Assembly And Overclocking
Assembling the system went smoothly and without much to talk about. Our sole irritation was that the Radeon 4850 X2 is a massive 11.25 inch behemoth––it is so huge, in fact, that it would not fit in the Cooler Master Centurion case (or not without some modifications).
To fit the beast in the Centurion, we had to remove the hard drive cage. Four screws later, the 4850 X2 was seated in the motherboard and good to go. Of course, we had to move the hard drive into one of the floppy drive bays but this was a minor nuisance at worst. It would have been more problematic with a two- or three-drive array, perhaps.
It turns out the 4850 X2 exhausts air out of the top of the card, which ended up fighting the Centurion case's side-panel cooler that was set at the factory as an air intake. We then reversed the fan so that it became an exhaust fan to help the 4850 X2 spit its heated air out of the case.
Other than this, we have no issues to report with the platform and hardware from an assembly point of view, which is a pleasant surprise considering that this is a relatively new chipset, motherboard, and CPU. Kudos to the hardware and driver teams.
Unfortunately, the Radeon 4850 X2 video card wasn’t quite as accommodating as the rest of the platform. First, it seemed Sapphire’s Radeon 4850 X2 was not recognized by any of the drivers on AMD’s Website. The card worked fine when we installed the driver off of the CD, but we were prevented from using the new Catalyst 8.12 driver suite. History has shown that AMD will support newer cards in time, but the 4850 isn’t exactly a spring chicken, and we would’ve liked to see the 8.12 drivers recognize the card.
Using slightly older drivers was a nuisance, but it wasn’t nearly as irritating as the inexplicable crashes we experienced with the card. At stock speeds when running graphics benchmarks, it would lock up, plain and simple. At first we suspected high temperatures but the 4850’s cooler was doing a fine job keeping the card under 60 degrees Celsius under load. Further experimentation showed us that the card would perform stably if the two GPUs were underclocked by 50 MHz, while our only thought was that we might have a bad apple.
Looking on the Web for similar experiences, it doesn’t look like this is a widespread issue with the 4850 X2. But while a 50 MHz underclock on the core probably won’t affect our benchmarks by more than a few percent points, it’s really going to hurt us when we overclock and try and bring the fight to last month’s 4870 X2 performance numbers.
In any case, we went forth on our overclocking mission hoping that the CPU would be more accommodating than the GPUs were.
This is my first time playing with the Core i7, so I have an opportunity to share some of my learning-curve experience with you folks.
Most of you probably know that the Core i7's frequency is determined by its multiplier multiplied by a base clock speed. The base clock works essentially like the front side bus (FSB) does when overclocking Core 2 CPUs and the i7 920 doesn't allow control of the multiplier. So overclocking an i7 is very similar to overclocking the Core 2 as you up the voltages a bit and raise the base clock (abbreviated BCLK).
Intel has implemented something called "overspeed protection" in its Core i7 models, which poses a small problem. The CPU will put a stop to things if it draws more than 130 W/100 amps. If you draw more when overclocking, the CPU will throttle back its multiplier and lower the clock speed. Ouch.
There are two ways to deal with this limitation. The first is to find a motherboard that has an option to disable overspeed protection. Intel's own boards have a CPU VR Current Limit Override option in the BIOS that turns overspeed off, and on the ASUS board we tested the option called "CPU TM Function." If your board supports a similar setting, you're in luck since you can then raise voltage as high as you like and overspeed protection should not kick in. The second way to deal with this limitation is to avoid it entirely. If we ensure that our CPU voltage is set low enough that it will never hit the 130 W limitation at 100 amps, we should be good to go. Now, 100 amps x 1.3 V = 130 W, so 1.3 V should be about the maximum setting you can use on the i7 920 without worrying about bumping into the overspeed-protection limitations. Since the stock voltage of the i7 920 is 1.2 V, this does give us some room to stretch our legs a bit.
We couldn't find an obvious overspeed-protection option in the Gigabyte BIOS, but in our case it didn't really matter since at 1.32 V, the CPU was overclocking to 3.7 GHz without the overspeed protection kicking in and lowering the multiplier. We did turn off Turbo Mode when overclocking because it constantly adjusted the multiplier and we preferred consistent performance.
In the end, heat limited us more than anything else. We settled on a final overclock of 3.7 GHz by setting our voltage to 1.32 V from the stock 1.2 V; we set the QPI/VTT voltage to 1.3 V from the stock 1.2 V; and we set the IOH Core voltage to 1.2 V from the stock 1.1 V.
One note about the IOH core voltage: when we tried to set it to 1.3 V, we experienced a hard crash when everything would lock up and the screen output would freeze. This is very odd, but when we backed up on the IOH Core voltage to 1.2 V, things were peachy.
We used two software utilities to record temperatures during our overclocking efforts. Speedfan reported a 42 degrees Celsius temp at idle and Realtemp indicated 52 degrees Celsius. The BIOS health monitor appeared to agree with Realtemp, which seemed very high for idle temps, even with the stock cooler.
The load temperatures reported by the Speedfan and Realtemp utilities were absolutely terrifying with the stock cooler. After a 1/2 hour Prime95 run, the temperatures plateaued as Speedfan reported 85 degrees Celsius, and Realtemp reported 95 degrees Celsius. (Chris: I can confirm--in my upcoming i7 920/Phenom II comparison, I'm also seeing the 920 peak at 85 degrees Celsius under load). Looking around on the Web, it seems that other folks are reporting high Core i7 temperatures as well. We experienced no stability problems with the machine at these temperatures so we decided to plod on, but we would certainly recommend a better cooler than the stock piece if you plan to put together an i7 machine for overclocking purposes.
For the Radeon 4850 HD X2, we did manage to overclock the memory from 993 MHz to 1,060 MHz even though the cores were unwilling to go anywhere. The 67 MHz increase wasn't great and probably won't make a difference in the benches, but it was better than nothing.
Now that we're armed with higher clocks, let's throw it down against last month's E8500 machine and its impressive 4+ GHz overclock.