Nehalem-EP......BLOOMFIELD

[Drwho?]

"
nehalems overclocks automatically, even to 3.4 ghz or more, so the inflated benchmarks corresponds to overclocked processors and that explains the 130 tdp, even for the 2.6ghz processor.

i think this is not good for (dirty) intel. using overclocking to improve results."

source; vr-zone from (http://www.tomshardware.com/news/Tur...ntel,6193.html , http://www.fudzilla.com/index.php?op...9455&Itemid=35 )

wow wow wow!
If you don't mind, The architect of the processor designed it to take advantage of the thermal "room" we have on the socket. it is not cheating or anything like you are saying, it is being smarter about thermal. Today's processors are very stupid in that sense, an opteron or a Yorkfield does not increase the frequency, even if it could, because it has room when not using 4 threads ... on single threaded program, 3 cores are in C6 states. Why giving penalty to the single threaded performance, because the Max power of 4 cores. PCU + Turbo fix this.
If you have a program using 4 threads, but not using a lot of thermals, because using a tight loop in the L1 (not using a lot of transistors), why not raising the frequency of 1 or 2 steps if your process technology allow you to do so and PAST EVERY QUALITY TEST !!!!



Using the thermal envellope in a better way is very smart, all other x86 makers are stock at their thermals because of the 4 cores max power.

On the top of this, Nehalem has Power gating for each Core, we can reduce the power usage to a neglisable quantity, it would have be foolish to implement this without the PCU (Power Controle Unit)

The Architect of Nehalem invested a lot of time in the PCU, it has more transistor than the original i80486 ... just to give you the scale.

the PCU and its reward will give nehalem an edge on all small form factor PC, it will allow you to overclock, it will make sure you minimize your carbon foot print.

nothing related to cheating, it is a new features that will save millions of tonnes of carbon, for Fudo ... it is Fudo ... what do you expect from "FUD"zilla ... some FUD , no?
and for Tomshardware ... well, this is a long serie of innacurate articles due to the fact the dresden fab is not doing the same ...

PS: This is my personal opinion for the conclusion ...

[Hornet331]

I`d like to ask for whole mounting socket dimension.. same guys are talking about 80x80mm but few other are talking abou 113x113... which is the true, jejeeje no body want to answer me..

JC (JCornell) already posted pics of the hole spacing with a ruler next to it, ots somewhere in the guts of this tread.

[Jamesrt2004]

wow wow wow!
If you don't mind, The architect of the processor designed it to take advantage of the thermal "room" we have on the socket. it is not cheating or anything like you are saying, it is being smarter about thermal. Today's processors are very stupid in that sense, an opteron or a Yorkfield does not increase the frequency, even if it could, because it has room when not using 4 threads ... on single threaded program, 3 cores are in C6 states. Why giving penalty to the single threaded performance, because the Max power of 4 cores. PCU + Turbo fix this.
If you have a program using 4 threads, but not using a lot of thermals, because using a tight loop in the L1, why not raising the frequency of 1 or 2 steps if your process technology allow you to do so and PAST EVERY QUALITY TEST !!!!

Using the thermal envellope in a better way is very smart, all other x86 makers are stock at their thermals because of the 4 cores max power.

On the top of this, Nehalem has Power gating for each Core, we can reduce the power usage to a neglisable quantity, it would have be foolish to implement this without the PCU (Power Controle Unit)

The Architect of Nehalem invested a lot of time in the PCU, it has more transistor than the original i80486 ... just to give you the scale.

the PCU and its reward will give nehalem an edge on all small form factor PC, it will allow you to overclock, it will make sure you minimize your carbon foot print.

nothing related to cheating, it is a new features that will save millions of tonnes of carbon, for Fudo ... it is Fudo ... what do you expect from "FUD"zilla ... some FUD , no?
and for Tomshardware ... well, this is a long serie of innacurate articles due to the fact the dresden fab is not doing the same ...

PS: This is my personal opinion for the conclusion ...

I agree with you totally on that + 1

I soo can't wait to make a nehalem system about july next year

[Drwho?]

JC (JCornell) already posted pics of the hole spacing with a ruler next to it, ots somewhere in the guts of this tread.

it is 8cmx8cm for Core i7 instead of 7cmx7cm for Yorkfield 775

Francois

[T_Flight]

wow wow wow!
If you don't mind, The architect of the processor designed it to take advantage of the thermal "room" we have on the socket. it is not cheating or anything like you are saying, it is being smarter about thermal. Today's processors are very stupid in that sense, an opteron or a Yorkfield does not increase the frequency, even if it could, because it has room when not using 4 threads ... on single threaded program, 3 cores are in C6 states. Why giving penalty to the single threaded performance, because the Max power of 4 cores. PCU + Turbo fix this.
If you have a program using 4 threads, but not using a lot of thermals, because using a tight loop in the L1 (not using a lot of transistors), why not raising the frequency of 1 or 2 steps if your process technology allow you to do so and PAST EVERY QUALITY TEST !!!!

Using the thermal envellope in a better way is very smart, all other x86 makers are stock at their thermals because of the 4 cores max power.

On the top of this, Nehalem has Power gating for each Core, we can reduce the power usage to a neglisable quantity, it would have be foolish to implement this without the PCU (Power Controle Unit)

The Architect of Nehalem invested a lot of time in the PCU, it has more transistor than the original i80486 ... just to give you the scale.

the PCU and its reward will give nehalem an edge on all small form factor PC, it will allow you to overclock, it will make sure you minimize your carbon foot print.

nothing related to cheating, it is a new features that will save millions of tonnes of carbon, for Fudo ... it is Fudo ... what do you expect from "FUD"zilla ... some FUD , no?
and for Tomshardware ... well, this is a long serie of innacurate articles due to the fact the dresden fab is not doing the same ...

PS: This is my personal opinion for the conclusion ...

I agree. Until now, CPU's have not been able to do this. Nehalem is just a superior technology, and I'm glad the designer had the forward thinking to pull that off.

I'm certainly glad I chose watercooling for this new CPU. I've been waiting a long time on this CPU. I need this CPU now. Actually I needed it a year ago. I'm really hoping that the release it no later than November. I absolutely must have a system before years end. The winter season is when I do some design work, and my computer is just hardly able to cope with the modelling programs I'm using now. I need multicore...not want, but *need*.

[LowRun]

"
nehalems overclocks automatically, even to 3.4 ghz or more, so the inflated benchmarks corresponds to overclocked processors and that explains the 130 tdp, even for the 2.6ghz processor.

i think this is not good for (dirty) intel. using overclocking to improve results."

source; vr-zone from (http://www.tomshardware.com/news/Tur...ntel,6193.html , http://www.fudzilla.com/index.php?op...9455&Itemid=35 )

Dude, this is XS, no one will complain because his CPU automatically overclocks as long as traditional OC is up to the expectations

[Jacky]

"nehalems overclocks automatically, even to 3.4 ghz or more, so the inflated benchmarks corresponds to overclocked processors and that explains the 130 tdp, even for the 2.6ghz processor.
i think this is not good for (dirty) intel. using overclocking to improve results."

What the hell is going on? "Dirty" Intel? Intel is as "dirty" as they used to be.
It's stupid *not* to use the overclocking potential if possible.
However, I believe none of the benchmarks used turbo mode, Nehalem seems to be only 5-10% faster in single-threaded apps. If they need turbo mode to achieve those gains, then nehalem would be slower clock for clock than penryn. A step backwards by the new Intel? Very unlikely, yet possible.

"Leaked information also indicates that production CPUs will self overclock by up to two speed bins" as per tomshardwere

EDIT:
The overclocks look lackluster as compared to mature core 2 quads, but are you sure we can compare them, I mean Nehalem features completely different circuits. Maybe the design with static CMOS reacts different to voltage and is more robust? Anyone with some knowledge care to comment? JJ?

[Newblar]

Dude, this is XS, no one will complain because his CPU automatically overclocks as long as traditional OC is up to the expectations

depends on what your definition of traditional ocing is
pushing 500+fsb, or gettin a couple of mhz out of the HTT

[Donnie27]

depends on what your definition of traditional ocing is
pushing 500+fsb, or gettin a couple of mhz out of the HTT

It also depends on IPC, IMC gains and etc...... Prescott overclocked like mad and seemed to be spinning its wheels. If at 3.4GHz, it could hang with the overclocked Yorkies at 3.7 to 4+GHz, who in the hell will care?

But as someone else said, please wait until the NDA's are over with for good and bad results to be supported. Too much hype for Nehalem is just as bad as too much FUD

What would I have liked to have seen? A straight Core 2 Duo or Quad with an IMC and NO L3 at all. No change to the cache setup as well. This would be to show just how much IMC is overrated on a Single Socket Desktop system.

[ken]

I`d like to ask for whole mounting socket dimension.. same guys are talking about 80x80mm but few other are talking abou 113x113... which is the true, jejeeje no body want to answer me..

I dont know about the new socket but LGA755 mounting is 80x80

[Donnie27]

Originally Posted by Monkeywoman
"
nehalems overclocks automatically, even to 3.4 ghz or more, so the inflated benchmarks corresponds to overclocked processors and that explains the 130 tdp, even for the 2.6ghz processor.

i think this is not good for (dirty) intel. using overclocking to improve results."

source; vr-zone from (http://www.tomshardware.com/news/Tur...ntel,6193.html , http://www.fudzilla.com/index.php?op...9455&Itemid=35 )

Let's see, you don't have a lot of folks complaining about Processors throttling down but cranking up is a problem, WOW! I think we've come full circle!

Please note that features can be turned off or disabled

[Monkeywoman]

Let's see, you don't have a lot of folks complaining about Processors throttling down but cranking up is a problem, WOW! I think we've come full circle!

Please note that features can be turned off or disabled

calm down people, all i'm saying that the initial "stock" benchmarks might seem high because during the run; the cpu will clock up instead of staying at the "stock" speeds i.e. 2.66Ghz-->during bench run to 3.2Ghz.

i know they will be fast but if they are to be tested; it should be with this feature turned off.

[informal]

I'm sure many sites will turn off Turbo mode and test Core i7 at its rated speed.It would be interesting to compare those results to the ones we have up until now.

[LowRun]

If this stuff is gonna be enabled by default for everyone and behaving identically i can't see why it should be turned off for benchs

[leoy]

If speculations are right then if u clock nehalem to 4ghz then with turbo mode on it will go up to 5!
Nah....
Then it would propably crash and u would need to keep turbo mode off, in order to be always at 4...

[Drwho?]

Let's be clear, when you overclock an XE Core i7, you 'll be using the Turbo max ratio as the max ratio ... YOU WILL NOT GET TURBO KICK IN WHEN YOU ARE OVERCLOCKING. Turbo IS the way you ll over clock, then, you ll be using the stock clock (133MHz) and you will increase it...
I recommand that you do not max out the Turbo ratio, and do a trade between Turbo ratio and clock increase via the base clock. This gives the best frequency results.

[Blauhung]

If speculations are right then if u clock nehalem to 4ghz then with turbo mode on it will go up to 5!
Nah....
Then it would propably crash and u would need to keep turbo mode off, in order to be always at 4...

I would think the smarter way to play this would be to leave turbo mode on, and consider the multi as always n+2 when calculating what the final clock will be.

This allows you to still make use of the intelligent clock scaling up and down in periods of loading and not loading for power efficiency.

As far as reviews go. I believe that any reviewer that is looking at the platform for stock ability comparisons should leave the features to default to show consumers the out of the box performance of the platform.

Let's be clear, when you overclock an XE Core i7, you 'll be using the Turbo max ratio as the max ratio ... YOU WILL NOT GET TURBO KICK IN WHEN YOU ARE OVERCLOCKING. Turbo IS the way you ll over clock, then, you ll be using the stock clock (133MHz) and you will increase it...
I recommand that you do not max out the Turbo ratio, and do a trade between Turbo ratio and clock increase via the base clock. This gives the best frequency results.

And exactly, unlike the FSB, the speed of the reference clock has no direct impact on overall performance of the platform. With FSB overclocking, quite often, a lower CPU multi and higher FSB speed can hit the same CPU clock, but speed up data transfer across the bus (providing you aren't switching to a looser strap or Trd as it's called nowadays).

Here, there are no performance gains for moving the reference clock other then fine tuning CPU/Mem/QPI speeds between multipliers that fall on either side of a stability zone.

[informal]

As far as reviews go. I believe that any reviewer that is looking at the platform for stock ability comparisons should leave the features to default to show consumers the out of the box performance of the platform.

That's indeed true for common users.But XS and other enthusiast communities are not "common user" groups and we'd like to see a real clock/clock comparison being made .That said,auto OC is really a good feature.

[Drwho?]

I would think the smarter way to play this would be to leave turbo mode on, and consider the multi as always n+2 when calculating what the final clock will be.

This allows you to still make use of the intelligent clock scaling up and down in periods of loading and not loading for power efficiency.

As far as reviews go. I believe that any reviewer that is looking at the platform for stock ability comparisons should leave the features to default to show consumers the out of the box performance of the platform.



And exactly, unlike the FSB, the speed of the reference clock has no direct impact on overall performance of the platform. With FSB overclocking, quite often, a lower CPU multi and higher FSB speed can hit the same CPU clock, but speed up data transfer across the bus (providing you aren't switching to a looser strap or Trd as it's called nowadays).

Here, there are no performance gains for moving the reference clock other then fine tuning CPU/Mem/QPI speeds between multipliers that fall on either side of a stability zone.

You will get more frequency on the extreme edition if you do not push too hard the frequency multipliers inside the dice. it is why you want to increase the stock clock as part of the mix. I am going into much details, but it is about clock distribution over the dice area.

[Movieman]

I read thru this thread and some of what you people know amazes me.
I always thought I was pretty good with hardware but with some of the people here I feel like a toddler in Kindergarten..

I have a question for you Dr. Who;
Will the Gainstown systems also have the ability with the 3200EE's to OC or will they be locked?

[Drwho?]

I read thru this thread and some of what you people know amazes me.
I always thought I was pretty good with hardware but with some of the people here I feel like a toddler in Kindergarten..

I have a question for you Dr. Who;
Will the Gainstown systems also have the ability with the 3200EE's to OC or will they be locked?

I am not allow to speak for Gainstown ... it is my Buddy Larry who is in charge of this server/workstation, I can't speak for his plan ...
sorry

(You always ask question I can t answer ... lol )

[Movieman]

I am not allow to speak for Gainstown ... it is my Buddy Larry who is in charge of this server/workstation, I can't speak for his plan ...
sorry

(You always ask question I can t answer ... lol )

That's ok, You answer honestly and thats all that matters.
So,ahh, what's Larry's telephone number?
Now I want you to understand that there's this 56 year Old guy sitting in New Hampshire that is dreaming of an 8 core gainstown running 16 WU at a time on WCG at hopefully a speed of 4000mhz so that explains my constant questions..

[Donnie27]

calm down people, all i'm saying that the initial "stock" benchmarks might seem high because during the run; the cpu will clock up instead of staying at the "stock" speeds i.e. 2.66Ghz-->during bench run to 3.2Ghz.

i know they will be fast but if they are to be tested; it should be with this feature turned off.

I stopped reading after calm down

[Donnie27]

I would think the smarter way to play this would be to leave turbo mode on, and consider the multi as always n+2 when calculating what the final clock will be.

This allows you to still make use of the intelligent clock scaling up and down in periods of loading and not loading for power efficiency.

As far as reviews go. I believe that any reviewer that is looking at the platform for stock ability comparisons should leave the features to default to show consumers the out of the box performance of the platform.

And exactly, unlike the FSB, the speed of the reference clock has no direct impact on overall performance of the platform. With FSB overclocking, quite often, a lower CPU multi and higher FSB speed can hit the same CPU clock, but speed up data transfer across the bus (providing you aren't switching to a looser strap or Trd as it's called nowadays).

Here, there are no performance gains for moving the reference clock other then fine tuning CPU/Mem/QPI speeds between multipliers that fall on either side of a stability zone.

Most important part here. Out of the box performance in more important than turning of stuff for theory and what ifs. It is far easier to work with something than to work against it. In this case it'd seem maxing out turbo tweaking it to the edge (over the edge = until its not stable anymore) would make all the sense in the world.

Go back to August 2006 (on this forum) and look at all of the features turned off or disabled to get better overclocks on the first C2D's? SO its not like folks aren't use to killing features to get better overclocks. IMHO, that's kind of counter productive

No matter how nicely said, some will always look for chinks and blemishes in the something shiny and new

[Blauhung]

You will get more frequency on the extreme edition if you do not push too hard the frequency multipliers inside the dice. it is why you want to increase the stock clock as part of the mix. I am going into much details, but it is about clock distribution over the dice area.

Something about signals between CPU/MEM/QPI being out of phase at higher multipliers? Or simply overworking the clock distribution circuits?