|CPU:||Intel Core i7 920|
2.66 GHz, 2 MB Cache
|Cooling:||Noctua U12P 1336 Edition|
Kindly supplied by Noctua
|Motherboard:||Foxcoon Bloodrage X58, BIOS P07|
Kindly supplied by Foxconn
|Video Card:||PowerColor Radeon X800XL 256MB|
|Harddisk:||Samsung P80 80 GB|
|Power Supply:||Deluxe 500W|
|Software:||Windows XP SP2, Catalyst 9.3|
Performance & Overclocking
Before We Get StartedThe new Core i7 in combination with the X58 is quite different than previous generations of Intel CPU & chipset combinations. The fact that the tri-channel memory controller is on the CPU poses a few unique limitations for memory makers. While we always bench DDR3 between 1.5 V and 2.0 V, this spectrum is not possible with the Core i7. Intel advertises and only guarantees their CPUs with up to 1.65 V on the memory, because the memory controller has now been placed on the CPU die. This means that there is a lot less room for higher clocks and it will be interesting to see how the memory fares at this maximum allowed voltage. There is no point in pushing everything further, risking to fry our plattform or enticing you to risk yours, by presenting you with results usually not safely attainable.
It should be noted that overclocking on the Foxconn Bloodrage is an art of its own. While most boards will only give you options which tend to be within certain specification, the Bloodrage gives you all the choices within the BIOS. This means that you may set something which will make your system not boot, even if the components can take the overclock. One thing you should always watch when OCing an i7 is the fact that the frequency of the onboard memory controller should be set to twice the speed of the memory you are running. This is simply achieved by setting the multiplier twice as high for the controller compared to the memory. The CPU voltage has been bumped to 1.45 V and any automatic fan control turned off, so that the Noctua CPU cooler blows full force all the time. Preliminary testing of our specific Bloodrage and i7 CPU seem to point to a possible base clock of 200 with a CPU speed of 4 GHz. Just to make sure, various other voltages have been tweaked to the maximum to allow the best possible overclock on the i7 920 - in fact we are even pushing the north bridge voltage and the so called "UnCore" voltage into the red numbers, to make sure that the CPU & mainboard are not the ones holding us back.
Base SettingsThis section is new and illustrates what settings were used to attain the advertised speed of the memory sample in this review:
- Base Clock: 160 MHz
- CPU Multiplier: 20x
- CPU Speed: 3200 MHz
- CPU Voltage: 1.45 V
- CPU VTT (UnCore) Voltage: +220 mV
- X58 I0H Core Voltage: 1.24 V
- Memory Multiplier: 10x
Starting out with the settings above, we did a complete run to give you the numbers you can expect when using this memory. The 533 MHz overclock on the i7 920 seems to be a lot, but there are plenty of reports of the CPU hitting 4+ GHz on air. If your CPU however does not manage to boot at 3.2 GHz you do have the option of reducing the CPU multiplier. Only the onboard memory controller requires a 2:1 ratio to the memory. From this point we simply pushed the base clock and managed to raise it all the way up to 177 MHz. This means that the memory was running at 1774 MHz. This was the highest attainable memory speed. We even tried to relax the CL timings and raise the Voltage above specifications for a very short time. None of these changes yielded better overclocking results, so the memory was holding us back.
Then dropping the voltage of the memory to 1.5 V had another surprising result. The memory managed to go just as far on 1.5 V as it did on 1.65 V. The performance graph below reflects that fact.
Considering the fact that relaxed timings did not yield any positive effects, we went on to tighten the timings slowly, until the memory would no longer boot at 1066 MHz, which is the slowest speed the i7 plattform is rated for memory wise. The three DIMMs were always run at 1.5 V followed by 1.65 V to check if there was a difference in overclockablity. The Foxconn Bloodrage has an option for CL5-5-5 but the G.Skill Tri-Channel kit did not boot at this setting.
Relaxing the timings to CL6-6-6 did the trick and the memory managed to climb up to 1211 MHz at 1.5 V with an 8x memory multiplier. Once again, changing the Voltage to 1.65 V did not yield a different result. Changing the CL to 7 also meant being able to boot at 1333 MHz at 1.5 V. Doing so we changed the multiplier of the memory to 10x. The memory managed 1412 MHz at rating no matter what voltage was applied. Raising the timings also allowed us to pass the 1600 MHz mark at CL 8. Remember G.Skill advertises a CL9-9-9-24 setting, but we managed to run the three modules at CL8-8-8-20 with 1.5 V with a bit of room to spare. The memory maxed out at 1614 MHz - once again - no matter how much voltage we applied. Looks like G.Skill wanted to give the user the possibility to overclock the kit, instead of advertising tighter timings with no overclockability.
There is no reason to give you a Voltage Scaling graph with memory run between 1.5 and 1.65V - especially if there is no difference with our sample in this review. Instead, the above graph gives you a nice visual cue what you can expect from the memory at each base CL setting. Note that we included CL 10 as well, just for reference, to show you that setting this latency did not improve the possible overclock. Remember, this is represents our sample - your mileage may vary.