We’ve already reviewed Intel’s new Core architecture so this time it about overlocking and just overclocking and who would be better suited than our very own OC guru Robert ”crotale” Kihlberg.
The successor to Northwood, Prescott, threw a spanner in the works for many overclockers and despite big ambitions the performance wasn’t much to brag about. When the enthusiasts had grown tired of these models and moved to the other camp an anonymous little creation appeared out of nowhere. Intel’s mobile processor Dothan turned out to perform very well considering its relatively low frequency. Together with an adaptor that made it possible to use better motherboards the results started to appear. The processor had no problem measuring up to the best on the market, but the platform it used was starting to become hopelessly outdated without the possibility to use dual graphics cards. It took more than a year before all pieces would fall into place, and thus a new era for enthusiasts.
The connection between Dothan and Conroe, the code-name for the Core 2 processors, have been discussed earlier here at NordicHardware: Intel Core – The New Ruler. Now it’s time to take a look at the overclocking potential of these well performing processors. This is our third detailed overclocking article here at NordicHardware and we will thoroughly dissect Intel’s top model, Core 2 Extreme X6800, in the same manner as we’ve done with AMD’s FX-57 and Intel’s Pentium 4 660. As usual we will use a number of different cooling solutions ranging from regular air cooling to more extreme compressor setups.
We get right to it and present our test system.
|Motherboard||Intel D975XBX ”Bad Axe”, rev. 304, BIOS 1334|
|Processor||Intel Core 2 Extreme X6800 (Conroe, 0617)|
|Memory||Corsair XMS2 5400UL (D9 Fatbody)|
|Graphics card||nVidia GeForce 7900GT|
|Power supply||OCZ PowerStream 520W|
|Operating system||Windows XP (SP2)|
|Drivers||Intel Chipset Driver 220.127.116.113|
|Test program||SuperPi 1.1e|
As the host for our project we have Intel’s own motherboard, D975XBX, nicknamed ”Bad Axe”. Bad Axe was partly one of the first motherboards to officially support Intel’s new processors and partly one of the first motherboards from Intel that allowed significant overclocking. It’s not really ”Plug ’n’ Play” though, but to activate all of these settings you had to move a jumper on the motherboard. This might not sound all that bad, but Intel chose to make it a bit tricky by not soldering any pins for you place the jumper on. Time to bring out the soldering iron and fix this.
In previous articles we’ve only used SuperPi to illustrate the performance scaling of the processor in regard to what cooling and voltage. We will do the same in this article and even if the 1M calculations we will perform goes spectacularly fast with this processor they put a surprisingly great load on the processor, which will comment further later on. To increase the number of tests with each cooling setup we’ve a number of tests to show you even better what you have to gain from the overclocking.
We start with the most common type of cooling; air.
We didn’t get a heatsink for our engineering sample of the X6800 processor so we’ve chosen to use a common retail cooler: Zalman CNPS9500-LED. Because of the current season we’ve chosen to place a number of fans in the best possible manner to supply the system with cool air. The summer heat we couldn’t do much about, we just had to accept an ambient temperature of 27°C/81°F.
As we can see the 65nm processors start at a voltage of only 1.275V and you have to be careful not to get to eager with the voltages when you’re used to a 90nm processor. As we can tell from the diagram we have a good margin for overclocking already at the default voltage. With a few increases we’re already over 3.5GHz and thus a 600MHz overclock. Even though Intel has made progress when it comes to leaking currents from the 90nm processors the temperatures start to rise when we get close to 1.50V. This lead to that the overclocking possibilities stopped at right above 3.7GHz and a voltage of 1.50V.
We can conclude that there is plenty of overclocking potential already when using air cooling and that Intel most likely has a good margin for releasing an even faster processor as it has been very careful with the voltages for the processors. Normally SuperPi doesn’t say much about the stability of the processor, which isn’t quite true when it comes to the Core 2 processors. There were many cases where we could start up the system and load Windows at a higher speed than what we could perform a 1M calculation at. With these higher frequencies we could run other types of synthetic benchmarks successfully. The highest stable frequency for a 32M calculation was 3740MHz
We move on to water cooling.
The water cooling consists out of a triple 120mm radiator from Asetek, an Eheim 1250 pump, and a Swiftech Apogee block for the processor. Because of the high temperature we didn’t expect much from the overclocking potential. With the radiator at the window we managed to stabilize the water temperature to 26°C/79°F, which is far from fantastic. Let’s see what this resulted in.
As we earlier mentioned the Core 2 processors are not as hot as the previous NetBurst models. This is pretty clear from the diagram above as we don’t get much of a difference when moving from air to water. SuperPi 1M is finished pretty quickly, which then doesn’t reflect the true potential of the water cooling. Thanks to the few degrees we earned we managed to gain a little more though.
Where we really hoped the water cooling would surpass the air cooling was with the more demanding tests. A 32M calculation passed at about the same frequency as a 1M, which proves the higher capacity of the water cooling. Below we see an indication for the increase in performance compared to the stock frequency.
We can conclude that the processor doesn’t get especially hot with the voltages we’ve fed it with so far, and it doesn’t respond to any further increases either. Our hopes are left to the more extreme cooling and the really cold temperatures it brings. Time for compressor cooling.
To be able to keep the temperatures way below the room temperature it is today common to use compressor systems. These work just like the cooling systems of a freezer, but that these have been constructed to work continuously and the absolute lowest possible temperature. With the refrigerant we use with our unit, R404A, we can during ideal circumstances reach below -40C. The room temperature plays in a little with these units and during our test day we had temperatures between -42°C and -38°C during full load. At temperatures under room temperature condensation happens, which is something you don’t really want when working with electronics. Therefore we prepared the motherboards with insulation and vaseline to keep the cold inside and protect from any possible condensation.
As we can see, and have seen earlier, when the temperatures go down the frequencies go up, this time down by 70°C. After increasing the voltage a bit we sprint by the air and water cooling without any problems. With any voltage higher than 1.575V the processor makes no real improvements and we barely managed to go past 4400MHz. Let’s take a look at the performance with our test programs.
As we can see the performance improves quite nicely with the increased frequency and even though the X6800 processor is unchallenged already during default frequency it gets pretty scary when overclocked.
We have yet another compressor setup to show you, namely a cascade.
A cascade unit works by the same principle as the ordinary phase change unit, the difference being that we’re talking about two compressors that are connected serially. Because of this, we can reach temperatures below -100 degrees Celsius. Not only is condensation a more serious problem compared to the previous phase change unit, there are also a lot of components around the processor socket that really aren’t comfortable with temperatures like these. Capacitors as well as power transistors receive different electrical properties and if worst comes to worst, they stop working. This means that you can’t use too much insulation, as you’re enveloping these components with almost too low temperatures. The fans are now actually used to heat up the capacitors.
Cold definitely does wonders for this processor and we’re seeing the same frequency increase now as when we went from the water cooling to the phase change cooling on the previous page. Thanks to the colder temperatures, we’re also able to increase the voltage more, which is making a lot of difference even up to 1.625V. When we pass 1.65V there are only minor increases in the frequency, which means that we’re closing in on the maximum frequency this processor can handle at these temperatures. Another indication of the processor running at its peak is that we have to decrease the frequency to be able to run the heavier types of benchmarking applications.
We can see that the performance increase in WinRAR is starting to diminish, which is largely because of us not having overclocked the memory as much in proportion to the stock frequencies. Cinebench continues to scale well and makes good use of the increased frequencies. Eventually we reach the highest frequency we managed to run SuperPi 1M with, 4916 MHz (1.66V vcore), which resulted in a time of 10.469 seconds.
Let’s move on and analyze the results on the next page.
Here is all the data compiled into one diagram. We can see the wide margin of the starting frequency of 2933MHz already by just using air cooling and default voltage.
The CPU scales very well from 1.275V to 1.35V, giving an increase in frequency of about 200-250MHz – irrespective of the cooling used. After that, the curves for the air and water cooling start to level out and for the last 50MHz, some substantial voltage increases were needed. Using the phase change cooling we noticed a weak trend of plateauing at 1.50V followed by a good scaling up to 1.60V. Just as with earlier Intel CPU’s we notice a bigger need of lower temperatures than higher voltages when overclocking.
An overclock of about 800MHz of the fastest model in a processor series using air cooling is quite impressing. The Phase change systems allowed us to sharply increase the overclock and we finally reach an increase of 2GHz. This is mainly because of the optimizations Intel’s done in its 65nm manufacturing process, along with a number of functions for improving energy saving. The substantially lower heat dissipation is clearly noticeable with our cooling systems. Even with voltages over 1.6V our coolers had no problem keeping the temperatures straight. At 4.6GHz the load was big enough to keep CPU temperatures over -100C with our cascade rig.
Let’s summarize our overclocking journey with Core 2 on the next page.
We are seeing many similarities in relative terms how this Core 2 Extreme processor is overclocking in proportion to the processors we’ve tested before, especially the P4 660. Intel is continuing to develop and improve their manufacturing processes and we can also see that even though spectacular improvements have been made in addition to a complete turn when it comes to performance per cycle, the processors still have a pretty large margin of frequency.
The purpose of this article was to give an overview of what one can expect from their Core 2 processor. The stepping of the one we tested is 5/B1 and is the direct predecessor to stepping 6/B2, which is the one that is available in retail boxes. When it comes to overclocking, there’s not much of a difference from what we’ve seen around the net so far. One thing that we stated early was that our sample didn’t belong to the crowd that overclocks very well. We’ve seen examples of several CPUs, even cheaper ones like the E6700 and the E6600, that exceeded our results by over 200 MHz in every level of cooling. As the processors are getting more and more common, we’ll be able to make better general conclusions, but we still want to say that the results we’ve presented is on the small side of how well the Core 2 processors will perform.
The next logical method of cooling would be liquid nitrogen, and the attentive reader will remember that we’ve already overclocked this processor by those means in the beginning of summer, more specifically during DreamHack. What we can conclude from that session is that like previous processors from Intel, we won’t really hit any aggressive obstacles concerning low temperatures. Our processor managed to take temperatures as low as -130 degrees Celsius, with some variations depending on the voltage fed to it. In other words, so-called cold-bugs is not a problem with these processors.
We’ll continue to push these processors to their utmost limits here at Nordic Hardware, so keep an eye out for more performance results in the future! We also want to thank Intel for lending us the processors for this test.