To test out the ECS A780GM-A Ultra I will be pairing it up with the following hardware:
AMD 9850BE Phenom X4
Swiftech H2O-220 Compact
2x1Gb OCZ Titanium DDR2-800
3x74Gb WD Raptor HDDs
I low-level formatted the drives and configured them in a RAID0 stripe array, then proceeded with a fresh install of 32bit Vista Ultimate. The AMD RAID driver was specified during the install and loaded from the motherboard software CD. All other drivers for the chipset, audio, video and NIC were installed from the same CD following the completion of Vista.
Using HD Tune I was only able to achieve ~120MB/s average transfer speed, whereas this setup had scored ~180MB/s on previous nVidia chipsets. I made sure the motherboard BIOS was up to date and also tried downloading newer versions of AMD drivers and software however there was no noticeable difference. Possibly the SB750 controller has an issue with the older SATA 1.5GB/s drives or the fact that they do not support NCQ.
This processor had previously achieved a peak stable overclock of 3.0Ghz in an nVidia GeForce 8200 board, up from the stock 2.5Ghz, at a reasonable 1.375V, so I was anxious to see what it could do with the advantage of AMD's new ACC feature. Although I was able to duplicate my 3.0Ghz overclock with ease, unfortunately through extensive testing I found I could not advance the chip to 3.1Ghz on the ECS A780GM-A Ultra either. Up to 1.45V was applied and temperatures climbed no higher than mid-40's under load, however attempts at stress-testing at this speed inevitably resulted in the "Secondary Processor Clock" BSOD. Enabling the ACC feature had no effect, no matter if I left it on the Auto setting or tried +2%, +4% or +6%. In fact, stability began to be compromised the more I increased this value, and I did not try settings above +6%.
I used both the BIOS and AMD OverDrive in my overclocking attempts and found the AMD software to be easy to use and quite extensive in its available options. OverDrive does allow each core to be overclocked individually, and while it may have been possible to increase some of them above 3.0Ghz, the process of testing this would likely have been tedious and time consuming, not to mention the questionable value of having the different cores running at disparate frequencies. Most of the settings in OverDrive can be modified on the fly, however some require a reboot to take effect and others simply can't be changed at all. One setting I found to be lacking in both was the ability to modify the NB Multiplier, although the HT Multi could be adjusted in either.
The processor, memory and hard drive stripe array each garner the maximum 5.9 performance rating from Windows Vista. The onboard Radeon HD3200 chipset however only gets a 4.2 for Windows Aero desktop and a 4.1 for 3D gaming graphics performance. That's actually not bad for an integrated graphics solution. Although the chipset seems to cause confusion in software monitoring programs that read core temperature from the processor. OCCT is just one of many different applications I use in testing, and as you can see it is only displaying three out of four cores from the Phenom X4. The temperature reading and some of the voltages are also significantly off, the core temp is showing appx. 25° C higher than normal and both the +3.3v and +12v are displaying incorrectly. Strangely enough, when the onboard graphics are disabled and a discrete card is used, the temperature and voltage readings work fine.
During testing the system began exhibiting stability issues that appeared to be related to the onboard graphics. The symptoms consisted of graphical anomalies, screen blanks and system hard locks. I noted that the chipset heatsink was extremely hot to the touch, and as I am using liquid cooling on the processor there is no direct airflow over it.
I removed the chipset heatsink and discovered that ECS used a thermal pad for the interface material. This typically provides poor performance compared to most pastes, so I opted to remove the pad and clean the chip and heatsink thoroughly. I then applied some Arctic Silver paste and reseated the heatsink. I also installed a fan onto the heatsink, in this case the 60mm fan from a stock AMD processor cooler. It overhangs the sink slightly but provides plenty of airflow to keep the chip cool. A 50mm or even a 40mm would probably also suffice.
Now that I could put a load on the chipset without locking up the system, I selected Futuremark's 3DMark06 benchmark to obtain a baseline. The result was 1933 3DMarks, and while low in comparison to most current generation discrete video cards, that's nearly double the score obtained by our previous GeForce 8200 integrated graphics test, in fact it nearly rivals the nVidia 8500GT! Unfortunately the HD3200 chipset is incapable of handling the extreme graphics of a modern benchmark utility like Vantage. While the program would load and get appx. two thirds of the way through, it eventually crashed back to the desktop. Framerates were alternating between 0 to 1 FPS which is similar to the results of the GeForce 8200 on this test.
Let's wrap things up with some final thoughts.