Socket 1366 obsolete, SMT a 'gimmick'

[randomizer]

Where's Jenny? No input ever since this thread was dug from the grave.

 

[andy5174]

They have almost exactly the same PCU and gate transistors as LGA1366 - the reasons they run cooler are:

1) The highest binned chips run 2.9GHz. This allows them to run at a lower voltage
2) No third memory channel
3) No ultra-high speed external connection (QPI)

Also, the most popular LGA1156 chip (the i5-750) has no SMT, which significantly cools it down. I would be willing to bet that if you ran all i7-975s at 2.9GHz, all i7-960s at 2.8GHz, and all i7-930s at 2.6GHz with no SMT, you'd find them running quite a bit cooler too.

This is incorrect.

Most of heat are generated by nothing but leakage current and the amount of leakage current depends the number of transistors instead of features(QPI/IMC.. etc). You get more heat from a circuit with more transistors given that the transistors applied are the same.

1) CPUs are basically millions of transistor circuits hooked together and IMC & QPI are nothing but some of the transistor circuits within them.
2) LGA1156 has more on-die transistors even without QPI and one less channel. Besides, LGA1156 also has on-die PCI-E controller, NB and DMI which are not on LGA1366.
3) The amount of leakage current is dependent on the number of transistors and the core temps are based on the leakage current.

Hence, LGA1156 should run much hotter if they had the same PCU and PGT as LGA1366. It is obvious that LGA1156 has much more efficient transistors that generates much less leakage current and therefore they run much cooler even with much more transistors integrated.

Again, heat is mainly generated by leakage current and leakage is dependent on the number of transistors instead of features. In addition, it is impossible for a CPU to run cooler than the other if it has more transistors on it, unless it is constructed by more efficient transistors. This is why C2Q runs hotter than C2D.


Yes, HT does make CPU run hotter. However, i7-860 has HT too and it still runs cooler and consumes less power than any higher binned i7-9xx even in load state.

 

[andy5174]

FYI,

Idle state is the state that the transistors should be turned off. Hence, the CPUs with lower idle power consumptions are obviously with more efficient transistors given that their frequencies are the same.

The chart shows that i7-860 consumes 30W less than the 920 in Idle state even with higher frequency.

In addition, P=VI and LGA1156 has higher voltage but still consumes less power!

What does it mean? It means that the leakage current of LGA1156 must be much much less than that of the LGA1366.

 

[cjl]

Note that that chart shows total system power consumption. The extra power consumed by the 1366 CPUs in that graph has absolutely nothing to do with the CPU. The extra power is because of the extra memory sticks and the northbridge that are present in the LGA 1366 setup, but not in the 1156 setup. If you were to measure the CPU alone, you would find that an i7-975 and an i7-980x use around 6-7 watts at idle. It would be completely impossible for any CPU to use 30W less than that, demonstrating that the power difference at the system level is entirely due to other components on the board (as I said, northbridge and memory).

See here for the numbers:

http://anandtech.com/cpuchipsets/showdoc.aspx?i=3763&p=2

(it's just below the die shots)

 

[fazers_on_stun]

Where's Jenny? No input ever since this thread was dug from the grave.

Maybe she & TC are enjoying their honeymoon 😀...

 

[jimmysmitty]

Note that that chart shows total system power consumption. The extra power consumed by the 1366 CPUs in that graph has absolutely nothing to do with the CPU. The extra power is because of the extra memory sticks and the northbridge that are present in the LGA 1366 setup, but not in the 1156 setup. If you were to measure the CPU alone, you would find that an i7-975 and an i7-980x use around 6-7 watts at idle. It would be completely impossible for any CPU to use 30W less than that, demonstrating that the power difference at the system level is entirely due to other components on the board (as I said, northbridge and memory).

See here for the numbers:

http://anandtech.com/cpuchipsets/showdoc.aspx?i=3763&p=2

(it's just below the die shots)

I would like to see then a setup with just 4GB of DDR3 to have a more on par comparison. But with the Northbridge it still wont be 100% comparable.

If anything though, the i7 980Xis more efficient than the i5 750 since the i5 750 is a 45nm part and the i7 980X is a 32nm part and would include any enhancements they got with 32nm, including gen 2 of the HK/MG.

And considering that I have seen a 980X compared to a older 45nm Core i7 using less power yet having 2 more cores it would make sense.

 

[EXT64]

I thought the x58 used quite a bit of power, thus skewing those figures more (than just a 3rd memory channel).

 

[warmon6]

Maybe she & TC are enjoying their honeymoon 😀...

[:jaydeejohn:4]

 

[cjl]

I thought the x58 used quite a bit of power, thus skewing those figures more (than just a 3rd memory channel).

Intel's spec for the X58 includes a 24.1 watt TDP.

 

[randomizer]

Maybe she & TC are enjoying their honeymoon 😀...

Of course. I forgot they got married... again :lol:

 

[andy5174]

Note that that chart shows total system power consumption. The extra power consumed by the 1366 CPUs in that graph has absolutely nothing to do with the CPU. The extra power is because of the extra memory sticks and the northbridge that are present in the LGA 1366 setup, but not in the 1156 setup. If you were to measure the CPU alone, you would find that an i7-975 and an i7-980x use around 6-7 watts at idle. It would be completely impossible for any CPU to use 30W less than that, demonstrating that the power difference at the system level is entirely due to other components on the board (as I said, northbridge and memory).

See here for the numbers:

http://anandtech.com/cpuchipsets/showdoc.aspx?i=3763&p=2

(it's just below the die shots)

@cil only:
I will say it once again. LGA1366's OFF-die NB has nothing to do with the """CPU's""" TEMP/CURRENT/POWER!


1) LGA1156 has much more on-die features which results in much more transistors(43 million more).

LGA1366 ON-DIE features: ( One more channel + QPI )
LGA1156 ON-DIE features: ( One less channel + on-die PCIE + on-die NB + DMI )

2) LGA1156 i7-860 runs cooler than 920 even with higher frequency and more transistors.
3) LGA1156 i7 has higher core voltage and P=VI, so LGA1156 has less leakage current EVEN if it use the exactly same power as LGA1366.

Hence, LGA1156 CPUs would have much more leakage current and run much hotter than LGA1366(which is not the case and completely oppose the truth) IF THEY HAD the exactly same transistors as LGA1366.

Thus, LGA1156 is obviously more efficient than LGA1366.

This is no surprise as LGA1156 comes with improved PCU and PGT.


@cil and Jimmy:
Yes, $999 LGA1366 32nm i7-980X is more efficient than LGA1156 because it has much further improved transistors from 32nm architecture WHICH IS NOT AVAILABLE ON OTHER LGA1366.

In addition, it is NOT affordable to 99.99%(cjl excluded) of the LGA1366 owners as I stated again and again and therefore it is no different from as if 980X is not there.

 

[archibael]

Just as a matter of clarification, the i7-860 is 45nm-- not 32nm-- and it has the on-die northbridge.

It is only the i3 and i5 Clarkdale's which are built on 32nm, and they have on-package northbridge (not on die)-- said northbridge being 45nm.

 

[andy5174]

Just as a matter of clarification, the i7-860 is 45nm-- not 32nm-- and it has the on-die northbridge.

It is only the i3 and i5 Clarkdale's which are built on 32nm, and they have on-package northbridge (not on die)-- said northbridge being 45nm.

My original sentence:

Yes, $999 LGA1366 32nm i7-980X is more efficient than LGA1156 because it has much further improved transistors from 32nm architecture WHICH IS NOT AVAILABLE ON OTHER LGA1366.

My original sentence is equivalent to:

Because it(i7-980X) has much further improved transistors from 32nm architecture WHICH IS NOT AVAILABLE ON OTHER LGA1366, $999 LGA1366 32nm i7-980X is more efficient than LGA1156.

The "it" here means LGA1366 i7-980X.

 

[cjl]

@cil only:
I will say it once again. LGA1366's OFF-die NB has nothing to do with the """CPU's""" TEMP/CURRENT/POWER!

Agreed. So why are you looking at total system power consumption if all you care about is the CPU?

1) LGA1156 has much more on-die features which results in much more transistors(43 million more).

LGA1366 ON-DIE features: ( One more channel + QPI )
LGA1156 ON-DIE features: ( One less channel + on-die PCIE + on-die NB + DMI )

2) LGA1156 i7-860 runs cooler than 920 even with higher frequency and more transistors.
3) LGA1156 i7 has higher core voltage and P=VI, so LGA1156 has less leakage current EVEN if it use the exactly same power as LGA1366.

Here's where your proof fails. I see no evidence provided that:

1) LGA1156 uses more voltage
2) the i7-860 runs cooler than the 920, all other things being equal

Because of this complete lack of proof, your conclusion is bogus.

Hence, LGA1156 CPUs would have much more leakage current and run much hotter than LGA1366(which is not the case and completely oppose the truth) IF THEY HAD the exactly same transistors as LGA1366.

Not necessarily. First, different sections of the CPU do different things, and necessarily require different amounts of power. You can't just say "LGA 1156 has more transistors so therefore it should use more power if it uses the same transistors". Even if the transistors are exactly the same, it's possible that QPI and memory controllers use more power than PCI-E controllers (I wouldn't know), in which case it makes sense that the LGA 1366 CPUs are more power hungry. Second, I am unconvinced that there's actually a significant difference in CPU power, since the complete difference in power in your total system numbers above can be accounted for as follows:

Idle: All Bloomfield (1366) chips use roughly 28 watts more than an i7-870, with almost zero difference between an i7-920 and an i7-975. The Bloomfield setups use an additional memory stick and also contain a northbridge not present in the Lynnfield setups. DDR3 DIMMs use 3-4 watts, and the TDP of the X58 chipset is 24.1 watts. This accounts for 28 watts, the entirety of the difference, indicating that the CPUs themselves are using almost exactly the same power at idle (which makes sense, since they use exactly the same power gate transistors).

At load, the i7-860 uses 33 watts less than the i7-920 (at similar, though admittedly not identical frequencies). 28 watts is easily explained due to the memory channel and northbridge, as explained above. The additional 5 watts can be explained as the 860 is a higher binned part (the 920 is the lowest binned Bloomfield), and likely runs at a slightly lower voltage.

Thus, LGA1156 is obviously more efficient than LGA1366.

Nope. Try again.

This is no surprise as LGA1156 comes with improved PCU and PGT.

Except that they don't.

@cil and Jimmy:
Yes, $999 LGA1366 32nm i7-980X is more efficient than LGA1156 because it has much further improved transistors from 32nm architecture WHICH IS NOT AVAILABLE ON OTHER LGA1366.

In addition, it is NOT affordable to 99.99%(cjl excluded) of the LGA1366 owners as I stated again and again and therefore it is no different from as if 980X is not there.

Fair enough. You'll note that I'm not comparing the 980 to Lynnfield above though - I'm comparing Bloomfield (specifically the 920 most of the time) to Lynnfield.

 

[andy5174]

@cjl:

Agreed. So why are you looking at total system power consumption if all you care about is the CPU?

It's you keep mentioning the non-related OFF-die NB of LGA1366 when I am talking about the "CPU" power/current.

Here's where your proof fails. I see no evidence provided that:

1) LGA1156 uses more voltage
2) the i7-860 runs cooler than the 920, all other things being equal

Because of this complete lack of proof, your conclusion is bogus.

Proof:
Anandtech show you that LGA1156 has much more transistors than LGA1366.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3634
(p.s. The number of transistors are shown in the table below the second picture of the page)

Anandtech and TOM shos you that LGA1156 require higher voltage.

You are just ignoring the links I provided so you see no proof.

Furthermore, where is your proof? You are just talking based on your feeling and said that LGA1156 doesn't come with more transistors.

Not necessarily. First, different sections of the CPU do different things, and necessarily require different amounts of power. You can't just say "LGA 1156 has more transistors so therefore it should use more power if it uses the same transistors". Even if the transistors are exactly the same, it's possible that QPI and memory controllers use more power than PCI-E controllers (I wouldn't know), in which case it makes sense that the LGA 1366 CPUs are more power hungry. Second, I am unconvinced that there's actually a significant difference in CPU power, since the complete difference in power in your total system numbers above can be accounted for as follows:

Idle: All Bloomfield (1366) chips use roughly 28 watts more than an i7-870, with almost zero difference between an i7-920 and an i7-975. The Bloomfield setups use an additional memory stick and also contain a northbridge not present in the Lynnfield setups. DDR3 DIMMs use 3-4 watts, and the TDP of the X58 chipset is 24.1 watts. This accounts for 28 watts, the entirety of the difference, indicating that the CPUs themselves are using almost exactly the same power at idle (which makes sense, since they use exactly the same power gate transistors).

At load, the i7-860 uses 33 watts less than the i7-920 (at similar, though admittedly not identical frequencies). 28 watts is easily explained due to the memory channel and northbridge, as explained above. The additional 5 watts can be explained as the 860 is a higher binned part (the 920 is the lowest binned Bloomfield), and likely runs at a slightly lower voltage.

1 more channel for 28W?!!

What about the on-die PCIE, NB and DMI of LGA1156? They are free of power?

The leakage current has nothing to do with performance or what.

The transistors applied is the only thing that has effect on the leakage current of transistors.

Nope. Try again.

Except that they don't.

Fair enough. You'll note that I'm not comparing the 980 to Lynnfield above though - I'm comparing Bloomfield (specifically the 920 most of the time) to Lynnfield.

Unreasonable replies!

You are just ignoring the fact that transistor is the only thing that has effect on leakage current. It has nothing to do with anything else.

More transistors means more leakage current given all the transistors are the same. You will only see more transistors with less current IF AND ONLY IF the transistors applied are more efficient than its competitor.

 

[cjl]

I didn't say LGA1156 had fewer (or the same) number of transistors. I said all transistors are not created equal, and it makes a HUGE difference what the extra transistors are doing.

As for why I keep mentioning the off-die NB? It's because you keep using full-system power consumption numbers to try to prove your point. The northbridge is part of the full system, and if you account for its extra consumption, it becomes obvious that the CPU alone in an LGA1366 system is not actually consuming a significant amount more power than in an LGA1156 system (assuming Lynnfield of course - 32nm duals will use much less power).

 

[andy5174]

I said all transistors are not created equal, and it makes a HUGE difference what the extra transistors are doing

Yes, the more efficient LGA1156 transistors has hugely less leakage current! HUDGE Difference.

Furthermore, the leakage current of a certain transistor is fixed! Otherwise datasheets of transistors are uselss.

It has nothing to do with performance or whatever.

As for why I keep mentioning the off-die NB? It's because you keep using full-system power consumption numbers to try to prove your point

That was my mistake by putting on the system power chart.

However, you kept mentioning NB of LGA1366 """FAR BEFORE""" I put those charts.
Proof:
page 3 of this thread, you said:

The same? No, but they are extremely similar. When running at the same voltage and clockspeed, Bloomfield and Lynnfield will use similar amounts of power. There are slight differences, in large part due to the additional memory channel and more power hungry northbridge, but the architectural differences are quite small between Bloomfield and Lynnfield chips.

 

[andy5174]

When we are talking about the (leakage current -> power lost -> heat), forget about everything but the "Transistos".

Applying more efficient transistors is the only way to build a circuit with MORE transistos but LESS leakage current.

 

[cjl]

The leakage current of a certain transistor is fixed at a given voltage and clockspeed. CPUs have multiple power and clock domains, and without knowing the voltage and clock speeds that the transistors are running at, you can't really extrapolate leakage.

LGA1156 lynnfield CPUs are made using the exact same 45nm HKMG process and transistors that Bloomfield (LGA1366) use, and exactly the same ones that the 45nm Core 2s used before them.

Do you have any proof that the 1156 CPUs are less power hungry at the same voltage and clockspeed by the way? I do mean CPU alone, so make sure any charts you post are the CPU by itself. You haven't provided any solid proof yet, and yet you are arguing as if the lower power consumption of the CPU alone is an accepted fact.

 

[cjl]

When we are talking about the (leakage current -> power lost -> heat), forget about everything but the "Transistos".

Applying more efficient transistors is the only way to build a circuit with MORE transistos but LESS leakage current.

If the additional transistors are not in the core clock domain, you can lower the clocks and voltage on the additional transistors and get less leakage (without having to lower the core clocks).

 

[andy5174]

The leakage current of a certain transistor is fixed at a given voltage and clockspeed. CPUs have multiple power and clock domains, and without knowing the voltage and clock speeds that the transistors are running at, you can't really extrapolate leakage.

LGA1156 lynnfield CPUs are made using the exact same 45nm HKMG process and transistors that Bloomfield (LGA1366) use, and exactly the same ones that the 45nm Core 2s used before them.

Do you have any proof that the 1156 CPUs are less power hungry at the same voltage and clockspeed by the way? I do mean CPU alone, so make sure any charts you post are the CPU by itself. You haven't provided any solid proof yet, and yet you are arguing as if the lower power consumption of the CPU alone is an accepted fact.

My reply:

Both TOM and Anandtech shows that LGA1366 runs much hotter than LGA1156 and no one has problem with that. ---1)

What is the reason of rise of temp? Heat

Where is heat come from? Lost of power

What is lost of power due to? Leakage current

What is leakage current dependent on? TRANSISTOR EFFICIENCY

Ok, I won't argue with you on the transistor type anymore as it will be endless debate unless I can show you the transistor spec which I can't.

Now, let's assume LGA1156 and LGA1366 has the same transistors.

Yes, leakage current is based on the voltage and frequency as well.

Hence, transistor has different characteristics(leakage current, EFFICENCY...etc) in different states. --- 2)

Conservation of Energy: Less heat(or power-lost) state is the higher efficiency state. --- 3)

The state of higher frequency with less heat is the higher efficiency state. --- 4)

i7-860 has higher frequency, more transistors and lower temps( -> less heat generated) than i7-920 AND 1), 2), 3), 4).

*[ Thus, 860 is in more EFFICIENT state compared to 920 and so it is still more efficient than 920 EVEN THEY HAVE THE SAME TRANSISTOR. ]*

p.s. You kept calling for proof through the whole thread. Where's your proof? Nothing, absolutely nothing since the start of this argument.

 

[andy5174]

If the additional transistors are not in the core clock domain, you can lower the clocks and voltage on the additional transistors and get less leakage (without having to lower the core clocks).

lower the clock? Why should we?

lower voltage? Yes, lower voltage give less power lost.
However, i7-860 still runs cooler( -> less power lost) EVEN with higher voltage, more transistors and higher clock than 920!!!!!

 

[jimmysmitty]

If what andy states is true, then a 920 D0 stepping would have the same efficiency since it came out about the ame time as LGA 1156 hit. I mean why would Intel change their process at all for one branch of CPUs? They both use the same 45nm HK/MG process.

Thats the thing with a process. Its easy to apply it to any arch you want. Hell they could probably take Netburst and use their 32nm HK/MG gen 2 process on it.

So it would make sense that any 920s sold would be just as efficient CPU wise since D0s have probably taken over.

 

[andy5174]

If what andy states is true, then a 920 D0 stepping would have the same efficiency since it came out about the ame time as LGA 1156 hit. I mean why would Intel change their process at all for one branch of CPUs? They both use the same 45nm HK/MG process.

Thats the thing with a process. Its easy to apply it to any arch you want. Hell they could probably take Netburst and use their 32nm HK/MG gen 2 process on it.

So it would make sense that any 920s sold would be just as efficient CPU wise since D0s have probably taken over.

Not at all. The D0 came out much earlier than the LGA1156.

i7-9xx D0 released in March/April, whereas, LGA1156 released in September which is 5 months later.

Is 5 months short? No, 5 months is very long for CPU industry. For example, you leap from 45nm to 32nm about another 5 months later than the release of LGA1156.

Do some research before arguing with others!

 

[andy5174]

You also made mistake here. Again, do some research before joining debate.

jimmysmitty wrote :

This is how it always goes with Intel. Highe end then slowly trickle down to lower parts. The Q6600 was the first quad and came out near $1K. Then the QX6700 hit and the Q6600 went to $550 and the QX6700 stayed at $1K. Then the QX6850, then the QX9650 and so on and so forth.


I am pretty sure that by years end we will see a sub $500 6 core Gulftown and possibly with Sandy Bridge, see 6 cores move to the mid end market much like the quads are now with 8 cores being the high end.

Actually, the QX6700 was first, at the beginning of Nov. 2006. The Q6600 came out in Jan. 2007 for $850 (rather than the $1k of the QX6700). It was cut to $530 in April, and $266 near the end of July (when the $530 Q6700 came out, along with the QX6850 for $1k).