Intel's Response For Q3 '07: Yorkfield

[Carfax]

yawn, more like a pocket monster...
basic math is Increase FP performance and add Cache liberally...
hmm almost expected.. Funny fact that most people miss is that K8 still beats Conroe in interger performance...

Are you kidding? Conroe slaughters the K8 in integer performance!

The K8L might be better than Yorkfield clock for clock in FP, but due to Yorkfield's clockspeed advantage, I think Yorkfield may have the upper hand overall.

i am abit dissapointed tho seeing that it won't share the l2

The cache is still shared, just not between all four cores. The reason is probably because it's difficult to get four cores to share cache effectively at the same time.

[Poodle]

Seems like an act of desperation to me. Intel must know that the Kenty sucks badly (Hallo Prescott!) We are seeing the Pentium D rutin again on the new quads from Intel. The non native thing seems to have a bad effect on power consumption. When Intel adds cores the heat multplies with the number of cores added. When Amd adds cores it stays in the same Tdp value as before the core addition.

And the k8L has some new advanced power feats. that will increase performance/w even more.


And then Intel have that fsb 1333 to worry about (it sucks). Amd will win easy at least on the server plattform, due to memory and bandwidth performance.

[accord99]

Seems like an act of desperation to me. Intel must know that the Kenty sucks badly (Hallo Prescott!) We are seeing the Pentium D rutin again on the new quads from Intel. The non native thing seems to have a bad effect on power consumption. When Intel adds cores the heat multplies with the number of cores added. When Amd adds cores it stays in the same Tdp value as before the core addition.

That's because Intel's TDP right now are much lower than AMD's. Even with the doubling, Kentsfield's power consumption is roughly the same as a FX-62.

[Helmore]

That's because Intel's TDP right now are much lower than AMD's. Even with the doubling, Kentsfield's power consumption is roughly the same as a FX-62.

There is one thing that you are missing, that is the fact that AMD is much more strict in TDP ratings then Intel. So if AMD would rate a CPU at 125W, then Intel would rate it at 100W. Those figures ought to be taken with a grain of salt, as I do not know how much of a difference there is, but it is quite a well known fact that AMD is stricter in thier ratings.

[vitaminc]

There is one thing that you are missing, that is the fact that AMD is much more strict in TDP ratings then Intel. So if AMD would rate a CPU at 125W, then Intel would rate it at 100W. Those figures ought to be taken with a grain of salt, as I do not know how much of a difference there is, but it is quite a well known fact that AMD is stricter in thier ratings.

Not more strict but different methodology. Intel is using RMS power, AMD is using maximum power.

[nn_step]

Not more strict but different methodology. Intel is using RMS power, AMD is using maximum power.

Tue, as well as AMD includes the Memory controller, Intel on the hand doesn't

[vitaminc]

Tue, as well as AMD includes the Memory controller, Intel on the hand doesn't

Memory controller doesn't consume much power or real estate.

If you really want to compare the actual power, you will have to take all CPU, MC, FSB/HT controller, PCIe switchs into account (aka include both CPU andd NB). Or even easier, measure power draw from the wall using the same power supply.

TDP is TDP, its for thermal mechanical design and does not matter where the MC is located. If you are packing 200W of heat on the CPU socket, its a bad thermal design compare to spreading the heat around CPU socket, NB socket, etc.

[Fred_Pohl]

There is one thing that you are missing, that is the fact that AMD is much more strict in TDP ratings then Intel. So if AMD would rate a CPU at 125W, then Intel would rate it at 100W. Those figures ought to be taken with a grain of salt, as I do not know how much of a difference there is, but it is quite a well known fact that AMD is stricter in thier ratings.

Since when is unsubstantiated conjecture a well known fact? Based on component wattage estimates a Woodcrest system consumes more power on paper than a comparable Opteron system, yet in real world testing the opposite is true.

I think you're confusing manufacturer specific TDP verbage with cpu family TDP ratings and reality. Obviously a 125W TDP FX62 consumes less than 125W just as a 65W TDP E6300 consumes less than 65W since neither cpu is at the top of their respective cpu family TDP ratings. At this time we can be fairly certain that a E6700 does consume 65W TDP but we still don't know if FX74 will be the top model in AMD's 125W family TDP, although it seems very likely.

[accord99]

There is one thing that you are missing, that is the fact that AMD is much more strict in TDP ratings then Intel. So if AMD would rate a CPU at 125W, then Intel would rate it at 100W. Those figures ought to be taken with a grain of salt, as I do not know how much of a difference there is, but it is quite a well known fact that AMD is stricter in thier ratings.

That's no longer the case:

http://www.xbitlabs.com/images/cpu/c...ut/power-2.png


A 31W TDP Yonah uses less power than a 25W Turion MT.

http://www.silentpcreview.com/article313-page5.html

[PetNorth]

That's no longer the case:

http://www.xbitlabs.com/images/cpu/c...ut/power-2.png

It depends on the link http://www.lostcircuits.com/cpu/low_e/5.shtml

A 31W TDP Yonah uses less power than a 25W Turion MT.

http://www.silentpcreview.com/article313-page5.html

I don't see MT, but ML (TDP 35W).

[accord99]

It depends on the link http://www.lostcircuits.com/cpu/low_e/5.shtml

Only the undervolted and slower EEs have lower power consumption. On the other hand, the E6700 barely uses half the power of the slower FX-62.

I don't see MT, but ML (TDP 35W).

The ML is a misprint, it's voltage is 1.22v which is indicative of a MT and it matches the power consumption of a MT in an earlier review:
http://www.silentpcreview.com/article300-page6.html

[uOpt]

There is one thing that you are missing, that is the fact that AMD is much more strict in TDP ratings then Intel. So if AMD would rate a CPU at 125W, then Intel would rate it at 100W. Those figures ought to be taken with a grain of salt, as I do not know how much of a difference there is, but it is quite a well known fact that AMD is stricter in thier ratings.

The TDP is not the power consumption of the processor.

The TDP is what the mechanical design of the casing can dissipate. It does not take into account what the die inside will actually take. That is why many AMD64 that have drastically different power consumption (same core same volts but at different clockspeeds) have the same TDP. Because it's the casing design and not the core.

Intel socket 775 and 771 CPUs have a lower paper TDP because they don't have as much surface (as much contact to the heatsink). But that doesn't mean the CPUs take more or less in practice.

To find out the actual consumption you have to measure. Neither Intel nor AMD specify that info. And it wouldn't make sense since as every overclocker know every die is different. We have all seen CPUs that take much more or less power at the same volts and frequency. In particular those CPUs that clock very well at low voltage already take a great deal of power more at that voltage that others.

This is why you cannot use TDP as the power consumption. It isn't. This is, BTW, one of the major reasons why the popular extreme PSU calculator is nonsense.

[nn_step]

Are you kidding? Conroe slaughters the K8 in integer performance!

The K8L might be better than Yorkfield clock for clock in FP, but due to Yorkfield's clockspeed advantage, I think Yorkfield may have the upper hand overall.

funny considering that K8 beats conroe at the same clock speed in
SM Cryptography ( 8.2% faster) and SM STREAM (4.3% faster) which If I remember correctly are both mainly integer programs that use lots and lots of memory

[accord99]

funny considering that K8 beats conroe at the same clock speed in
SM Cryptography ( 8.2% faster) and SM STREAM (4.3% faster) which If I remember correctly are both mainly integer programs that use lots and lots of memory

Funny considering a C2E scores 18.5 on the single-threaded SpecInt2006 benchmark while a 3GHz Opteron only scores 13.3. Cryptography is dependent on a few instructions and Stream is a memory test.

[nn_step]

Funny considering a C2E scores 18.5 on the single-threaded SpecInt2006 benchmark while a 3GHz Opteron only scores 13.3. Cryptography is dependent on a few instructions and Stream is a memory test.

rather unique since techinically conroe should be able to do 4 integer operations per cycle while AMD can only do 3...
Funny, then again conroe has 1 complex integer unit and 3 simple and AMD just has 3 complex

[Carfax]

rather unique since techinically conroe should be able to do 4 integer operations per cycle while AMD can only do 3...
Funny, then again conroe has 1 complex integer unit and 3 simple and AMD just has 3 complex

NN_step, this is overtly biased even for you..

Seriously, the P-M family and all it's derivatives are well known to be extremely strong in integer.

Even Yonah rocks the K8 in integer, let alone Core 2..

[Theli]

I'm wondering, what exactly are the advantages of reducing 90nm to 65nm and on to 45nm? Someone mentioned reduced power consumption.

[Carfax]

I'm wondering, what exactly are the advantages of reducing 90nm to 65nm and on to 45nm? Someone mentioned reduced power consumption.

Reduced power consumption, aswell as a higher transistor budget meaning they can add more features to the core, while not increasing the size.

Also, they can make more processors at lower prices.

All in all, process mastery is extremely important, and it's how Intel has managed to become such a giant in the semi-conductor industry.

[zabomb4163]

I'm wondering, what exactly are the advantages of reducing 90nm to 65nm and on to 45nm? Someone mentioned reduced power consumption.

http://www.intel.com/technology/sili...nm_silicon.htm

"More than 20 percent improvement in transistor switching speed or more than a five-fold reduction in transistor current leakage."

[3NZ0]

funny considering that K8 beats conroe at the same clock speed in
SM Cryptography ( 8.2% faster) and SM STREAM (4.3% faster) which If I remember correctly are both mainly integer programs that use lots and lots of memory

But the question is, does that have a noticable impact on the real world use?

[fhpchris]

I like aggravate discussions wiiiiiiiii

[metro.cl]

funny considering that K8 beats conroe at the same clock speed in
SM Cryptography ( 8.2% faster) and SM STREAM (4.3% faster) which If I remember correctly are both mainly integer programs that use lots and lots of memory

that means nothing, maybe a crappy car can beat a ferrari at 20km/h but the ferrari at 300km/h beat it.

not saying C2D is a ferrari nor amd is a crappy car.

K8L will be good and also intel will have a good cpu this "new" cpu is just a conroe with some minor features added + the process shrink.

After this one Intel will have a awsome chip with many new things

[[XC] hipno650]

Are you kidding? Conroe slaughters the K8 in integer performance!

The K8L might be better than Yorkfield clock for clock in FP, but due to Yorkfield's clockspeed advantage, I think Yorkfield may have the upper hand overall.



The cache is still shared, just not between all four cores. The reason is probably because it's difficult to get four cores to share cache effectively at the same time.

What i mant was that i wanted it to share it! look at conroe and how it benifits surely it would multiply with 4 cores!