Overclocking, and Performance

Now, normally I am one to jump in and tinker with the BIOS settings of a new motherboard. But I was curious to see how the OC genie button worked, and how much of an overclock it was able to achieve. The process really could not be any simpler. Simply power down the machine, open up the side of the case, press down the OC genie toggle button, then power back up. When you power the system back up again there is a series of quick test power ups, followed by a screen after post saying the OC genie is active, and that it isn't recommended that any settings in the BIOS be changed. Then you are on your way.

Once into windows I loaded up CPUz, which indicated I was running at a nice overclock of 3.36GHz. The same process for stability testing was used as before. To ensure total stability Prime95 sessions were extended to 8 hours. There is a few things to mention when looking at the overclock results. First, the idle temperature.

You will notice it is a good deal higher at idle than the stock 27C. I attribute this number to the speedstepping being turned off. The lower power draw from the CPU when the multiplier is lowered allows the CPU to run at very low temperatures. Combine this with the higher core voltage value and you get a bit higher idle temperature. I then turned on Prime95 and monitored the temperatures for an hour.

As you can see the max temperature reached on core 1 was 64C. Note how long it took to reach max values, the longest period it took to peak on each core was about 20 minutes. After that point there was no change in max values. This is very impressive performance from a CPU cooler that comes in below 40 bucks after taxes.

I decided to run some synthetic benchmarks after the overclock was in place to see what sort of results the overclock yielded. The three tests used included Cinebench 10, Valve Particle Simulation, and Valve Map Compilation Benchmark.

Valve developed a particle simulation benchmark within the source engine that fully utilizes multiple cores, and takes advantage of hyper threading. The i5 750 does not benefit from hyper threading, so clock for clock it shouldn't perform quite as well as the i7 920, which has the same base clock speed, but supports hyper threading.

From the results the stock number is quite a bit lower than what the i7 920 would achieve. But with the simple OC genie overclock in place the results increase to a very respectable 144. 144 is as good, if not better than what the i7 920 is able to achieve at stock. The i5 750 can be had for almost $100 less than the i7 920. The more modest motherboards like the GD65 usually come in a bit cheaper than the x58 chipset boards that the i7 920 requires. The i7 920 is about equal, and in some occasions a bit behind the i5 750 in applications that don't support hyper threading. Now, the i7 920 also overclocks rather well to bring it's hyper threading results above the i5 750 results again, but you might want to ask yourself if slightly better results in hyper threading applications is worth $120+.

For the second test, another Valve developed benchmark software, known as the Map Compilation Benchmark was used. Basically the utility processes a map from Half Life 2 and measures how long it takes. Faster CPUs will take less time. Lower results here are better.

Similar to the results of the previous test, at stock the numbers for the i5 750 lag behind what the i7 920 would come up with. When overclocked things tighten up a bit, but from what I've seen the i7 920 is still usually able to bring up slightly better results.

The final synthetic benchmark used is the Cinebench 10.0 benchmark software. Yet again the benefits of hyper threading come into play. The way the program benchmarks is by testing single core performance, followed by the performance of multiple cores. Three runs are averaged to give the results below.

While there is some modest gains from the overclock in the single CPU test, in the second test taking advantage of all cores, the gains are huge. In the end, the multi core test showed almost 2500 additional points over the stock speeds. At 3.5times the points awarded it looks like the additional cores are utilized well. As a point of reference, this again gives the i5 750 performance as good if not better than the i7 920.

Now with the synthetic benchmarks out of the way, it is time to move over to what really counts. In game performance! To compare the results of the overclocked i5 I will be setting it against the results shown in the previous review for the 4890, using the Q6600 at 3.2GHz as the CPU. The video card will be running at stock settings for this one, so I will only be comparing the stock results. The games under the microscope include Crysis Warhead, and FEAR 2 : Project Origin.

Initially a lower resolution of 1680x1050 will help show more exaggerated differences in performance between the two systems. Keep in mind that the only difference between systems is CPU / memory / motherboard combo used. The video card is the main piece of hardware stressed in games these days, especially in a game like Crysis.

As expected, the max and average values stay much the same. However, the frames per second low shows a rather large difference. It seems that with all the various things going on during Crysis gameplay even a powerful system like the Q6600 system has trouble keeping up at the most stressful moments. At the higher resolution even more stress will go to the 4890, so the margin between the two systems will shrink even more.

Things are still pretty close between the average and maximum frames per second, and now the minimum numbers look evened out as well. The Q6600 system remains about the same, The i5 750 drops from 30 down below 25 frames per second.

The next game used was FEAR 2 : Project Origin. Because FEAR 2 doesn't stress the 4890 all that much I expect some noticeable differences between the two systems.

So much for expectations. In fact the i5 750 even falls short on the max frames per second. Let's see how things looks at 1920x1200.

Same scene here. Only the slightest edge on average and minimum frames per second.

<-Back | | Next->