Post Cards from the Edge Update - AMD 780G

We have had a flurry of activity in the labs since posting our diatribe a few days ago. We have had almost round the clock conversations with the manufacturers on the items we discussed along with additional testing of newly arrived replacement products. There has also been an avalanche of comments, emails, and forum messages asking for additional information, commentary, or clarification of our statements and/or opinions. Here are the current 780G updates and we will discuss any additional information we gather over the next couple of days in the upcoming articles. We will provide an update to the 790i and Gigabyte 680i in our next blog shortly.

AMD 780G Update

We would like to reiterate that our problems with the AMD 780G motherboards are not chipset related. The problems revolve around the capability of current designs to properly handle the 125W TDP processors at stock settings or when overclocked. As we stated in our first article, the vast majority of users will never install a Phenom 9850BE or Athlon 6400+ X2 in these boards.

The motherboard manufacturers designed these particular products for the SOHO, casual gaming, and HTPC audiences with a price target in the $70~$90 range. This market sector typically will utilize a processor such as the LE1600, 3800+ X2, 4850e, or maybe even a 5000+ BE for their needs. As such, the power delivery design and electrical layout on the boards target the 45W~95W TDP processors. We have yet to have a failure utilizing any processor that meets these criteria after a few thousand hours of collective testing on our 780G motherboards.

However, we typically step outside the bounds of a given market sector in order to fully test the capabilities of the board and in hopes of finding one of those rare hidden jewels in the budget sector that offers performance equal to the high-end products. That is what leads us down the path of trying AMD’s finest series of desktop processors on their entry-level chipset. Of course, we found that path to be full of obstacles after suffering through several failures. We discussed this with the manufacturers at length and with AMD.

While a board’s PWM/MOSFET count generally indicates its capability to handle a particular load rating, it does not always indicate its ability to properly regulate, correct, and deliver clean power. That is why a board with a properly designed four-phase system can offer improved quality/performance over a poorly designed five- or six-phase system. All of our boards in the roundup utilize a decent three- or four-phase PWM design, with the exception of the ASRock board that features a five-phase design. Does this mean that the ASRock board will handle the 9850BE without a problem? Not necessarily, as there is more to our story than PWM design.

The critical aspect of running a 9850BE at stock speeds on the current 780G boards centers on the cooling of the PWM circuitry. Each and every manufacturer along with AMD agreed that cooling the MOSFETS properly was critical to the successful operation of the board at stock or overclocked speeds with the 9850BE - and to some degree, the 6400+ X2. The lack of space around the CPU area in the uATX design results in a crowding of components leading to additional thermal loads. This type of thermal loading is not prevalent in most ATX designs utilizing similar PWM components or the space that an additional/improved phase design (with associated components) affords.

We have measured MOSFET temperatures ranging from 128C~146C utilizing the stock CPU cooler on the 9850BE in our Silverstone SG03 case (temps were about the same in our Thermaltake and Zalman cases). While cooling of the PWM/MOSFET area is critical, most of the manufacturers admitted that improved component choices and electrical layouts similar to the 790FX boards is also required to ensure a robust design for users who are on the more enthusiastic side. However, this means additional costs that would place the board out of its intended market, a market where probably 98% of users would not worry about running a 125W TDP processor. For those who demand this extra level of performance, we expect to see several 780G ATX designs in May offering improved PWM designs.

Our primary concern about using the 9850BE or any other processors with this excellent chipset centers on the CPU support lists and general product marketing information available at the various manufacturer websites. We are happy to report that several of the manufacturers have been busy updating their 780G product information the past few days - something we appreciate, and we hope you will also.

First up is Gigabyte; they have added the newly released Phenoms to their CPU Support list and another surprise listing. The 9850BE has now appeared, although it's listed as unofficially supported. We asked Gigabyte about this and they responded that AMD has not qualified the 9850BE for the 780G uATX platforms. However, after running significant stress tests, Gigabyte is confident their GA-MA78GM-S2H board will operate correctly at stock speeds with the 9850BE. They have provided us with significant test data of the board running in a hotbox (no additional cooling) at 50C ambient temperatures for the past several days. This includes both stock and mildly overclocked settings.

We have received our replacement GA-MA78GM-S2H and have experienced varying results with it the past couple of days. Our particular setup has been prone to automatic shutdowns every 17 to 44 minutes when running OCCT, Crysis, or PCMark Vantage without additional cooling around the CPU area, primarily the MOSFETS. However, unlike our first board, the system recovers after a short cool down period. At first glance, an automatic shutdown would lead us to believe that the CPU is overheating. This is not the case, as the CPU never exceeded 48C during testing. We strongly believe the culprit is the lack of airflow across the MOSFETS as our next test session will indicate.

After placing an additional case fan over the PWM/MOSFET area, our board completed three hours of OCCT, two hours of our Crysis demo looping, and over five hours of PCMark Vantage looping. We only had one shutdown of OCCT and that was at the one hour, six minute mark, but it has not occurred again. We are still working with Gigabyte on our shutdown scenario without additional cooling along with a few BIOS related problems we have experienced when trying to overclock the 9850BE. At this time, we will defer to Gigabyte’s “official unofficial” support of the 9850BE as being their stamp of approval for running this CPU on the GA-MA78GM-S2H.

Next up is ASRock; they added the Phenom 9850BE to their CPU Support list this week for the A780FullDisplayPort. ASRock has stated to us that they are officially supporting the 9850BE at stock and overclocked settings. They based this qualification on their board passing AMD’s internal CPU test tool, SST version 4.7.4, without any problems or errors. However, our test results have been similar to the Gigabyte board. We have not noticed any stability or shutdown problems to date with additional cooling over the PWM/MOSFET area, even with the processor overclocked slightly to 2.7GHz. Like the Gigabyte board, the A780FullDisplayPort has recovered from the shutdowns after a cooling off period.

Random shutdowns have occurred without extra cooling utilizing the same applications tested on the Gigabyte board. ASRock just sent a revised BIOS that has additional tuning and support for the 9850BE that we will test later today. ASRock believes strongly that the retail heatsink is the problem. It appears not enough airflow is generated in a radial pattern to properly cool the MOSFETS. We will perform testing today with a new Thermaltake low profile fan/heatsink to test this theory.

We just started testing (who would have thought having three 9850BE processors on-hand would not be enough?) on the Jetway PA78GT3-DG with the 9850BE today. Jetway is confident their board will support and operate properly with the 9850BE after internal testing this week. Jetway has not updated their CPU Support List yet, but we expect them to shortly.

ECS, ASUS, Biostar, and J&W will not be supporting the 9850BE on the boards we discussed in our previous article. J&W did state that with proper cooling around the PWM/MOSFET area that their board would run the 9850BE, but they do not recommend it. We wholeheartedly agree. ASUS did provide their ATX based M3A-H/HDMI board with a 4+1 Phase design. We just received an updated BIOS for the B3 stepping processors and will commence testing later today. All of the manufacturers plan to have their CPU Support Lists updated shortly. Currently we have three 780G uATX boards that will support the 9850BE with full backing from the respective manufacturers. All of the manufacturers committed to providing updated CPU Support Information and additional warnings or advisories to the user if required. We will have a progress report this weekend on our test results and any additional information from Gigabyte, ASRock, and Jetway.

Other Items of Interest

Our sources have told us that the Intel G45 is suffering growing pains. The current revision is locked at a 667MHz GPU core speed and the VC1/H.264 decoder capabilities are turned off. A new revision will be available in July with the 800MHz GPU core speed and VC1/H.264 decoder capabilities turned on. In essence, the first chipsets (if they are available) will be nothing more than a G35+. We will update this information as we receive additional information.

We have been banging our heads against the wall for the last few weeks testing various sound cards for a roundup. Our ASUS Xonar DX and D2X PCI Express cards would not work properly on ASUS’ own 680i/780i/790i motherboards, but worked superbly on the Intel X48/P35 and AMD 780G/790FX boards. It turns out the Oxygen chipset that ASUS utilizes (AV100/200) is not designed for PCI Express operation. ASUS gets around the problem by utilizing a PLX bridge chipset to provide PCI Express functionality. According to ASUS, NVIDIA has confirmed the problem and will be providing a core BIOS update shortly to fix it.