Intel brings forward Nehalem Launch

[Donnie27]

If one can't OC nehalem, then why getting one? ...unless early 2 GHz Nehalem is faster than 3.8-4.2 GHz Core2Quad.

...which I doubt.

The slowest model will be 2.66GHz and what Intel Quad Cores overclock to 3.8 to 4.2GHz right now? Current models might need about a 400MHz head start.

[Blauhung]

I doubt the reference clock is designed to be overclocked on AMDs either but they seem to go high enough on select chips.

If they don't overclock well via the reference clock or the board manufacturers don't figure out a way to sneak in higher multipliers than intended through 'Turbo' then I must say I'm less tempted. A $999 CPU is probably the last thing I'll ever buy.

I'm kinda out of it on how high the ref clock on AMD chips goes, but is a 50% increase there within the realm of possibility? It's a very similar system so if 300MHz clocks from a base of 200 on AMD are possible I would assume that it's may be possible to get 200Mhz from a base of 133 on Intel.

I haven't been paying too much attention to AMD clocking, but I don't think i've seen 300. Please correct me if I'm wrong.

EDIT:
hehe, and I found some
http://www.xtremesystems.org/forums/...d.php?t=167502 There's 310

So right now it's all just a waiting game. We will have to wait and see how well the reference clock does scale.

[demonkevy666]

front side bus clocks is said to make things hotter (right?)

so 266mhz down to 133mhz may lower heat & power some how.

[Movieman]

I'm kinda out of it on how high the ref clock on AMD chips goes, but is a 50% increase there within the realm of possibility? It's a very similar system so if 300MHz clocks from a base of 200 on AMD are possible I would assume that it's may be possible to get 200Mhz from a base of 133 on Intel.

I haven't been paying too much attention to AMD clocking, but I don't think i've seen 300. Please correct me if I'm wrong.

EDIT:
hehe, and I found some
http://www.xtremesystems.org/forums/...d.php?t=167502 There's 310

So right now it's all just a waiting game. We will have to wait and see how well the reference clock does scale.

I don't want to wait..

[Blauhung]

front side bus clocks is said to make things hotter (right?)

so 266mhz down to 133mhz may lower heat & power some how.

You can't really compare the 2 in that way. It's a totally different interconnect. The move from the FSB to the QPI, i think does have some positive effects on power usage, but in this case the power usage doesn't have anything to do with clock speed.

I don't want to wait..

I know, I'm all giddy too

SQUWEEEEEEEEEEE

[Movieman]

You can't really compare the 2 in that way. It's a totally different interconnect. The move from the FSB to the QPI, i think does have some positive effects on power usage, but in this case the power usage doesn't have anything to do with clock speed.



I know, I'm all giddy too

SQUWEEEEEEEEEEE

This is sitting behind me waiting for a Gainstown system.
Just the case, the guts will be not Koolance..
http://www.koolance.com/water-coolin...product_id=198

[Extelleron]

Held back?

The Yorkfield core is capable of 4GHz, but the Q9300 / Q9450 are held back because of the multiplier. 4GHz on a Q9450 would require a 500MHz FSB, absolutely impossible to make stable on air cooling with a quad-core.

[chuckbam]

How hard will it be to get the base clock to 200MHz? I'm wondering how easy/hard it will be to get the 2.66GHz Nehalem to 4.0GHz on air cooling. The actual Nehalem core should be capable of similar (or even better perhaps, because of the deeper pipeline) clocks to Core 2, so the question is whether or not the non-XE versions will be held back as much as a CPU like the Q9300 or Q9450 is held back now.

It does have a low core volt. I think it is 1.075v. I think we will soon start seeing some OC Nehalem reviews.

http://www.nehalemnews.com/

[chuckbam]

The slowest model will be 2.66GHz and what Intel Quad Cores overclock to 3.8 to 4.2GHz right now? Current models might need about a 400MHz head start.

That is just the slowest first release Bloomfield (4 cores w/ hyperthread). They call the 2.66 Bloomfield mainstream. I don't think any Bloomfield is mainstream. Many LGA1366 CPUs will follow.

[Zytek_Fan]

That is just the slowest first release Bloomfield (4 cores w/ hyperthread). They call the 2.66 Bloomfield mainstream. I don't think any Bloomfield is mainstream. Many LGA1366 CPUs will follow.

No mainstream parts will be available at launch. It will be all server and high end at launch, with mainstream and performance mainstream trickling in over what probably will be a few months.

Remember, Intel still has some Penryns to sell off, and the Christmas buying season is perfect for that.

[Kasparz]

No mainstream parts will be available at launch. It will be all server and high end at launch, with mainstream and performance mainstream trickling in over what probably will be a few months.

No.
Bloomfield isn't mainstream, but cheapest 2.66GHz 285$ is mainstream. Whole mainstream platform is Lynnfield, that will be available late Q2/Q3 2009.

[Movieman]

If one can't OC nehalem, then why getting one? ...unless early 2 GHz Nehalem is faster than 3.8-4.2 GHz Core2Quad.

...which I doubt.

A 3000mhz dual Gainstown is "faster" than my Harpertown at 3600..
I needed to break 3750 to match it and remember, thats matching a pre release beta system not a refined retail.
This was using CineBench10 as a test..

[JumpingJack]

front side bus clocks is said to make things hotter (right?)

so 266mhz down to 133mhz may lower heat & power some how.

Yes and no .... The system clock is used to derive the clocks for many of the other components in a system, some other oddites have their own clock gens, but the major components, memory - CPU - bus /chipset time themselves from a common clock.

Each device operates at a frequency of some multiplier of that base clock. The power given by any one chip is really, to a simple form, a sum of two terms.

Ptotal = Pdynamic + Pstatic

The Pdyanamic term is the electrical power as a result of an oscillating current against a load formed from a capacitor. The Pstatic term is the rest power or idle power which comes from leakage (but not all the potential leakage, just the 'off' state leakage). Think of Pdynamic as the power to flip the switch (transistor) and the Pstatic term as the small amount of current that may flow through a faulty switch when in the off position.

The Pstatic term is very hard to define, so for the sake of argument let's say this is small. Of course it isn't terribly small but it use to be in the past

The main contribution then is Pdynamic which is simply defined as P=C*V^2*F with C as the total capacitance for the device (chip), V is voltage, and F is frequency.

So while the system clock is 133 MHz, if the chip is multiplied up to 3.0 GHz, it will consume the same amount as if the system clock was 300 Mhz and multiplied up by 10. This is no different.

However, there are other system components that clock against this, and the new QPI implementation ... it will not be readily known how the power will compare until we actually get to see the stuff in action.

Jack

[fragmasterMax]

holy $hit
Intel and its f'ing blitzkrieg product releases. This is going to totally kick amd in the ballz.... again
I'm definitely not complaining though.

[Donnie27]

That is just the slowest first release Bloomfield (4 cores w/ hyperthread). They call the 2.66 Bloomfield mainstream. I don't think any Bloomfield is mainstream. Many LGA1366 CPUs will follow.

I only did reply because the others replied so well. There are currently Mainstream Xeons and Bloomfield will fit right in with these in that same market. One thing Intel has always done well by Hook or by Crook is fill all market segments half-way nice choices.

Mainstream is the defined differently by almost all tech companies. Intel's road-map lists that slowest Nehalem as Mainstream, not be confused with Budget Mainstream.

Edit for example.
http://www.expreview.com/img/news/20...oomfield_1.png

[natty]

Nehalem is not overly well optimized to do much of anything with single threaded apps, which represent most of the bulk of the software sold out there in cyberland. At most you will see about a 10% performance increase clock per clock, quad core per quad core. So a Bloomfield running at 2.66 GHz will outrun a Q9450 by roughly 10% on benchmarks that involve single threads. However, feed the Bloomy a blooming multithreaded app, and watch it leave the Penryn in its trail with up to double the performance.

is all this information correct?

[Stumbo]

Yep, pretty correct.

Im tired of my E6600. When does the waiting stop?!?!?!?

[JumpingJack]

is all this information correct?

No. because of this -- " ... represent most of the bulk of the software sold out there in cyberland" .... it is probably 1/2 or more is multithreaded. Everything else is about right, though i would disagree with the 10% number ... it will vary pretty wildly from just a little to 20% or more for some single threaded apps. However, I would not say he is wrong because either of us could be right, we will know when the processor arrives.

Jack

[m^2]

No. because of this -- " ... represent most of the bulk of the software sold out there in cyberland" .... it is probably 1/2 or more is multithreaded. Everything else is about right, though i would disagree with the 10% number ... it will vary pretty wildly from just a little to 20% or more for some single threaded apps. However, I would not say he is wrong because either of us could be right, we will know when the processor arrives.

Jack

I'd say that 10% is far too high estimation.
50%+

Most programs don't need it though.

[JumpingJack]

I'd say that 10% is far too high estimation.
50%+

Most programs don't need it though.

http://www.digit-life.com/articles3/...-e4x00-p1.html

His software list is pretty well representative of both professional an non-professional applications available today.

Gen-2 games of course are not, but the majority of contemporary games are. Multicore processors have been out > 3 years now.

A month or two ago I was also when someone said that same thing, then when I went out and actually looked... it's true, about 1/2 of what you can find is multithreaded (excluding the sharecrapware) and there is a multithreaded app available for most all catagories of usages you can find.

Video encoding -- yep.
NLE -- Yep
Photography -- yep
Games -- yep
compression -- yep

Software has done a more to catch up that what we realize.

On the single threaded material... it will vary, but you will be shocked when the numbers come out I suspect. What is more shocking, and make note of this ... it will be hard to find more single threaded benchmark data since there will be much much less of it when the reviews come.

[Boschwanza]

http://www.digit-life.com/articles3/...-e4x00-p1.html

His software list is pretty well representative of both professional an non-professional applications available today.

Gen-2 games of course are not, but the majority of contemporary games are. Multicore processors have been out > 3 years now.

Video encoding -- yep.
NLE -- Yep
Photography -- yep
Games -- yep
compression -- yep

Software has done a more to catch up that what we realize.

You are right but just opening multiple threads and utilizing all this threads is still a very big different. I other words .. whats better ? : 2 full utilized Core or all cores with 12.5 % (quad core).

For examble:

Just found out that video encoding (Nero Suite) scales very well from 1 to 2 Cores (About 1.8x) but from 1 to 4 its only about 2.7x, which means one core is totaly obsolet.

[Shintai]

You are right but just opening multiple threads and utilizing all this threads is still a very big different. I other words .. whats better ? : 2 full utilized Core or all cores with 12.5 % (quad core).

For examble:

Just found out that video encoding (Nero Suite) scales very well from 1 to 2 Cores (About 1.8x) but from 1 to 4 its only about 2.7x, which means one core is totaly obsolet.

It doesnt mean 1 core is obsolete. Its about diminishing returns.

[NeedMoMegaHurtZ]

You are right but just opening multiple threads and utilizing all this threads is still a very big different. I other words .. whats better ? : 2 full utilized Core or all cores with 12.5 % (quad core).

For examble:

Just found out that video encoding (Nero Suite) scales very well from 1 to 2 Cores (About 1.8x) but from 1 to 4 its only about 2.7x, which means one core is totaly obsolet.

Again, depends on the specific encoder and settings used.

x264 is one that scales well, 100% usage for 8 cores (it's been reported 16 cores as well and probably more when it comes out) when settings are maxxed out

[JumpingJack]

You are right but just opening multiple threads and utilizing all this threads is still a very big different. I other words .. whats better ? : 2 full utilized Core or all cores with 12.5 % (quad core).

For examble:

Just found out that video encoding (Nero Suite) scales very well from 1 to 2 Cores (About 1.8x) but from 1 to 4 its only about 2.7x, which means one core is totaly obsolet.

Good point, I never said that the multithreaded apps were well multithreaded Windows Media Encoder will only split off two threads .. is another example.

The reason I made this point is about 2-3 months ago, I was developing a multitasking test scenario. What I wanted was one case where I had one dual threaded app, and two single threaded apps, in other cases I wanted all just general threaded apps.

It is now hard to find general single threaded apps that represent a mainstream like applications that puts any kind of load on the processor.

Jack

[Boschwanza]

Again, depends on the specific encoder and settings used.

x264 is one that scales well, 100% usage for 8 cores (it's been reported 16 cores as well and probably more when it comes out) when settings are maxxed out

You are right x264 is a perfect example for very well optimized SMT (But thats the absolulty minor at least for Desktop APPs). Just try to recode a DVD to xvid and you will not scale as well.

I just wanted to say that you wont get a massive perfomance gain with a nehalem just because of the fact the application uses multithreading. Still depending on how good SMT scales with the cores.