The Athlon XP has injected new life and improved performance for fans of AMD CPUs. Unfortunately, users are faced with a dizzying array of choices for motherboards, and picking the right motherboard for your AMD CPU can seem like a daunting chore. In some ways, choosing the best motherboard for an Athlon XP is more critical than for users of Intel CPUs. For one thing, the rate of turnover of core logic chipsets supporting AMD CPUs is more frequent. This means that technology improvements (such as faster DDR memory) often show up on Athlon systems first, and it also means that the system BIOSes can be immature and the designs sometimes seem more rushed.
The recent release of DDR memory support for the Intel CPU line has given tech enthusiasts more options if they favor Intel CPUs. However, DIYers with limited funds may still want to steer clear of Northwood because of its higher cost for roughly the same performance level. Its great to have the newest toys on the block, but to get em you also have to have the biggest wallet on the block. While its true that RDRAM seems to be slightly cheaper these days than name brand DDR memory, its also true that good Athlon motherboards and a fast Athlon XP CPU still costs less.
So AMDs range of solutions continues to be a favorite among gamers and DIYers, both for its gratifying price/performance ratio and its accessibility when it comes to tweaking, overclocking and otherwise breaking the terms of their processors warranties.
A while back, we reviewed whole swath of AMD DDR chipsets and motherboards. (Check out “DDR Solutions for the Athlon XP”) Round two is upon us, but its not as diverse as the opening volley. Today were looking at a six pack of AMD DDR motherboards, five of which are built upon the sturdy foundation of AMDs wonderfully stable and efficient VIA KT266A chipset. The sixth is built around the ALi MAGiK1 SDR/DDR chipset.
One thing to remember before we take the plunge: in seeking a home for their processors, RAM, and other components, different folks are drawn toward different features. Most of the boards reviewed here are high-quality bits of equipment, but that doesnt mean theyre right for everybody. Tweakers and overclockers look for boards that have as few jumpers as possible and feature lots of BIOS setup controls for memory timings, FSB and clock ratio settings, and so on. Hardcore gamers want tweakable performance–the ability to squeeze every last drop of bandwidth and speed to crank up their frame rates as high as possible. Less driven folks might look for a good, stable platform that they can set up as a server, or use to crunch spreadsheets and compile graphics without having to worry about it crashing. Theres something for all of you here.
VIA KT266A and ALi
VIAs KT266A DDR chipset is quickly becoming ubiquitous. If you frequent PC enthusiast message boards, youll probably note that boards based upon it are discussed a good ten times more than boards based on any other chipset–including AMDs own DDR core tech–put together. It is, of course, VIAs follow-up to the somewhat disappointing KT266, and it more than makes up for the weaknesses of that core.
When we covered the KT266A in detail here, in the same article, we made note that a good deal of the KT266A boards available at the time may not have been ready for prime time. Some of the boards were less than stable, and performance from one KT266A board to the next varied. We theorized that it was only a matter of time before the KT266A hits its peak.
The time has come. The late bloomers rounded up here shine with professional polish and flaunt more dependability than the early comers in the previous roundup. Likewise, in terms of performance, the five KT266A boards reviewed in this piece are remarkably similar in many cases–but that doesnt mean theres not a great deal of other factors separating one board from the next.
ALi MAGiK1 Chipset
A single gladiator in this roundup dares to feature a different chipset. The Soyo K7ADA is based instead on a new spin of the ALi MAGiK1 core. Created in the depths of Acer Labs, the ALi MAGiK1, which we covered here, has been getting mixed reviews thus far. While the VIA boards all churned out similar numbers in our death-defying series of benchmarks, the K7ADAs scores were decidedly different, some for the better, but most for the worse.
Note, though, that it wouldnt be fair to judge a chipset by a single board. The Soyo board reviewed herein raises concerns for the memory bandwidth efficiency of the MAGiK1–but it could very well be a characteristic the board itself. It suffered in a few of the more intensive tests, while the KT266A boards showed, for the most part, little variance. Its definitely worth following the commentary of DIYers in our forum who have worked with this and other ALi MAGiK1-based motherboards to get a better feel for the chipsets capabilities across the board. nForce Chipset?
No boards based on nVidias nForce chipset are featured in this set of reviews. Were taking an especially close-up look at one nForce board, and will report on that later.
Our benchmarking is done using custom designed, NASA-inspired machinery thats calibrated to a hundredth of a micron. It takes place in a sealed room thats 1000 times cleaner than any hospital operating room. Were observed by independent auditors from three different agencies, who themselves are observed by other auditors, and all results are notarized and sealed on the spot. (No, Arthur Andersen is not involved).
Okay, so were exaggerating a bit. That may be going a little overboard, but suffice it to say our test conditions are very strictly controlled. We reformat the hard drive between each and every motherboard test, and we install all the same versions of the same applications and games in the same test bed for each board. Tests are always run in the same order, and we reboot the system between each test run.
Before testing a board, we make some mandatory settings changes in its BIOS to even the field. We do not cache any BIOS feature, including video BIOS. We set the memory settings for auto, but if the board has a default optimal setting we give it a try to give the board the best possible circumstances. If the optimal setting proves to be unstable, we make a quick attempt to tweak the memory settings, but if it takes too much work we drop back to default. While we comment on the accessibility a board offers to overclockers, we do not overclock anything during our benchmark runs. We go for an out of the box experience.
During benchmark runs, onboard audio, game and MIDI controllers are all disabled. We do use the onboard network adapter if there is one. Once weve completed our entire battery of tests, we go back and listen to the audio quality of the onboard audio, and then we play several games and run applications to get a flavor for the real-world experience the board offers.
We install each board according to its manufacturers instructions. We use the BIOS and drivers that were shipped with the board, only seeking updates to fix apparent instabilities, incompatibilities, and other problems.
Our testbed consists of the following components:
Antec SX-635 case
AMD Athlon XP 1800+ with Cooler Master active cooler
640MB DDR memory
WD400BB ATA100 hard drive, Maxtor D740X-6 ATA100 hard drive, Toshiba 16x DVD ROM drive and 1.44MB, 3.5″ floppy drives
GeForce3 Ti 500 (driver v.21.85)
KeyTronic 104-key PS/2
Logitech USB Optical mouse
ABIT KR7A RAID
A few years ago, half the world learned how to overclock with an ABIT BX6 motherboard (or its splendid follow-up, the BX6 2.0) and an Intel Celeron 300A processor. ABIT unleashed upon the tech world its SoftMenu, a component of the BIOS setup utility that allowed even staunch amateurs to tinker with their boards FSB and multiplier settings, and subsequently it became a champion of DIYers everywhere.
Cut to the present. In the KR7A and other recent ABIT boards, the SoftMenu is up to version III now–and its no less useful than it was on the BX6 series. The only jumper on the board is the clear CMOS block.
The board features a well-engineered and convenient layout. It has six PCI slots and four DIMM sockets, is hampered only by the CPU sockets close proximity to the edge of the board. That can make it tricky to remove and replace a CPU cooler without hauling the whole board out of the case. Two of the PCI slots allow for full-length cards–or five do if you dont take advantage of the HighPoint HPT372 RAID controller.
Setting up RAID on this board will be tough for neophytes, mainly due to the poor documentation. The RAID section of the manual is a six-page afterthought, and contains such insightful advice as “For detail RAID introduce and concept, you can found it on our WEB site…or you can search the concerning information on the internet. We do not description it on this manual. (sic)” In other words, techies only need apply.
Otherwise, the KR7A RAID is a solid board, especially for performance tweakers and overclockers. SoftMenu III lets you change voltages, FSB frequencies, and multipliers to your hearts content without ever having to open the case.
Best of all, the board cleans up in our Windows 2000 benchmarks. It scores top or near-top honors in every Win2K run except the SPECViewPerf MedMCAD test. It was equally competitive in the majority of our gaming tests, and paired with a powerful processor (like the Athlon XP 1800+) this board is a great chassis for a gaming speed demon.
The only things missing from this board are extras. At its price, wed hope for onboard audio or at least a LAN adapter. What you get is a no-frills RAID board with very low-frills documentation. But its not aimed at the value-conscious: this ABIT board is for hardcore tweakers. And, as is the usual case with an ABIT board, were delighted with what it has to offer.
Pros: Great performance; SoftMenu III rocks; well-laid out board
Cons: Pricey; lousy documentation
-7VTXH+”> The itty-bitty blue Gigabyte GA-7VTH+ has one thing that none of the other motherboards in this roundup have: a great, big Iomega-style poster to walk owners through the installation and setup process. We loved it because it has an almost life-sized, clearly labeled diagram of the board. Newbies will love it: it shows in glossy color pictures how to insert the motherboard, cooler and RAM, and it even goes so far as to show a sample case and what physical challenges one might encounter in installing a board. Likewise, the manual is thick and mostly well-written.
We already know all that stuff, though. There are some other perks that turned our heads, like the excellent layout. All five of the PCI slots can handle full-length boards. It has three DIMM sockets, but their left side clips are impeded while an AGP card is in place. Theres plenty of room for a big CPU heat sink. The northbridge is covered with a passive cooler, which in theory should free up a fan connector–but there are only two.
The onboard audio chip is a Creative Labs CT5580, and features pretty decent quality, hardware accelerated sound. It only supports two channels, though, so if youve been eyeing a 5.1 system youll have to spring for an add-in sound board.
Theres also a RTL8100L 10-base-T 10/100 LAN adapter which works perfectly well. We didnt actually benchmark it, but we used it to download some of the tests for our report and to transfer some of the results over our LAN to the PC upon which this is being written, and it functioned predictably and reliably.
A DIP switch toggles the FSB frequency, and another is provided to manually adjust the clock ratio, so overclockers will need to do some case diving to apply their craft. Inexplicably missing from the motherboard is a clear CMOS jumper! The label is there, and solder caps mark where its pins should be, but the jumper is not to be found. If you ever need to clear CMOS, youll have to manually short the two caps.
The BIOS setup program contains a page dedicated wholly to system performance. Unfortunately, the system is unstable if you set it to its default high performance mode, which sets the DRAM timing to 1T and latency to 2. It hanged and crashed regularly during benchmarking. We able to achieve stability by riding the timing setting back to 2T, and we were able to leave the latency setting at 2.
All that fun stuff aside, the GA-7VTXH+ offered solid performance, turning out impressive numbers in Content Creation Winstone and gaming tests–it completely owned the Quake 3 Arena Windows Me benchmark. The only area in which it faltered was in the SiSoft Sandra SSE memory benchmark, topping only the bandwidth-challenged Soyo K7ADA.
With its outstanding documentation and overall commendable performance, the GA-7VTXH+ is a perfect board for a first-time DIYer. The high quality onboard audio and LAN adapters are terrific value-added components goodies that make the board worth its already competitive price. Overclockers would be better served to check out a more tweakable board, but newbies will really score with this Gigabyte board.
Pros: Incredible documentation; very newbie-friendly; solid performance
Cons: No CMOS clear jumper; default high performance mode isnt stable
The lone ALi MAGiK1 board in our roundup stands out from the group for more than just its chipset. It proved to be the least reliable board of the bunch, and it appears to suffer from bandwidth problems, at least according to some of our tests.
The board itself is tiny and very cluttered. It features five PCI slots, three of which could house full length boards. Although the chipset supports both DDR and SDRAM, this board only provides three DDR DIMM slots–and theyre practically on top of the CPU socket. The onboard audio consists only of vanilla AC97 CODEC for software controlled sound–and its very low quality.
The documentation is as thin as the board. Youll need it, though, because for one thing, the case wiring pinouts arent labeled on the PCB. Neither are the three DIP switches, which you must use if you wish to manually adjust CPU multiplier and core voltage. A jumper toggles the FSB frequency. One of the DIP switches, labeled FJ1, doesnt appear to be in the manual at all.
For its part, the BIOS is mainly generic but I couldnt find an option to disable the onboard audio CODEC. Maybe thats what that mystery DIP switch is for…
In action, the K7ADA was hot and cold (and were not talking about core temperature). It hung up when running the 3DMark2001 and 3D Winbench 2001 low-level benchmarks, but it did a terrific job with the game benchmarks. And real-world gaming went smoothly. We ran a few current graphics-intensive titles like Tom Clancys Ghost Recon, Max Payne, and Red Faction, and the board didnt falter.
The K7ADA did startlingly well in some of the processor tests, like our 3D Studio Max and Lightwave rendering benchmarks. However, it coughed up blood over some of the bandwidth tests–particularly SiSoft Sandras SSE memory benchmark. We looked for BIOS and driver updates that might have helped its performance, but the software included with the board was more recent than anything online.
With its erratic scores, its pathetic documentation, and its reliance on jumpers and DIP switches for serious tweaking, wed be hard pressed to recommend the Soyo K7ADA to anyone. Its price is nice, and gamers on a serious budget might find solace with that–but although our real-world gameplay didnt result in crashes, its inability to cope with 3DMark2001 and GameGauge would cause us to shy away from even that perk.
Pros: Inexpensive; commendable game performance
Cons: Inconsistent benchmarks; terrible documentation; tweaking requires DIP switch manipulation
Like the ABIT KR7A RAID, this is a truly jumperless motherboard. Other than a clear CMOS jumper (which is necessary) and a CMOS write protect jumper (which is optional), you dont have to touch the board to tweak the daylights out of it. Everything, including the CPU voltage and multiplier, the memory timing, and even the FSB frequency, can be adjusted via the BIOS setup utility. That alone makes this board tweakers dream.
With six PCI slots, four DIMM sockets, a HighPoint HPT372 RAID controller, and an onboard C-Media 8738 5.1 audio chip, Shuttle did an incredibly efficient job of laying out this board. It could have been an overburdened crowded mess, but everything is easily reachable and theres room for whatever you want to stick in there, including a couple of full-length PCI boards. The only caveat we have with the layout is that our oversized heat sink actually made contact with the side of the northbridge cooler, but neither fan was impeded and the CPU stayed cool throughout our tests.
A major concern with the AK35GTR is its stability; motherboard forums are filled with threads of woes with crashes and other mishaps on earlier boards, especially when paired with Athlon XP processors.
Our battery of tests on the AK35GTR went with refreshing and relieving smoothness. In fact, the Shuttle the only board of the bunch that ran perfectly with the DRAM command rate set to 1T, with nary a crash or a lockup. It performed very well on all of our benchmarks, taking the top score in Business Winstone 2001 and remaining competitive through the rest of the tests.
RAID administration was trickier than on other boards, but with the help a bit of experimentation and the separate RAID manual we had it running smoothly after only a few trial runs.
The onboard sound chip deserves special mention. Its capable of very convincing 3D positional audio and, since the line-in jack doubles as a rear speaker out jack, it supports four-speaker gaming rigs like the Logitech Z560 and the Klipsch ProMedia 4.1. Testing only by ear, its definitely on par with the SoundBlaster Live! used in our test bed.
Shuttle knocked one out of the park with the AK35GTR. Between its excellent performance, its honest jumperless convenience, and Rock of Gibraltar stability, it should please even the most finicky of DIYers.
Price: $ 115
Pros: Truly jumperless; awesome performance; stable at optimal settings; good onboard audio
Cons: Large CPU coolers encroach upon the northbridge cooler
ECS K7VTA3 V
.2″> Not to be confused with the K7VTA3 v.1.x (which is based on the VIA KT266 chipset), the 2.x version of this little no-frills board is as much of an outcast from this bunch as the ALi-powered Soyo K7ADA. While its overall scores were in line with the other KT266A, it was thwarted by other problems.
Its layout is crowded and inconvenient, with the CPU socket encroaching upon the three DIMM sockets. Theres room for oversized rectangular heat sinks, but users of oblong Orb-style coolers have a scant few millimeters of clearance. The board features a generic, software-controlled AC97 audio CODEC, five PCI slots and, of all things, a CNR slot.
Installing the K7VTA3 was simple enough, until it came to connecting the case wiring. The connector layout doesnt match the diagrams presented in the manual: while we found the proper pinouts, theyre not labeled on the motherboard and theyre surrounded by connector blocks that arent even in the book. Perhaps ECS didnt update the documentation when it updated the board.
Meanwhile the little board completely bucks the jumper-free industry trend. Youll need to set jumpers for the FSB frequency, the BIOS voltage, and even to toggle power to the extra USB ports. Meanwhile, there are only two three-pin power sockets for cooling fans.
In other words, this is not an overclockers board.
On the happy side, once we got the little pink motherboard (yes, its pink) configured and installed, it proved to be flawlessly stable. Its gaming and productivity performance was sub-par compared to the rest of the crop, but it held its own in bandwidth testing, placing third in the SiSoft Sandra SSE Integer Streaming memory benchmark.
Furthermore, its price is right. Its easy to score a K7VTA3 online for $75 plus shipping. DIYers certainly wouldnt want one for themselves, but its a decent, cost effective board for building up a web surfing board for Mom, or for putting to use as a server on a SOHO LAN.
ECS K7VTA3 v.2
Pros: Very stable; nice price; its pink
Cons: Too many jumpers; documentation doesnt match board; low end performance; its pink
-SR”> Another board fit for the masses but a bane to DIYers, the DFI AD70-SR is packed with features but needs too much physical handling to score points with tweakers.
Jumpers allow you to toggle 5V power for the integrated USB ports and the pinouts for USB2 and 3. You also set the FSB voltage with a jumper, and changing the CPU multiplier requires manipulating a DIP switch–although you can adjust the voltage through the BIOS utility.
Considering all the real estate on this big board, DFI did a stunningly inefficient job laying it out; they completely failed to take advantage of its space. Theres enough empty room around the north bridge to build a house. There are only five PCI slots, but there should be room for six; it sports three DIMM sockets, but theres sufficient acreage for a fourth. The CPU socket is on the very edge of the board, which puts it within an amoebas breath of the power supply. The EIDE and floppy connectors are clustered tightly together, but the Promise 20265R-powered RAID connectors have some breathing room.
DFI scores points for documentation. In addition to the motherboard manual and a separate manual for RAID administration, the board includes an old fashioned, glue-backed, notated diagram of the board that you can stick to the inside of your case for reference.
The DFI is an all-around solid performer, kicking particular butt in Windows Me game testing. It took the checkered flag in the 3DMark tests and the GameGauge composite score–but suffered in our Quake 3 Arena high-polygon test (using the VIA demo).
With its impressive stability and friendly documentation, this board might be the foundation for a decent server or a power hungry, but tweak-shy gamer. Unfortunately, it doesnt have much in the way of features for its price, and its layout is atrocious. DIYers who have a penchant for experimentation would definitely be better off looking elsewhere.
Pros: Very stable; great game performance
Cons: Poor layout; DIP switch reliance for CPU ratio tweaking
Score: 7/10 www.dfi.com
Business Winstone and Content
These benchmarks use real-world applications to determine how well a given system can handle business applications and content creation functions. They each put the thumbscrews to the CPU and memory subsystems, and give a good idea of how a system stacks up in productivity situations.
Performance variations with general benchmarks like these tend to be slim, since were using the same processor, memory and other components. In fact, the variance among our contenders in Business Winstone 2001 is a mere 3%. However, in Content Creation Winstone 2001 we see the ALi MAGiC1-based Soyo board coming in significantly lower than its VIA KT266A contenders. This, dear readers, is the beginning of a trend.
SiSoft Sandra Memory Bandwidth
As we did in the last AMD DDR motherboard roundup, we used SiSoft Sandras SSE integer streaming test to measure the memory bandwidth of this batch. The trend continues: the results are telling of a serious deficiency on the part of the Soyo board. It came in more than 18% lower than its nearest competitor, the Gigabyte K7VTXH+, and a whopping 25% lower than the ABIT KR7A, which, by the way, burned through this benchmark without breaking a sweat.
Whether that harkens to a memory bandwidth deficiency with the ALi MAGiK1, or just with the Soyo K7ADA, cannot be determined without a wider sampling of ALi product. Its also possible that BIOS updates could improve the memory performance of the relatively new K7ADA.
OpenGL Performance with SPEC
Our OpenGL test showed almost no variance whatsoever between the various motherboards. Chalk that up to all the systems using the same graphics subsystem, and given the test is quite graphics-intensive, performance of all systems is near identical.
Encoding puts a lot of strain on the system processor and the memory subsystem. Notice in the MPEG test that the troubled Soyo K7ADA pulled ahead for this benchmark, but not by a convincing margin. This could mean that the ALi MAGiK1 chipset manages to eek a bit more efficiency out of the CPU in Windows 2000.
Under Windows Me, the Soyo came out last in both the Windows Media Encoder and Quicktime Encoder tests, the former by about six percent, and the latter by about 7 percent. Though the margins are narrow, comparing the Soyo boards encoding performance between the two operating systems would appear to indicate that the K7ADA makes more efficient use of the system processor through the Windows 2000s NT kernel than it does through Windows Mes 9x kernel.
Meanwhile, both the DFI and the Gigabyte boards showed excellent encoding performance in Windows Me. Among the KT266A boards, the ABIT KR7A RAID and the ECS K7VTA3 were collectively the least efficient motherboards.
.0b and 3D Studio Max 4.2 Rendering”>
3D rendering engines puts heavy emphasis on a CPUs FPU performance. Its surprising that the 3D Studio Max test saw such heavy variance among processors on the same chipset. It really battered the Soyo–which until this test seemed to be fairly adept at processor intensive applications under Windows 2000. The 3D Studio test uses the architectural model from the old SPEC 3D Studio 3.0 benchmark. There are over 600,000 polygons in this particular scent–enough to really hammer both the CPU and the memory subsystem. The Lightwave raytrace test is also quite floating point intensive, but has relatively few vertices. So most of the time is spent in the rendering codepath, not in moving data to and from memory.
SolidWorks 2001 CPU Test
Solidworks 2001 is a mid-range, professional 3D CAD package. Its very CPU and system I/O intensive. Further emphasizing the paradoxical performance of the Soyo is the SolidWorks 2001 CPU test. While the KT266A boards hardly showed any variance at all, the Soyo once again came up short.
: 3D Winbench 2000″>
Not only did the Soyo come up short in all of these tests, it couldnt even complete the 3D WinBench test in Windows Me without crashing. Its definitely more stable in Windows 2000. Among the other motherboards, results varied little.
To get a real feel for the processors FPU power and the motherboards subsystem efficiency, we run the wholly synthetic 3DMark2001 in low res mode and turn off hardware support for transform and lighting, forcing the processor to do the work. At last, we saw some real variance among the individual motherboards.
In Windows 2000, the ECS K7VTA3 limped to the finish, lagging about 8% behind the benchmark-leading Shuttle AK35GRT. Notice that the Soyos score is missing from the chart; it couldnt complete the tests in either operating system without crashing.
Meanwhile, in Windows Me, the results were closer. The DFI AD70-SR reared its head and took the lead by a nose–or should we say a resistor?
Our other composite test, GameGauge, is actually a conglomerate of seven real-world games: ReVolt, Expendable, Unreal Tournament, Janes USAF, Quake 3 Arena, NFL Fever, and Panzer Elite. That pretty much covers the range of graphics-heavy titles youll run into in the real world.
In the composite score, which is calculated by determining the geometric mean of the frame rates returned in each game test, we see mild variance. The ever-paradoxical Soyo K7ADA board keeps up with the pack, while the ECS K7VTA3 lags behind, but not by an alarming margin.
More interesting are a couple of specific tests, which returned surprising results. The Soyo K7ADA kicked complete and total butt in ReVolt tests (we run three, and the results of the test shown here is typical of the results of all of them). Apparently, that motherboard has a particular knack for the high speed rendering challenges associated with racing titles (yes, we know that ReVolt is an RC racer, but its game engine is inherently similar to other action-oriented racing games).
Meanwhile, we also discovered that both the Soyo K7ADA and the Gigabyte K7VTXH+ hate football. We were so shocked by the Gigabytes seemingly anomalous score that, when wed finished all our testing, we reinstalled the board and reformatted the drive to run the test again. The results turned out to be repeatable. Then, we loaded up Madden 2002 and cranked the detail settings up. It seemed to play just fine. Maybe the Gigabyte board has something against one of the teams in our benchmark demo…
: Quake 3 Arena”>
No benchmarking suite would be complete without Quake 3 Arena. For motherboard testing, we run the benchmark at its fastest setting, which gives a better test of a motherboards bandwidth efficiency since it doesnt require much in the way of video card muscle.
In Windows 2000, the ABIT KR7A RAID came out ahead followed very closely by the ECS K7VTA3, while the Gigabyte K7VTXH+ lagged behind the others. It made up for the difference by taking the flag in the first Windows Me test, and competing quite well in the Windows Me high polygon benchmark.
Some gearheads will stand by ABIT to their graves. Its SoftMenu alone made lifelong fans out of first-time overclockers when it was introduced, and the company has a sturdy reputation for building fast, stable and vastly tweakable motherboards. The KR7A RAID carries on that tradition, and we can comfortably recommend it to hardened DIYers.
But as strong as the pull of brand loyalty may be to a techie, this techies opinion was swayed away from ABIT by the speedy, stable, feature-packed and surprisingly affordable Shuttle AK35GTR. This board has power to spare, and configuring it is as easy as wading through the BIOS setup menu; it may not be as neatly organized as ABITs SoftMenu III, but it offers the same level functionality. Its a killer board, and its likely to be in my next desktop system.
ExtremeTech readers tend to be a fairly hands-on bunch, but if youre new to the DIY experience or know somebody who is, steer toward the Gigabyte GA-7VTXH+. It wasnt the benchmark winner and its not as easily configurable as the ABIT or Shuttle boards, but its outstanding documentation gives it an excellent set of newbie training wheels, and its included audio and NIC can save a bargain hunter a few bucks.
We were underwhelmed by generic nature and the mediocre design of the DFI AD70-SR. Take it if someone gives one to you for free, but dont go out of your way trying to find one. Theres very little to set it apart from the vast hordes of motherboards on the market, and tweakers will be miffed by its dependence on physical switches and jumpers.
The bottom-feeders of the bunch are the ECS K7VTA3 v.2 and the maddeningly inconsistent Soyo K7ADA. The ECS doesnt seem ready for sale yet–its layout is better suited for a prototype. And the Soyo seemed stable under normal use, but the benchmark inconsistencies and failures raise a big red flag. We only had the board running for a day or two; theres no telling what flaws weeks or months of normal usage might uncover. And whoevers responsible for its pathetic documentation should be flogged with a dozen dictionaries.
The medal goes to Shuttle, and may there be many other boards like the awe-inspiring AK35GTR.