Intel E6400 2.13 GHz
Samsung P120 250GB SATAII
NEC DVD Burner & floppy drive
Corsair XMS2 DDR2 PC2-5400 2 x 1GB
Gigabyte GF6200TC 128 MB RAM (256MB with System memory)
PCI WiFi card (B/G)
Scythe Ninja + with Nexus 120mm
Arctic Silver 5 thermal paste
Antec P180 with Silverstone ST30NF (Fanless)
2 Nexus 120mm fans – one in lower chamber, the other as exhaust on the side of the case opposite the CPU
Windows XP Pro SP2
All three of the fans were set to 45% using Speedfan which reported the speeds at ~ 620 RPM. The hard drive was resting on acoustic isolation foam and located in the bottom chamber of the case; AAM was set to the quietest mode.
The stock VCore for both CPUs is 1.325V at full speed and 1.15V when Speedstep drops the speed to 1.6 GHz.
The BIOS doesn’t allow the VCore to be set lower than 1.225V. I was able to under-volt using CrystalCPUID, but no lower than 1.16V.
You can’t set a voltage manually in the BIOS and still use Speedstep.
I measured the power consumption at three states: idle, running two instances of CPU Burn-in and running two instances of Prime95 (In-place large FFTs (Max heat & power consumption)). The information below is in the format:
GHz, VCore, Watts (Idle, CPU Burn-In x2, Prime95 x2), Ambient Temp, CPU Temp, FSB (if not stock), No of Hours Tested Stable with dual Prime95.
xx is used where no reading was taken.
1.60 GHz / 1.150V / 088 097 105 Watts / 25 xx
2.00 GHz / 1.150V / 088 100 111 Watts / 25 37 / 250 / 2
2.13 GHz / 1.150V / 088 102 112 Watts / xx xx / FAIL
2.13 GHz / 1.325V / 100 115 126 Watts / 25 xx
2.80 GHz / 1.325V / 103 124 135 Watts / 26 48 / 350 / 12
2.90 GHz / 1.325V / 103 124 136 Watts / 22 46 / 362 / 2
3.00 GHz / 1.325V / FAIL
3.00 GHz / 1.400V / 111 137 149 Watts / 26 55 / 375 / 1
3.20 GHz / 1.450V / 119 150 163 Watts / 26 60 / 400 / 1
In the last two tests I had to raise the CPU fan speed in an attempt to keep the CPU temperature below 55C. Even at 100% it couldn’t keep the temperature from rising uncomfortably.
I suspect that the thermal paste may not have been optimally applied due to the fact that I swapped out the CPU whilst the motherboard was still in situ. I found it very difficult to install the Ninja in this scenario and the paste may have been spread slightly away from the optimal contact zone. If I was keeping the motherboard I would certainly have removed it from the case to install the Ninja.
I didn’t have the time to test every permutation of CPU speed and VCore fully, so it may be possible to squeeze a bit more efficiency from this particular CPU.
I did some testing with an E6300 initially and when running that at 2.80 GHz with stock VCore the system consumed 5W more than the E6400 when running dual Prime95. This is due to the FSB needing to be run at 400 versus 350 for the E6400. The E6300 has a multiplier of 7 versus 8 for the E6400.
One thing to consider when over-clocking a C2D is that motherboards using the P965 chipset don’t currently allow the RAM to run slower than the FSB. The Asus board that I used supports running the RAM/FSB with a ratio of 3:4. (It also supports running them synchronously and with the RAM faster than the FSB).
The advantage of this is that you can run an E6300 at 2.8 GHz using inexpensive PC2-5300 RAM. In fact, even at 3.2 GHz you can still use PC2-5300 as it will only be very slightly over-clocked.
In a P965 based board you’d need to use PC2-6400 (800), as 2.8 GHz with an E6300 equates to a FSB of 400.
PC2-5300 typically uses 1.8V, whereas PC2-6400 is typically 2.1V, so you pay a double price for the higher speed RAM, in $ and Watts.
The downside to the 975X chipset is that it’s built on an older process than the P965, so should consume more power. Hopefully, one of the forthcoming non Intel chipsets will also allow a RAM/FSB ratio of 3:4 and be manufactured at 90nm. That would seem like the best option. Although, it’s possible that the P965 based boards may get BIOS updates that allow the RAM to run slower than the FSB, unless this is a limitation of the chipset!
To use Speedstep and over-clock with this setup you have to use the stock voltages, which limits you to the range 2.0 to 2.9 GHz; 2 GHz being the highest stable speed at 1.15V and 2.9 GHz being the highest stable speed at 1.325V.
For the E6400 used this implies a FSB of 1333, which gives a Speedstep range of 2.0 – 2.67 GHz and an optimal RAM speed of 667.
For an E6600 (2.4 GHz, multiplier = 9) at the same FSB you will have a Speedstep range of 2.0 – 3.0 GHz. It’ll be hit and miss whether it will be stable at that speed on stock Vcore, so you may need to back off the FSB to make it stable. I’m not sure if the extra cache of the E6600 and higher will make a difference to over-clocking stability!
It might also be possible to over-clock more aggressively using CrystalCPUID. By this, I mean that you use custom Speedstep voltages set with CrystalCPUID and then over-clock the FSB to a higher level than is stable using stock Speedstep voltages. You obviously choose a higher FSB value that is stable with the CrystalCPUID voltages that you set.
The danger in doing this is that the system will be running at an unstable setting for the 30 – 60 seconds or so it takes for Windows to load and CrystalCPUID to take over the voltage settings. If the settings chosen aren’t that aggressive and are unstable at the lower Speedstep frequency rather then the higher one, then I don’t think it will be such an issue. I suggest that stability at the higher Speedstep frequency with stock Vcore is more important, as whilst Windows is loading the system is more likely to be running at the maximum clock speed as it is under load.
In fact, even though the E6400 that I tested failed at 2.13 GHz at 1.15V, this test was conducted with both cores running at 100%. In everyday usage, if this system was over-clocked, it would never run at 2.13 GHz at 100% load, since Speedstep would bump it up to the max clock speed before that happened. This makes it more likely that the system would be stable in the period before CrystalCPUID took charge of voltages.
As an example for the E6400:
It was stable at stock Vcore at 2.9 GHz (FSB = 362).
This gives a lower Speedstep clock speed of 362 * 6 = 2.17 GHz.
It was not Prime95 stable at 2.13 GHz with stock Vcore, but was at 2.0 GHz
Well, there’s only one way to find out
Conclusions:
Cooling a C2D silently with this setup is easy provided you keep the VCore at stock.
Using Speedstep will reduce your idle power consumption by up to 15W using stock VCore values.
If you disable Speedstep and over-clock by increasing the VCore, your idle power consumption will climb very quickly for little performance gain.
E.g. 2.9 GHz with Speedstep = 2.17 GHz @ 1.15V = 88W
3.2 GHz @ 1.45V = 119W.
That’s a 31W increase at idle for a < 10% clock speed increases. The wattage difference at full load is slightly less at 27W.
Note. The difference would be slightly less, as in this scenario you would need to increase the Vcore with CrystalCPUID to gain stability. (See above).
If you are using Speedstep then you don’t need a motherboard that supports memory dividers, as a FSB of 1333 is about as far as you can push it at stock VCore values. This is synchronous with PC2-5300.
Higher FSB and RAM speeds consume more power for the same CPU clock speed, for typically little performance gains.
If you want to go for a large over-clock with an E6300 and a P965 chipset motherboard, you will need very fast and expensive RAM. It would probably end up costing you more than using a 975X based motherboard with slower RAM and very likely consume more power.
A power consumption comparison of the different chipsets for C2D would be useful. ASRock have a very cheap VIA based C2D board that might well consume little power. But, I assume it doesn’t support Speedstep, so it will be at a 12 – 15 W disadvantage at idle.
The system at almost all times during testing was quiet to the point of being ‘almost’ totally inaudible, even after midnight when the ambient noise floor is very low. The only time the noise level rose was when I increased the CPU fan speed due to over-volting the CPU in the last two tests.
Notes.
The power consumption of Prime95 at its most aggressive settings is I suggest atypical for most systems. The dual CPU Burn-in power values may well be more typical. This will lead to lower CPU temps than reported also.
The Asus P5W DH Deluxe uses the Intel 975X chipset and is fully featured; Crossfire support, eSATA, WiFi, Firewire, 2 Gigabit NICs, 2 PATA connectors. It uses heatpipes for cooling and comes with a fan that can be optionally used to improve cooling. It has 5 fan headers (one 4 Pin), at least 3 of which adjustable with Speedfan; there’s a fourth Speed control in Speedfan but I didn’t test to see whether it controlled the 2 fan headers that weren’t utilised.
