Intel's Pentium Extreme Edition 965: The Last of a Dying Breed
by Anand Lal Shimpi on March 22, 2006 1:51 PM EST- Posted in
- CPUs
We were given a pleasant surprise by Intel at IDF just a few weeks ago: a fully functional and benchmark-able system based on Intel's next-generation desktop processor, codenamed Conroe. The performance that we saw was stunning and the power efficiency promised is a welcome change from the Pentium 4's NetBurst micro-architecture. But after leaving sunny California and returning to reality, we found another Intel chip in our labs; it just wasn't Conroe.
With the heatspreader on top of the chip, you could hardly tell it apart from a Conroe. It looked and smelled like one, but that's where the similarities end. While the fastest Conroe based processors will barely exceed 3.0GHz, this chip is clocked at a mighty 3.73GHz. Indeed, what Intel had sent us was no Conroe. It was the new dual core Pentium Extreme Edition 965, a chip that's being announced today, but one that won't actually be shipping in retail systems until next month. And yes, it's another very high clocked, deeply pipelined CPU from Intel. It's not the power efficient, high performing, competitive part that we want from Intel; it's yet another bump in the rocky road of NetBurst.
Needless to say, it's not a chip that we were too excited about, especially after having our appetites whet by Conroe earlier this month. But the new Extreme Edition did manage to surprise us in a number of areas, some of which may surprise you as well. When we last looked at the Pentium Extreme Edition 955, we found its power consumption to be a bit troubling. The move to 65nm didn't seem to do much for the Pentium 4, as AMD was still able to maintain a significant advantage in power consumption. As it turns out, the initial 65nm parts from Intel did not support all of the power management features that were introduced in later versions of Intel's 90nm silicon. Mainly, support for EIST (Enhanced Intel SpeedStep Technology) and the new C1E enhanced halt state, both of which were introduced over a year ago initially with C1E on the 5xxJ series of processors and later adding EIST on the 6xx series of processors. For whatever reason, neither technology made its way into the Presler EE chips (65nm dual core Extreme Edition). With the new Extreme Edition 965, Intel re-added support for C1E, but EIST is still not supported (note that C1E support was added to later steppings of the EE 955, just not the initial stepping). The two technologies vary as follows (taken from our Pentium 4 6xx review):
C1E - Enhanced Halt State
Note that AMD is penalized a bit here because we're testing on an ASUS A8N32-SLI, which effectively features two North Bridges (one masquerading as a South Bridge) to offer two dedicated x16 slots. The two chips together consume more power than NVIDIA's 2 x8 single chip design, so depending on what chipset that you use, AMD's power consumption could be a bit lower.
Also, keep in mind that AMD's Cool'n'Quiet is enabled, and the EE 965's C1E support is kicking in to keep its power consumption low at idle as well.
Under load, the comparison is also very close:
With power consumption finally very close between AMD and Intel, who is the true winner here? From AMD's perspective, the FX-60 is a 90nm chip that still consumes less power than Intel's 65nm CPU, and its power consumption will only go down further when AMD ramps up its 65nm process. From Intel's perspective, AMD won't be producing 65nm in volume until 2007, the same year when Intel will start playing with 45nm chips. Not to mention that the EE 965 is made up of 376 million transistors compared to AMD's 233 million, running on a much more power-hungry architecture. So to Intel, being able to almost equal AMD's power consumption on a NetBurst based part is pretty impressive. From our perspective, it's nice that the EE 965 brings power consumption down to a more reasonable level, but what really matters is how well it performs at that relatively similar power envelope. And that's what we'll find out next.
In our first look at Intel's Presler core with the Extreme Edition 955, we investigated the benefits of Hyper Threading with a dual core CPU as well as the performance impact of Presler over Smithfield. So, be sure to consult that article if you want a more in-depth look at Intel's 65nm dual core desktop CPU.
With the heatspreader on top of the chip, you could hardly tell it apart from a Conroe. It looked and smelled like one, but that's where the similarities end. While the fastest Conroe based processors will barely exceed 3.0GHz, this chip is clocked at a mighty 3.73GHz. Indeed, what Intel had sent us was no Conroe. It was the new dual core Pentium Extreme Edition 965, a chip that's being announced today, but one that won't actually be shipping in retail systems until next month. And yes, it's another very high clocked, deeply pipelined CPU from Intel. It's not the power efficient, high performing, competitive part that we want from Intel; it's yet another bump in the rocky road of NetBurst.
Needless to say, it's not a chip that we were too excited about, especially after having our appetites whet by Conroe earlier this month. But the new Extreme Edition did manage to surprise us in a number of areas, some of which may surprise you as well. When we last looked at the Pentium Extreme Edition 955, we found its power consumption to be a bit troubling. The move to 65nm didn't seem to do much for the Pentium 4, as AMD was still able to maintain a significant advantage in power consumption. As it turns out, the initial 65nm parts from Intel did not support all of the power management features that were introduced in later versions of Intel's 90nm silicon. Mainly, support for EIST (Enhanced Intel SpeedStep Technology) and the new C1E enhanced halt state, both of which were introduced over a year ago initially with C1E on the 5xxJ series of processors and later adding EIST on the 6xx series of processors. For whatever reason, neither technology made its way into the Presler EE chips (65nm dual core Extreme Edition). With the new Extreme Edition 965, Intel re-added support for C1E, but EIST is still not supported (note that C1E support was added to later steppings of the EE 955, just not the initial stepping). The two technologies vary as follows (taken from our Pentium 4 6xx review):
C1E - Enhanced Halt State
"Whenever the OS executes the halt instruction, the CPU enters what is known as the halt state. Architecturally, what's going on in a halt state is the clock signal is shut off to the CPU for some period of time. With no clock signal, none of the logic in the chip will do anything and thus power consumption is reduced. Performance is also significantly reduced; however, the halt instruction isn't usually called during application usage, so the performance aspects of the halt state aren't very important.EIST:
The problem with the halt state is that it does nothing to reduce voltage, only current draw by stopping clocks from going to the CPU. Since Power varies linearly with both current and voltage (P = I * V), you're effectively only addressing half of the problem. The Enhanced Halt State, as Intel calls it, does two things: it reduces the clock speed of the CPU by decreasing the clock multiplier down to its minimum value (on the EE 965 series, that's 14x, or 2.8GHz), then reducing the voltage. The clock speed is reduced and then the voltage is dropped, to maintain stability.
Intel insists that the enhanced halt state is a significantly lower power state than the conventional halt state, thanks to the reduction in voltage in addition to the reduction in clock speed. While the standard halt state causes a linear reduction in power, Intel's enhanced halt state causes an exponential decrease in power, potentially offering better power savings than the standard halt state. The real world impact obviously depends on how idle your system happens to be."
"What EIST does is very similar to AMD's Cool'n'Quiet. It is demand based reduction in CPU clock speed and voltage. Using the same mechanism of adjusting clock speed and voltage, based on the application demand, the processor will dynamically increase/decrease its clock speed between its minimum clock and its normal operating frequency, as well as voltage, in order to optimize for power consumption.Because of the support for C1E and Intel's 65nm process, power consumption is finally competitive with AMD. Let's have a look at idle power consumption first:
Because of the way EIST (and AMD's Cool'n'Quiet) works, there's inherently a drop in performance. The idea is this: if you're performing a task that's not using 100% of the CPU, the CPU will operate at a slightly reduced frequency in order to conserve power. So, while some tasks will require that the system run at full speed, others will run at lower speeds. "
Note that AMD is penalized a bit here because we're testing on an ASUS A8N32-SLI, which effectively features two North Bridges (one masquerading as a South Bridge) to offer two dedicated x16 slots. The two chips together consume more power than NVIDIA's 2 x8 single chip design, so depending on what chipset that you use, AMD's power consumption could be a bit lower.
Also, keep in mind that AMD's Cool'n'Quiet is enabled, and the EE 965's C1E support is kicking in to keep its power consumption low at idle as well.
Under load, the comparison is also very close:
With power consumption finally very close between AMD and Intel, who is the true winner here? From AMD's perspective, the FX-60 is a 90nm chip that still consumes less power than Intel's 65nm CPU, and its power consumption will only go down further when AMD ramps up its 65nm process. From Intel's perspective, AMD won't be producing 65nm in volume until 2007, the same year when Intel will start playing with 45nm chips. Not to mention that the EE 965 is made up of 376 million transistors compared to AMD's 233 million, running on a much more power-hungry architecture. So to Intel, being able to almost equal AMD's power consumption on a NetBurst based part is pretty impressive. From our perspective, it's nice that the EE 965 brings power consumption down to a more reasonable level, but what really matters is how well it performs at that relatively similar power envelope. And that's what we'll find out next.
In our first look at Intel's Presler core with the Extreme Edition 955, we investigated the benefits of Hyper Threading with a dual core CPU as well as the performance impact of Presler over Smithfield. So, be sure to consult that article if you want a more in-depth look at Intel's 65nm dual core desktop CPU.
41 Comments
View All Comments
AnandThenMan - Thursday, March 23, 2006 - link
"Would you have preferred we not have this review at all?"What are you suggesting? That the review can't be published unless Conroe benches are in there as well? Let's be very clear here. Anandtech is publishing Intel sanctioned benches as the gospel truth. Anandtech is publishing benchmarks on an un-released product, running on unknown hardware, unknown settings, under restricted testing conditions, and passing them off without question at all.
If this is the way Anandtech plans to do all articles from now on, then I want to see benches that include future products from ATI, NVIDIA, and others as much as possible. And it won't matter if ATI or NVIDIA supplies the hardware, doesn't allow anyone to see it, and puts restrictions on what benches can be run and for how long. Anandtech will STILL publish these benches and assure us all they are 100% accurate.
It is very, very dissapointed to see Anandtech pull this crap. Conroe could very well live up the the hype and more. That is not the point. Integrity of testings means you control the testing environment, you control the hardware, and you control the software. And if you DON'T, then you SPECULATE and make it very clear that the results are speculation. You don't announce, "Intel takes back the performance crown" based on tests Intel let you perform. Very poor job.
AnnonymousCoward - Friday, March 24, 2006 - link
You're overreacting. Any reader knows that the original Conroe tests couldn't be 100% verified, since Anand made that very clear. So then you read this review, and you keep that in mind. Big whoop.Thanks, Anand, for showing the Conroe numbers. It's information that any sane builder would want to see, even though the numbers haven't been independently confirmed.
PrinceGaz - Thursday, March 23, 2006 - link
"If this is the way Anandtech plans to do all articles from now on, then I want to see benches that include future products from ATI, NVIDIA, and others as much as possible. And it won't matter if ATI or NVIDIA supplies the hardware, doesn't allow anyone to see it, and puts restrictions on what benches can be run and for how long. Anandtech will STILL publish these benches and assure us all they are 100% accurate."It would be a bit pointless for the companies to supply the hardware, but not allow anyone to see it and put restrictions on the benchmarks run. It would be far more sensible if the companies simply ran the benchmarks on future products themselves, and provided Anand with the figures to publish :)
Zebo - Wednesday, March 22, 2006 - link
Anand is doing intel's dirty work. It's called osborne effect - leak an upcoming products when you have nothing to undercut competitors sales. Brillaint move by intel - and Anand is being played like fiddle.Zebo - Wednesday, March 22, 2006 - link
How do you know there was'nt a 3.6 Ghz conroe in there?cornfedone - Wednesday, March 22, 2006 - link
SOS, DD.They ain't got a clue.
ksherman - Wednesday, March 22, 2006 - link
not sure if it is real or not, but it goes down in my bookas one sweet word: "obsolescence" (found in the closing paragraph)Chadder007 - Wednesday, March 22, 2006 - link
Who in their right mind would purchase this processor??? ....on yeah, the one that buys the new $10,000 Dell. :kekekeke:jojo4u - Wednesday, March 22, 2006 - link
EIST was introduced in the 6xx series and not in the 5xxJ as the article suggests.Anand Lal Shimpi - Wednesday, March 22, 2006 - link
You are correct, I've made the appropriate change to the article :)Take care,
Anand