:shrug:
They already demoed the chips in action. I dont know what exactly is holding up the launch at this point. 45nm production? Additional testing? I just dont know. But the silicon should be good to go.

Some of the delay will probably be from them wanting to milk all they can outta the Conroe/Penryn? Maybe?

 

[Manny Calavera]

:shrug:
They already demoed the chips in action. I dont know what exactly is holding up the launch at this point. 45nm production? Additional testing? I just dont know. But the silicon should be good to go.

AMD having fallen down and cant get back up.ie competition.And what Jason said,milking what they have now or all its worth.

 

[ilkhan]

New info! http://en.wikipedia.org/wiki/Nehalem_(CPU_architecture)

I wonder what the differences between bloomfield and lynnfield are going to be (besides the third mem channel) as well as why theres no desktop octo version (besides no real demand for 16 thread) in a desktop socket.

Anything else think bloomfield might all be unlocked? Be a real incentive to move up....which would have a significant cost associated with it. Not just in the CPU but also in memory. With DDR3 youd have go from 2x2GB to either 3x1GB (cheap but possibly too little in late 2008) or 3x2GB (expensive). And fast DDR3 2GB chips are going to be expensive.

/me starts saving.

 

[fluxion]

:shrug:
They already demoed the chips in action. I dont know what exactly is holding up the launch at this point. 45nm production? Additional testing? I just dont know. But the silicon should be good to go.

i wouldnt be surprised if intel is "stockpiling" it's technology at this point in lieu of any serious competition from amd atm. seems like they could hit H208 if they wanted, but riding out penryn for a few months extra might be more financially viable

seems like penryn clock speeds have already been scaled down for similar reasons (just speculating here).

 

[ilkhan]

i wouldnt be surprised if intel is "stockpiling" it's technology at this point in lieu of any serious competition from amd atm. seems like they could hit H208 if they wanted, but riding out penryn for a few months extra might be more financially viable

seems like penryn clock speeds have already been scaled down for similar reasons (just speculating here).

Certainly possible. Im just trying to decide if its worthwhile to upgrade now, wait the little bit for a yorkfield, or hold out for a bloomfield. Id lose a least several hundred dollars of value by the time bloomfield hits, so its a hard choice. Especially having a month of no school coming up...grabbing an X3210 and OCing the bejeezes out of it is sounding very attractive.

 

[Menelmarar]

Intel internal policy AFAIK requires an insane amount of testing and qualification. Which is a good thing.

Additionally, Intel is a corporation, they will release it when it makes economic sense for them.

 

[Manny Calavera]

interesting. Since we have the transistor count I wonder what the die is is going to be for the octo version (16 friggin threads!)

are there any known plans to run a dual die version? That would go from 4 threads per socket straight to 32 threads in a single socket.

And finally DDR3 controller only or will there be a DDR2 version?

Individual die size will take a large jump up,from 19~ million transisitors on Penryn to 30~
on Nehalam

 

[Digital Viper-X-]

I remember reading that the Quad will be out next year late, while the octo will have to wait for 32nm as even @ 45nm Yorksfield is pushing 150W TDP on the top end parts?

 

[ilkhan]

Individual die size will take a large jump up,from 19~ million transisitors on Penryn to 30~
on Nehalam

What are you talking about?
Transistor counts on the quad are already released. By die sizes I meant the physical size of each die (which along with yield rates determine the cost to intel). The conroe die is 143mm^2 (kentsfield 2x143mm^2) wolfdales are 107mm^2 (yorkfield 2x107mm^2).

Im wondering what the sizes are going to be for those 731million transistors on a bloomfield (or lynnfield, possibly, since they only said native quad and both of those are native quads...)

As to TDP wiki says that the non-extreme yorkfields are pushing 95. Plenty of room on high end cooling to scale up (go go 4GHZ!).

 

[Manny Calavera]

What are you talking about?
Transistor counts on the quad are already release. By die sizes I meant the physical size of each die (which along with yield rates determine the cost to intel). The conroe die is 143mm^2 (kentsfield 2x143mm^2) wolfdales are 107mm^2 (yorkfield 2x107mm^2).

Im wondering what the sizes are going to be for those 731million transistors on a bloomfield (or lynnfield, possibly, since they only said native quad and both of those are native quads...)

NM then,I am half asleep and have not had a coffee as yet.

 

[HarkovAdm]

still even if octo-cores do come out in 2008, they will likely have "xtreme" prices of $999+

Most of us will probably either get a nehalem quad for $316 in 2008 or wait for the octo "price cuts" of 2009. Just my guess

 

[Manny Calavera]

still even if octo-cores do come out in 2008, they will likely have "xtreme" prices of $999+

Most of us will probably either get a nehalem quad for $316 in 2008 or wait for the octo "price cuts" of 2009. Just my guess

Even then I dont think many here will be running octo-core procs.

 

[ilkhan]

Even then I dont think many here will be running octo-core procs.

I agree. Not a lot of people need 16 threads running at once. Its rather specialized. Even with a quad, running full res fraps while playing a quad optimized game is only going to take 8 threads which is a quad nehalem. Now if you wanted to realtime transcode that to an alternate format I can see running several threads on that, but how many people have HDDs that would keep up with that much activity? (Would take a couple of raptor/SSDs to rip WHILE transcoding.

 

[Mystofolyse]

BOINC.

 

[Neuffy]

I agree. Not a lot of people need 16 threads running at once. Its rather specialized. Even with a quad, running full res fraps while playing a quad optimized game is only going to take 8 threads which is a quad nehalem. Now if you wanted to realtime transcode that to an alternate format I can see running several threads on that, but how many people have HDDs that would keep up with that much activity? (Would take a couple of raptor/SSDs to rip WHILE transcoding.

Well, you actually should need far less hdd speed if you're transcoding on the fly (without copying to hdd first). If I can encode with FRAPS (or equivalent) H.264 it will take far less hdd speed than if I'm just dumping to mpeg2 or whatever.

Also, can just write to multiple hard drives. Multiple standard hard drives should be able to take partials that can be later combined.

God it'd be nice to have that much power. I guess I'll be getting an octo when they drop to ~$350 or less...

 

[Big D.]

I'm really happy with my Q6600 and 4 cores right now. When I saw Mental Ray and Brazil R/S using all 4 cores at max and my render times going from 8 seconds (3 cores disabled) to 1-2 seconds I almost jumped out of joy.

But yeah, apart from when I render in 3DS Max, I rarely see anything giving my CPU more than a 50% usage, so octas really won't be needed by most people. As it is programmers are barely managing to using 2+ cores.

 

[note235]

im getting the octo

need a lot of rendering power
especially terragen 2...

 

[ilkhan]

Everything Ive heard says the octocores will only be released on the >2 chip socket. So good luck getting a single octo. Besides, thats 16 threads, quite a few by any count.

 

[DemonDiablo]

I wonder how long it will take them to get to 16 core cpus with HT enabled. Imagine that beast. Its a shame that software developement is light years behind processing developement. It almost makes all of this pointless. I may as well still be using a P1 at the developement rate of software.

Still a MP 16 core setup will go along way to folding at whatever you are folding for. The power of 64 cores in only 2 physical sockets. Nobody would have thought it possible 5 years ago and here we are looking at such a feat within the next plausible 5 years.

 

[JSpecGC8]

But how long will it take current software makers to make full use of this new power?

I am going to agree that Intel is going to wait on a strong demand from the market for a CPU of this caliber. I mean honestly the current CPUs they have out are insane and can handle pretty much anything, why release something that there isnt a high demand for?

 

[DemonDiablo]

But how long will it take current software makers to make full use of this new power?

I am going to agree that Intel is going to wait on a strong demand from the market for a CPU of this caliber. I mean honestly the current CPUs they have out are insane and can handle pretty much anything, why release something that there isnt a high demand for?

Which is exactly why they are probably only releasing the octo-core for servers only as GLSauron pointed out.

For desktops yea theres not much need for 8 processing threads let alone 16 but for the server world, the more the merrier. With servers 16 core HT enabled cpu has the processing potential of replacing an entire room full of servers with just a few boxes. What Intel is close to embarking on doing for the server world is similar to the leap in technology from a 1960's mainframe computer (things were huge just so you know) to todays 15" laptops.

Also you have the fact that Nehalem will have a higher ipc than current cpu's so yea, good things for the server world.

 

[Neuffy]

Well, not quite the mainframe->15" laptops jump...

The thermal output of these is still going to be huge. That necessitates significant cooling hardware, which means that a quad-socket 64-core box is still going to need to have a certain amount of cooling applied.

It will be nice to have, though.