Overclocking

As cooling solutions do a better job of keeping the CPU at a lower temperature, it is reasonable to expect the overclocking capabilities of the CPU will increase. In each test of a cooler we measure the highest stable overclock of a standard X6800 processor under the following conditions:

CPU Multiplier: 14x (Stock 11x)
CPU voltage: 1.5875V
FSB Voltage: 1.30V
Memory Voltage: 1.90V
nForce SPP Voltage: 1.35V
nForce MCP Voltage: 1.7V
HT nForce SPP <-> MCP:
Auto

Memory is set to Auto timings on the 680i and memory speed is linked to the FSB for the overclocking tests. This removes memory as any kind of impediment to the maximum stable overclock. Linked settings on the 680i are a 1066FSB to a memory speed of DDR2-800. As FSB is raised the linked memory speed increases in proportion. The same processor is used in all cooling tests to ensure comparable results.

Highest Stable Overclock (MHz)

The Scythe Ninja Plus B with the stock fan could only reach 3.83GHz overclock with stability. This is a little better OC than the OCZ Vindicator with the stock fan and matches the performance of the Scythe Infinity with a single stock fan. It is not, however, performance comparable to the top tier of the heatpipe towers tested.

However a change to the 72 CFM 14 dB-A SilenX IXTREMA 120 allowed us to push the overclock to 3.90GHz with complete stability. This does match the top tier of overclocks achieved by the Tuniq Tower 120, Thermalright Ultra 120/Scythe SFLEX, OCZ Vindicator (with SilenX), and Scythe Infinity with dual push-pull fans.

The only air cooler that has reached higher than 3.90GHz is the Thermalright Ultra 120 Extreme, which was featured at CeBIT. We managed 3.94GHz with that top air cooler. The hybrid TEC/air Monsoon II Lite reached 3.96GHz in benchmarking, but that cooler combines active TEC cooling with passive air cooling and would be expected to reach higher overclocks than an air alone solution.

Noise

For many enthusiasts upgrading cooling the goal is maximum stable overclock, and they will live with the inconvenience of a louder system. For other users silence is the most important factor, and these users will forgo maximum overclocking if that increases system noise levels.

We measured noise levels with the Scythe Ninja Plus B with both stock and SilenX IXTREMA fans under both load and idle conditions. The fans have specified noise ratings of 20.9 dB-A and 14 dB-A, and the measured noise with both fans was below our system noise floor at both 6" and 24" positions above the open side of our system case.

There are virtually no power supplies that do not include a fan. While Zalman and a few others do make an expensive fanless power supplies, we have not seen a fanless unit larger than 500W, or one that would be used for seriously overclocking a system. With that in mind the noise level of the system with all fans turned off except the power supply was measured. The power supply used for the cooling test bed was the OCZ PowerStream 520, which is one of the quieter of the high performance power supplies.

We have also measured the Corsair 620-watt and Mushkin 650-watt power supplies which are reported to be quieter than the OCZ. Both the Corsair and Mushkin are indeed quieter at idle or start up speed. However, as soon as load testing begins and the PSU fan speed kicks up the measured noise level is almost exactly the same as the OCZ PowerStream 520 watt power supply.

We are currently in the process of evaluating "quieter" power supplies for an update to our cooler test bed. We will make changes to that test bed as soon as we are confident in the noise measurements and test procedures with a variable speed quiet PSU. We plan to evaluate additional power supplies and configurations in our upcoming 120mm fan roundup, at which point we will complete the transition to a revised and lower noise cooler test bed.

The noise level of the power supply was 38.3 dB from 24" (61cm) and 47 dB from 6" (152mm). The measured noise level of the test room is 36.4 dB, which would be considered a relatively quiet room with a noise floor slightly below the OCZ PowerStream 520 PSU.

Noise Level - 6

Noise Level - 24

Measured noise levels in this chart should be considered worst case. Measurements were taken with an open side of a mid tower case 6" and 24" from the HSF. Real world would be a completely closed case resulting in a further reduction in noise.

Any 120mm fan that is a standard 25mm thick should be mountable on the Scythe Ninja Plus B. The fan clips connect to the outside mount hole, which means both open post and the more common closed post fans will work properly. The stock Ninja fan is closed post and the SilenX is open post. This means the SilenX will work on any cooler we have reviewed thus far - including the Thermalright Ultra 120 and Ultra 120 Extreme.

Scaling of Cooling Performance Final Words
POST A COMMENT

36 Comments

View All Comments

  • Lord Evermore - Saturday, April 14, 2007 - link

    How is the Ninja and Infinity "not particularly great for overclocking", when even with their stock fans, when they can hit 3.83GHz? 3.9GHz is only 66MHz higher, a whopping 1.7% higher. Individual CPU differences could easily make that up for other users, and the "better" coolers might not get an equal increase in their maximum speed in those cases. At those levels to me things seem for all intents and purposes essentially equal.

    With the stock fan that's a 31% overclock at 3.83G. 33% at 3.9G. Only in comparison to the very best you've tested do these come out near the lower end of the scale. Right at the end you group it together with the top tier of other coolers. And even if it was at the lower end of the "good" coolers, stock or overclocked, the actual numerical differences are so tiny as to be statistically insignificant.

    I suppose 66MHz difference is a HUGE deal to rabid overclockers who are willing to spend money for it, but the majority of even enthusiasts I'd think wouldn't consider it particularly impressive given how large the stock fan overclock is. (Note I'm only addressing the frequency obtained since that seems to be the focus of what makes a cooler "good" here. Temperatures obtained at those speeds could change the lineup.)
    Reply
  • Wesley Fink - Monday, April 16, 2007 - link

    We use the exact same CPU and test bed for testing all coolers - to remove the "individual CPU" variable. But you are right to put the performance results into perspective. In the grand scheme of performance the difference in 3.83GHz and 3.90GHz is very small as you point out. There are other advantages to the very top performing coolers, however. If you check the cooling at various temps those that do 3.90GHz easily also generally provide the lowest CPU temperaturesa at any given speed.

    Had we devised our overclocking test a little differently and tested say, every 266 MHz - 2.93GHz, 3.2GHz, 3.466GHz, 3.633GHz, 3.80GHz, 3.966GHz - the true differences might be put in better perspective. However, we started the test with the assumption that if a cooler couldn't outperform the Intel Retail there was no point in buying it. Perhaps if the Intel Retail HSF were poorer we could show more, but the fact is the Intel Retail HSF is really very good and it takes a decent cooler to beat the one that comes at no additional cost with the Intel processor.

    Your comments are fair, but we also believe our cooler test methods do identify the very top performers in the marketplace. One notch down is still pretty remarkable performance as you point out.
    Reply
  • strikeback03 - Thursday, April 12, 2007 - link

    It is kind of hard to tell in the photos, does one set of 3 heat pipes run above the other set, or are they somehow woven together so they are all approximately the same distance from the CPU? Assuming they are mounted with one set above the other, did you do any testing to see whether cooling is better with the fan mounted on one side or the other (i.e. blowing directly on the upper set of heatpipes or the lower set). Reply
  • Wesley Fink - Friday, April 13, 2007 - link

    one group of 3 heatpipes passes through an upper plate and the other three pass through the mounting plate below. As you guessed they do cross over the lower set of heatpipes in the center of the base. We did not test with different orientations relative to the heatpipe level, but it is an interesting question. Reply
  • Wesley Fink - Thursday, April 12, 2007 - link

    We have received the following statement from Scythe regarding the similarity of their Ninja Plus B Cooler to the OCZ Vindicator:

    "Scythe is a stock holder of the factory where we produce our coolers,
    and also the factory is a stock holder of the Scythe Taiwan office where
    we have a very tight relationship almost as close as a group company.
    Within this relationship we are acknowledged and aware of all shipments
    done from the factory. But OCZ is not one of them.

    With the above said, we are definitely sure that OCZ coolers are manufactured
    in another factory. The only components that are common between the two are the
    hex caps on the heat pipes, which are supplied outside of our cooler factory.

    We also have not approved OCZ's design in any form or shape, nor did we receive
    any OEM request. Their products are just a simple copy of Ninja and we are
    currently being troubled by this untrue rumor. "
    Reply
  • Lord Evermore - Saturday, April 14, 2007 - link

    My thought is maybe they use a different working fluid. They're just too similar in design for the hardware to have any performance difference, even if OCZ did just copy the design (if they had though, you'd think Scythe would be suing). I suppose the exact alloy composition for the base plate or other parts could be different, if they actually are made somewhere else, and then the parts just end up looking the same and being assembled in the same shape.

    Then again, Scythe doesn't say NOBODY gets this design but them...
    Reply
  • ceefka - Thursday, April 12, 2007 - link

    On the packaging of the Ninja it says: "supports fanless model". Now that of course will be dead silent, but it will require adequate airflow in the form of case fans. What are the temperatures and dBs in that situation? Reply
  • Wesley Fink - Thursday, April 12, 2007 - link

    Fanless cooling seems to perform best with a PS with a downward facing exhaust fan and excellent case entry and exhaust cooling. Our test case does not really have these features, but we will definitely be paying more attention to airflow and the PS fans in choosing components for our new test bed. With the current configuration in mind, it seemed unfair to test fanless mode temps with a poor configuration for fanless cooling.

    We did run a quick 5-minute gaming loop at the X6800 stock speed with a fanless Ninja Plus B and it ran fine. Temps were much higher than fan configurations just as you would expect. We will be doing more in testing fanless claims in the future - once the fans are selected from testing and the new test bed is launched.

    We do confess concerns over fanless given the current crop of excellent perfroming tower coolers. It appears fanless just moves the fans from the cooler to the case, but fans are still required. There are many large quiet case fans, but there are also many quiet 120mm fans for coolers these days. Fanless seemed to make more difference when ccoolers used buzzy loud 60mm and 70mm fans. These days the best coolers use heatpipes and 120mm fans and the noise levels are MUCH lower with these coolers than the screaming wonders of the past.
    Reply
  • fic2 - Thursday, April 12, 2007 - link

    I am curious about the Noise Limit heatsinks: http://www.noiselimit.com/index2.asp?id=16">http://www.noiselimit.com/index2.asp?id=16. Any chance this will be included in a review? Reply
  • Stele - Wednesday, April 11, 2007 - link

    I'm not sure if this question has been dealt with before - my apologies if it has.

    I've always wondered how is it that a certain cooler can produce lower CPU temperatures but yet fail to make it at a higher CPU frequency? The Zalman CNPS9700 is a good example - at 3.83GHz the CPU runs at 40°C, which is lower than many of the others, yet does not make it to the last two CPU frequencies. A Scythe Infinity dual, however, registers 44 at the same frequency and yet makes it at least into 3.90GHz. The anomaly is even more pronounced in the stress test. In all cases the coolers seem to scale similarly, so it's not as if one of them had reached its design limit (compared to, for example, the stock Intel HSF). Has an explanation for this been given or proposed before?

    I'm guessing that one factor is the mounting method. It's remarkable that some Socket-775 coolers, despite weighing some 1.5-2x that of the official 450g limit with fan, still use the stock plastic push-pin method. Without a backplate. Makes one wonder sometimes whether the cooler is indeed in good contact with the CPU, and whether after some time the cooler would fall out - or the motherboard flex and crack.
    Reply

Log in

Don't have an account? Sign up now