DTK PEM 2500
This is typically what you'd see in an early 386. Notice at the top left of the board, no fewer than three crystal oscillators. Just right of the 386 CPU is a full 64kB of SRAM L2 cache - this would have been an amazingly fast 386 board. Back then, boards weren't obsolete in ten minutes and most chips were socketed to facilitate replacement, since motherboards were so expensive that labour costs paled into insignificance. This perhaps explains why this board was supported up until at least 1990 (the BIOS date, but that's not the original shipping BIOS) and long after 256kB cache boards using more integrated chipsets had appeared.
Note that there's no chipset, just bundles and bundles of logic. The only thing approaching a chipset is the VLSI chip which is probably an interrupt controller just left of the CPU.
The image is from an older site I used to have, so it's annotated with...
Points of Interest
1. This is a 387 FPU socket.
2. 10 SRAMs to make 64kB of cache. These are 64kb chips with two extra for parity.
3. AT keyboard connector...just look at the placement!
4. Slot for a (I believe) ZIP static column, up to 4MB.
5. SIPP RAM holes.
Yellow box - The size of a Baby-AT (the 'full size' things from the early Pentium and 486 era!) motherboard in comparison.
In 1993 there was a fire at a factory supplying DRAM and SRAMs. As a result, cache RAM became socketed and boards could be ordered without it fitted, because it was very expensive. It wasn't even for upgradablility, the sockets could only take one type of SRAM! This is a board from that era, but one sold to the very bottom of the barrel.
The board itself is very good and highly regarded, but this one shipped with no cache. Whoever bought a system with this board, the 486SX25 and 4 to 8MB of 30pin, 16bit SIMMs was rather stupid. It'd have been a good deal more than a proper 386DX40 and just as much slower.
Removing the CPU's L2 cache was an enormous downgrade, the CPU would then be perhaps 40% of the performance it would be with cache.
This one could be upgraded to a DX50 or DX75 (it was intended for a DX50) but was fixed to a 25MHz system clock.
Points of Interest
1. One of the worst CPUs ever, the 486 SX25.
2. VESA slots. This board layout was nearly canonical at that time to accomodate VESA.
3. Eight 30-pin SIMM slots for a maximum of 8MB RAM.
4. UMC 481 chipset. Highly regarded.
5. Nine empty SRAM sockets! Eight plus parity, about $30 USD (this would have been $60 or more at the height of the crisis) of cache chips is quite a saving when you're making a really crap PC. Boards could be ordered without cache, but it was very silly to do so.
By 1995, Intel's Pentium still hadn't caught on despite being introduced in 1993. Intel had a lot of problems producing it and just couldn't get the price down. Indeed, 1993 was the year of the 386DX40. This is what you'd see in most 486s sold by shops intent on keeping a good name for themselves if they weren't a UMC-only house, which were quite common. (The rest would be doing PC Chips...)
The OPTi 895 chipset was a nightmare to work with, coming in many different version each of which had their own style of jumper setting. Note the long DIP chip just left of the main bank of ISA slots. That's the AMIKEY keyboard controller - nothing more than a licensed 80C42 microcontroller, as seen in boards ever since the 286. You can see one in both previous boards. (Very top left on the DFE, same position as here on the MG.) When PCI took over and brought the LPC bus to handle standard I/O and ISA, the seperate keyboard controller would be annexed into the southbridge.
We were now down to the Baby-AT form factor which would survive as standard until ATX took over. The OPTi chipset was one of the early "Northbridge" and "Southbridge" combinations. The northbridge, the larger chip, handled RAM and fast I/O, like the three VESA Local Bus slots. The smaller rectangular chip was the southbridge (this would become just the "Super I/O controller" a few years later) and handled I/O, so all the rear ports and the keyboard.
Th8is board is also notable in that it has eight cache SRAMs by the bottom right. This means the cache was not ECC or parity protected.
Points of Interest
1. AMD 486DX100. If you weren't a big OEM, you'd be using far more Cyrix and AMD chips than Intel. They worked just as well (or better) and were a lot cheaper.
2. OPTi 895 (revision 3) chipset. OPTi would later lose a lot of influence and eventually leave the business with a hell of a broken (and highly anticipated) Pentium chipset.
3. 30 pin SIMMs were still popular in banks of two. Two 16bit SIMMs would satisfy the 32bit bus of a 486.
4. 72 pin SIMMs had been released by this time - a single SIMM could match a 486 if you used 72 pin, but many liked to re-use their older 30 pin and add wait states to match the faster bus. That'd have been a performance killer in a system as fast as this, which was really wanting at least 70ns FPM RAM and preferably 60ns EDO.
Submitter Lee Caskey writes:
"This was salvaged during my college years and used as a testing and programming platform while I was system administrator of my college. It was also fun to play old-school games on. It actually ran Win95B rather well when loaded with 64Mb of EDO RAM, the onboard 256Mb cache chips didn’t hurt either. It ran the gamut of 486-era processors from a TurboChip 133, to a AMD DX4-100 and even a Intel DX2 66 before I finally settled on the Cyrix chip. It’s lack of a hardware IDE drive interface required a VESA bus controller card. Also it required an external battery to maintain it’s BIOS settings, fortunately these seem to last forever."
This board is right at the time when building PCs stopped being a duopoly between large companies and bearded techs in a darkened room and the parts started becoming available in retail channels. The time was also a transitional one in many other ways. The bandwidth limits of EISA had long been feared and were being reached, but there was no consistent solution. The brown ports you see here are VESA Local Bus ports where peripherals were actually added to the CPU's local bus, the FSB. Rising FSB speeds and uncertain FSB clocks made VLB a chancy proposition and there was little support beyond certain high end video cards. When Pentium arrived with 50, 60 and 66MHz FSB clocks, VLB was abandoned entirely, its successor was PCI.
Visible on the top left is another legacy of older motherboards, onboard cache. Level 1 cache is sometimes called "internal cache" and level 2 "external" because L2 cache indeed was external. This particular cache consists of UM61256AK-15 CMOS SRAM in a 28 pin DIP3 package with organisation 32K x8, giving each chip a capacity of 256 kilobits. With eight of them, the bus becomes 64 bits wide and 256KB large, but the bus is actually only 32 bits wide and the SRAMs are organised in two banks of 32 bits. A single lone chip at the top right of the cluster is the parity buffer, as evidenced by the two 74xxx series 8-bit CMOS buffers right next to it.
At the bottom left is another sad legacy of PC history. Back before time, before the Big Bang, before even IBM, keyboards were just serial devices and they had their own microcontroller onboard to give them their 'smarts' and send the correct byte code when a key was pressed. This keyboard controller was moved internally very early in the PC's history and became the "Keyboard BIOS", though it was not truthfully a BIOS. Near that is the real BIOS, an Award system which would still be familiar today, BIOS UI having all but ended its evolution even this far back. AMI had introduced an utterly horrendous mouse-driven BIOS interface not long before and as a result, board makers stopped using the once-dominant AMI BIOS. A GUI BIOS is not necessarily a bad thing but AMI's attempt certainly was and this cast a long shadow over any future attempts of GUI BIOS
The system chipset is under the Soyo Energy Star sticker, it was very common back then to put a sticker on the main chipset (the bus controller, the Northbridge) and the Southbridge, the I/O controller, was just starting to emerge as one chip. It can be seen just below the main bus controller with the pink sticker on it. It seems Soyo didn't want people to know the chipset in use! The chipset in use was an SiS. Back then, of course, the actual chipset didn't much matter. System architecture was quite simple, you had a host chip which the FSB originated from, going to the CPU. This also handled L2 cache and tag RAM. From that you had a memory controller, the "southbridge" on this motherboard which sat on the CPU's address bus. The data bus went through the "northbridge" to the CPU's FSB.
The board layout, number of ISA slots, positioning of the chipset, CPU socket, cache, RAM, VESA slots was absolutely typical for an early 1990s 486 motherbard; Almost all of them had some variation on this basic layout. One of the first was the PX486P3 and, any manufacturer who didn't lay out his board like that had to reduce something. Maybe fewer RAM slots (like on this, only four, the standard was eight), maybe less cache, maybe fewer ISA slots...but regardless, he was almost certain to lose money.
The layout of this particular board looks more like something out of the 386 age than mid-486, components absolutely everywhere. When you recall the standard VLB/486 board fashioned after the PX486P3 you realise what's going on with the Soyo board. It's tiny! Where are the hard disk connectors? The floppy header? Headers for serial and parallel ports? The SY-25P had none of that, instead relying on a port controller card which sat in an ISA port but also used the VESA ports. This card would provide the serial and parallel on the back or give headers for them. It'd have the IDE channels onboard and finally the floppy disk controller port. But why? Why would any manufacturer do that? Namely, the SiS chipset used likely did not support LBA addressing, so hard disks above 520MB would cause problems. An add in card was free to support whatever it liked. The alternate explanation is that the SY-25P was produced to order by Soyo for an OEM customer, such as Compaq or Gateway, who needed a specific configuration.
So there you have it, a mid-486 era motherboard designed much more like the old early 386 motherboards which was certainly not a bad thing but not for the faint of heart.
You're looking at one of the best Super-7 motherboard ever made, just the component placement should tell you that. While boards from the likes of FIC using Via's powerful MVP3 chipset were sometimes faster (especially with 1MB or 2MB of cache), they were never quite stable due to issues with the SDRAM controller which Via would finally correct in the KX-133 Athlon chipset.
ALi's AladdinV chipset was capable of doing what it claimed it could, doing it reliably and quickly, but it had one major flaw; AGP. While this board does have an AGP slot, few were willing to take the chance. Way back when I had the baby brother to this (the GA5-AA), an ATI Rage128 worked well with some issues, a Voodoo3 was perfectly happy, but a Voodoo Banshee didn't work at all.
By this time, motherboard design had more or less standardised. Everything hung off the PCI bus, the Southbridge was a PCI device connected to the Northbridge by a PCI interface. It then did a PCI to PCI bridge to the system's main PCI slots and other onboard devices. The two ISA slots (one or two was still common) were again on a PCI bridge. So we had the core system based around the Northbridge, with the CPU, L2 cache and RAM hanging off the Northbridge, then a PCI bus going to the Southbridge, where everything else was connected. The original Southbridge use, to control external I/O, had been passed over to a WinBond or ITE "Super-IO" chip. This did the serial and parallel ports, JTAG, I2C and a few other things. Many of them could also handle an ISA bus. The Super-IO chip here is the "EliteMT" chip just left of the Northbridge. Below it is the L2 cache.
Before MSI decided to start giving their boards names it was just MSxxxx and so it is with the MS6156, a completely unremarkable mini-ATX Slot1 motherboard often used by Fujitsu-Siemens. This one was absolutely bog-standard for the day. The rear had two serial ports under the parallel, three audio ports (line out, line in, mic in) under a game/MIDI port and two USB next to the PS/2 ports.
Sound wasn't bad, being provided by the Ensonique ES1373 (aka SoundBlaster PCI128) but otherwise the motherboard was exactly what you'd expect from something intended to be sold to and by OEMs. Minimal silkscreening (the white print on a board), no provision for a case fan and the cheaper VIA VT82C693A (Apollo Pro 133) chipset which had some reliability issues but could run a 133MHz bus and 133MHz RAM.
This particular board was supplied with a PentiumIII 600E, the Slot-1 Coppermine with 256k on-die L2 cache but only a 100MHz FSB. Due to the larger capacity of a Slot-1 heatsink, that chip could almost certainly be clocked to 133MHz FSB and run at 800MHz which was more standard for Coppermines.
Legendary. Truly, truly legendary. Very few motherboards had the kind of impact on enthusiast or high-end home computing that the Abit BP6 did. You see those images right: That's two Socket 370s.
Intel never made a BIOS or a specification to allow Socket 370, then for Celerons, to run SMP - But Intel did not disable the interface on the Celeron which allowed it to select being CPU 0 or CPU 1. Abit stepped in, wrote the proper BIOS for it, laid the traces and the result was legendary. Two Celeron 300As at 450 MHz each were enough to stomp all over anything and everything. £400 would get you a high end dual-CPU workstation. Before the BP6, your SMP options were Pentium II Xeons or elderly Pentium Pros, where a single CPU alone would cost much more than an entire BP6 system. In about 2006, I came across a pair of Pentium II Xeon 450/2MB CPUs, the highest end Xeon for some time. They were £2,100 each at the same time a BP6 was on the market.
Later, Intel was to realise there was a market for SMP outside huge servers and officially sanction SMP on Slot-1 (a Shuttle HOT-649A I had was one of the best here). BP6 was first, though.
One of the legends. The KT7 was the first Socket A motherboard to be not crap. In the days of the first Thunderbirds and Spitfires (Athlon-B and Duron-A), the KT7 was the board to have. It ran paranormally stable, was capable of overclocking well beyond what you'd expect. In my use, it took a stick of memory rated for 66MHz to an incredible 152MHz.
The KT7 was incredibly popular among enthusiasts, as was its successor, the KT7-A. Primary among the reasons why was the VIA KT133 and KT133A chipsets, the highest performing SDRAM chipsets ever made on any platform. The blue and white wire you see in the image is a voltage mod, which adds 0.25V to the CPU core voltage.
The RAID variants of this motherboard used the Highpoint 370 and 370A chips. Users of Windows Vista will find that the XP drivers for the HPT370 work just fine and in fact perform a little better under Vista, due to the streamlined disk access stack in Vista.
The MOSFETS on this board show Abit's then-trademark overengineering. The CPU supply is three-phase and even the RAM has its own single phase supply. This overengineering was expensive and Abit's motherboards were not cheap. Not long later, Abit would run into financial difficulty and the engineers responsible for this glorious motherboard would go to work at DFI, transforming DFI from a middle of the road uninteresting vendor to the enthusiast's prime choice, just as Abit had been.
This particular board powered the very first Hatserver.
At this time, dated to late 2001 by the presence of the VIA 686B southbridge, VIA were a strange company. Nobody could beat their Athlon chipsets, KT133 was spanking 440BX in SDRAM performance, but their Intel chipsets were, for a word, dire.
FIC, for their part, had always had a very close relationship with VIA and were usually first to market with a new VIA chipset. In this case, the Apollo Pro133T, newly qualified for operation with the new Pentium-IIIS Tualatin-core parts.
Against the RDRAM-crippled i820, the VIA 133T easily won, but against the crop of old BX 133 motherboard and even some of the newer i815s the comparison wasn't so clear cut. VIA chipsets on Intel were never quite stable and they took a lot of tweaking to get peak performance but once at peak performance they were certainly competitive.
The FA15T was from the first (and only) generation of revised Tualatin-compatible boards, others included Asus' well regarded TUSL-2 and TUV4X (also using the 133T chipset) but all followed the standard layout seen here which was something of a de facto standard during the later SDRAM and early DDR days, RAM to the right of the CPU with one AGP and five PCI below the CPU with the southbridge just to the right of them.
This one powered the server you're reading this page from until mid-March 2008, running a 1GHz Celeron underclocked to 750MHz. It was never really happy because the only RAM I had was PC100 ECC and VIA chipsets were never too keen on ECC RAM, especially not when mixed with normal unbuffered.
Micro-ATX with onboard-everything and based on VIA's highly integrated KL-133 chipset, which was the KT-133 with onboard S3 ProSavage video, a rather feeble 120MHz 1x2 3D accelerator. The KL-133 was rather similar to the KM-133 but lacked support for an AGP slot, which brought down its pincount and cost.
While this board does have a "Host clock" jumper listing 100/133MHz, this has no function as the KL-133 is based off the KT-133 and not the KT-133A, so has no support for a 133MHz FSB. This motherboard, lacking 133MHz bus support, could not run any CPU faster than a 1200MHz Athlon and with only two SDRAM slots could only take 1GB of RAM using fiendishly expensive 512MB PC133 modules.
As can be seen around the VT8364 northbridge, several smaller electrolytic capacitors are bulging and two have started leaking orange electrolyte. Four out of the six GSC 1000uF 6.3V capacitors are showing signs of fault. Though operating fine with the Duron 900 in the CPU section, it would not have been for long due to the leaky caps so was retired from service.
The KX7-333 was a member of a series of boards started with the KR7 on the disappointing KT266 chipset. Next was the KR7A on the KT266A chipset, which was much better. The KX7-333 was next, on the KT333 chipset. The actual difference was negligible both physically and logically. The KT266 was a dog, but the KT266A and KT333 were near identical.
VIA was quickly iterating its EV6 bus interface to get the clocks up and fixing bugs as it went. From the first DDR chipset, the KT266, which was mostly poor (stable, but slow), VIA released the KT266A, mostly a bugfix, but could run PC-2100 DDR RAM, at a base clock of 133 MHz. They'd usually overclock to around 150 MHz (300 MT/s via DDR) but RAM of the day almost always couldn't go that fast. By the time of the KT333 chipset, able to run PC2700 (166 MHz / 333 MT/s), VIA had ironed out all the remaining quirks.
KX7-333s were reknowned for being very, very fast, mixing it with the nForce motherboards of the day but never entirely trustworthy, they were at the level of an average motherboard and nowhere near the ungoldly stability of the KT7. Overclocking was mostly poor, some got theirs as high as 410MHz but even with some very hardcore mods, this specimen was never able to pass 385 with any stability. This board ran runesource.co.uk, its predecessor runesource.cjb.net and hatserver.cjb.net.
Anonymous SiS 730s-Based SocketA Motherboard
Now what could this be? From a starting point, we know:
1. SiS 730s (which was under a completely blank passive heasink)
2. Poor component layout and what the hell is that AMR doing above the AGP?
3. VIA VT6103 fast-Ethernet PHY when the SiS 730s has one built in.
4. Very high revision number (bottom left) of 7.1
The component layout is a hallmark of ECS as is the very high revision number (they get a lot of things very wrong) and the SiS chipset. But back then, ECS didn't design motherboards, they just made them (and made them very well I might add) for other people, providing manufacturing for the likes of Intel, Dell and most top-tier OEMs as well as extra capacity for top-tier motherboard makers like Asus and MSI.
Our next clue is from that ECS didn't design motherboards. They had just bought a company that did...the infamous PC Chips, maker of the World's Worst Motherboards. Have a look on the BIOS chip, bottom right. "M810LM V7.1A". Yes, this is the PC Chips M810LM, perhaps the worst Socket A motherboard ever made, tied only with the ECS K7S5A (which was designed by PC Chips too). It's no surprise that this board came out of a dead PC which I'd diagnosed to a faulty motherboard and neither is it a surprise that it took some detective work to identify it as PC Chips; They know they have a bad name.
A PC Chips/ECS board will work, but not quite right, never quite stable and eventually it'll just crap out completely. Their ONLY redeeming point was that they were VERY cheap. In 2001, when you could pick up a very overclockable Duron 650 for £15, one of these motherboards for £20 and 128MB of PC133 for £20, all you needed to do was add a £25 15GB hard disk and you had a full working system for £80! Wouldn't work for too long, mind.
See the two holes above and below the CPU socket? Two holes is all they are. They don't correspond to AMD's heatsink mounting specs nor to anywhere a screw should be on an ATX motherboard. See the white plastic stickers saying XP on them? That's alll they are too. The SiS730s was never qualified to run AthlonXP
Absolute trash and not even comparable to the similar Asus A7S-VM which also used the SiS730s, same size, but without the AGP and with four PCI slots...but who'd use AGP video with a board like these?
The -X variant of the successful A7N8X series (nForce2) means it's the middle-of-the-road model. There's no onboard IEEE-1394 or serial-ATA such as with the A7N8X Deluxe but onboard ethernet and four mounted USB2.0 ports are present, with headers for four more USB2.0 ports.
It used Asus' then trademark large passive northbridge heatsink (also seen on the A7V8X) to avoid low clearance fan reliability problems which plagued other motherboards after the first year.
In general the A7N8X-X is a typical Asus motherboard. Stable, well built, a little temperamental, all the expected features and some minor compatibility issues. It had the mounting holes and enough socket clearance for the largest of Socket-A heatsinks, the gargantuan (and ferociously effective) Alpha PAL-8045.
The good ends there. The BIOS was poor, unable to set CPU voltage below 1.65V (standard voltage for AthlonXP). It hated Crucial Ballistix memory, killing two 512MB PC3200 modules in quick succession. Overclocking performance was unimpressive, the board was unable to pass 420MHz on either the FSB or RAM, which was with a better cooler on the northbridge and high quality memory.
This board powered the servers when they were still on the main workstation and eventually became an AthlonXP 3200+, Radeon 9700 and 1GB RAM system which it remained as a secondary PC and was still quite capable until around 2010. It would run Windows Vista quite well, and even ran Windows 7 reasonably well, but 1 GB RAM was not the same in 2010 as it was in 2005.
Socket A came to full maturity at this point. The FSB was running at 400 MHz (200 MHz DDR), RAM was running at 400 MHz, PC-3200. AGP had matured to 8X and 2.1 GB/s transfer rates. All motherboards by this time claimed AGP 3.0, but rarely were. AGP 3.0 was a 0.8V signal spec, which broke backwards compatibility with cards expecting 1.5V signalling. Every AGP 3.0 card was a 1.5V AGP2.0/3.0 universal, and every motherboard was the same 1.5V AGP2.0/3.0 universal.
Asus A7V8X-MX SE
Based on the VIA KM400 chipset (KT400 with integrated video), this is everything an entry level integrated motherboard should be and nothing it shouldn't. It's not meant to be upgraded, just put in a system and left to run. The image is poor and looks very odd, the board was too wide for my scanner and too high, but shows off most of the features of the board. Three PCI, four rear USB, positions for two more USB, two DDR SDRAM slots, three PCI, AGP 8x and the Socket-A ZIF socket. It was the standard square mini-ATX to fit in smaller tower cases and came with a large passive heatsink on the KM400 northbridge the same as the A7N8X-X did, so no unreliable low-profile fan.
Due to one capacitor near the CPU socket, 80mm coolers can't be used, such as the PAL8045.
In all, a decent enough cheap low-end motherboard with onboard-everything.
Asus A8N-VM CSM - 2005
Nvidia's many AMD K8 chipsets gave a lot of choice in motherboards. The Asus A8N-VM CSM and A8N-VM looked identical, but within the chipset lie a small difference. The A8N-VM CSM is based around the Nvidia GeForce 6150 and nForce 430 MCP, offers one PCI Express x16, one PCI Express x1 and two PCI slots, as well as four DDR DIMM slots, all populated with 512 MB modules in this example. Audio is provided by the Analog Devices AD1986A 5.1.
The A8N-VM used the Nvidia GeForce 6100 and nForce 410 MCP. They were exactly the same silicon, so perfectly pin compatible, meaning the same board design can be used for both. nForce 410 disables the onboard gigabit ethernet, so Asus added a Realtek ALC8201. Two of the serial ATA ports are disabled on the 410, so they're simply not fitted to the A8N-VM.
The second difference was the GeForce 6150 and GeForce 6100. Here the difference is even smaller. The 6100 clocks at 425 MHz, while the 6150 clocks to 475 MHz. The 6100 enables only a VGA output, while the 6150 can also handle DVI and component. Finally, the 6150 has a video decoder able to handle high definition media. The GPU integrated is a stripped down NV44 (NV44 was GeForce 6200), and the chip holding it is the Nvidia C51, belonging to the nForce 4 family. C51 was intended for lower cost, highly integrated systems and notebooks/laptops. It has two pixel shaders, one vertex shader and two pixel pipelines in both 6100 and 6150 configurations
nForce 4 was the last Nvidia chipset anyone was likely to see, however Nvidia made three subsequent lines on AMD processors, nForce 500, 600 and 700, then dropped the nForce brnading for two more chipsets, GeForce 8000 and then nForce 900 (A rehash of nForce 700). Intel chipsets from Nvidia stopped at Core 2 level, as Intel was unwilling to allow competition in the chipset space for Nehalem and onward. Nvidia did not make an AMD FX chipset and exited the chipset market in 2010.
Class action lawsuit later, Nvidia took quite a hit against the GeForce 61x0 series. The chipset northbridge used a poor solder formulation, causing its BGA solder balls to detach from the motherboard.
Back to the board. This one has the AMD Opteron 165 (from the CPU section!) fitted, 2 GB RAM, and was running Windows 7 as a home-office machine until 2016. It's still quite functional, simply obsolete.
Asus A8NSLI Deluxe
A very high end motherboard based around the NForce4 SLI chipset, this was paired with the Opteron 165 in the CPU section. Alas, my overclocking ambitions were somewhat limited by a chipset which wasn't at all happy beyond 300MHz (giving a CPU clock of 2700, far beneath what even the worst Denmark Opterons were capable of). In attempting to fit a better heatsink to the chipset, a mistake was made and the chipset took fatal damage. The slight chip in the top left corner killed the motherboard.
With two ATA, eight SATA, three legacy PCI and four PCIe, this was a motherboard designed for flexibility and power. The board did have some glaring oversights, however. The fan PWM controller could only control the CPU fan (the chipset fan was linked directly to it) and the first chassis fan (bottom right corner) and even then, the chassis fan would only be controlled in software such as Speedfan. The fan headers near the PCIe x16 slot and just above the CPU socket had no control at all.
When an overclock failed and the system wouldn't POST, most motherboards would reboot to default settings but 'remember' the failed settings so they could be modified. The A8NSLI Deluxe required clearing CMOS with a jumper under the bottom PCI slot, a location very difficult to get to if all three PCI slots are populated.
The SATA ports are badly positioned (why aren't they were that barcode sticker is?), Asus only know what the ATX power connector is doing all the way over on the other side of the motherboard when it should be on the top edge and the stock fan for the chipset was on an extremely poor heatsink, necessitating a very loud fan.
When working, the A8N series of NF4 motherboards (all this design) were good and fast and stable. When being worked on, tweaked or pushed to limits, they were not happy with it nor fun to work with. Not what a high end motherboard should be.
Gigabyte GA-MA69G-S3H - 2007
This is written as a review, as the product was bought retail in 2008, and this review posted then. Most of the points were timely for the day but may be dated or obsolete today.
After acquiring ATI, AMD suddenly had a chipsets division. This chipset is the ATI RS690 or simply, the AMD 690G. Onboard it features a Radeon X1250 (actually a descendant of the Radeon 9700 and sharing no feature at all with the X1xxx family), but it's more conspicuous for what it lacks: No ethernet! AMD's southbridges did not gain ethernet until the AMD 800-series! The SB600, 700, 700S, 710 and 750 did not support ethernet!
Seriously AMD, WTF? No ethernet? What year is this, 2001? No, SiS and Intel were both putting ethernet on their chipsets in 2001.
Let's make this completely clear: Onboard video has to drive a monitor. Nothing else. I don't care how many pixels it can texture or how good the shaders are. As long as it can run a Windows desktop, it's what's needed. Intel have had this down for ages. Instead AMD have wasted complexity which could have been better used, perhaps, putting a goddamned ethernet MAC on the bloody chip!
The video isn't bad, though. It's about half the performance of a Radeon 9700 (from 2002!) which puts it in line with the Geforce FX 5600. It has onboard HDMI (and a DVI dongle for it), VGA and doesn't need one of those ungodly whiny fans on the Northbridge, just a small passive heatsink.
The IGP identifies as "Radeon X1250" which puts it right against Nvidia's Geforce 6150. The Geforce 6150 is built into NV's NForce 430 chipset, so how do they compare?
GPU Core Clock Arch. SM
X1250 RV370M 400MHz 4x1 2.0 (4x pixel, 0x vertex)
6150 ~NV43 475MHz 2x1 3.0 (2x pixel, 1x vertex)
ATI say that the GPU is half the RV410 in the X700 but the GPU has no support for pixel shader 2.0b, this makes it in the R300 (Radeon 9700) descent, not from the rework to R420. It is architecturally almost identical to Radeon 9800 SE, X300, X550 and X1050. R420 decoupled the shaders from the ROPs, this does not have such decoupling, so it is not a derivative of RV410. Instead, it's a version of RV370 (X300, X550, X1050) and quite old.
This makes the Gigabyte motherboard a little strange. Of the ten possible USB2.0 ports, it has four on the back and headers for six more, this is good as it's utilizing the chipset's available features. It adds three IEEE-1394 ports (two headers, one rear port) courtesy of a Texas Instruments 78AF5RW, one has to ask why? Surely the board space would be better used in adding an ethernet controller than adding four channels of a dead technology? Instead, we're stuck to a single Realtek 8110C gigabit ethernet when almost every other board other than bargain basement nothings can sport dual gigabit. Where the second ethernet port would be on the back is an IEEE-1394 port. If we want it, we'll add the headers ourselves!
About a year into the operation of this motherboard, the onboard ethernet disappeared. Plain wasn't detected as present by any OS and the smart cable detect in BIOS just hung.
The port layout and the ridiculously out of place IEEE-1394.
I'm left wondering who's going to buy this board. Gamers don't give a flying f--k about the wimpy onboard video. It's not mATX, so small system fans aren't going to chase it. It can't even control the system fan, so quiet PC guys aren't going to take advantage of the onboard HDMI output because it'll be too noisy.
So why did I buy it? It was cheap and I didn't know it only had one ethernet port.
I rated it 65% when I first bought it, as it was a little better than typical (I use the whole scale, so 50% is a perfectly acceptable middle-of-the-road product).
It's a bit confused, the board features make little sense, but it's well laid out, the AMD 690 chipset is a proven solid workhorse and the DIMMs don't foul a PCIe video card. It's certainly worth considering, but it's nothing special. If it could control the system fan it'd be good for a HTPC (home theatre PC). It's as though Gigabyte started designing a mini-ATX board then decided to upgrade it to full ATX and just added random features without consideration to what would make sense. To me, that puts it 15% above average but really only because it's a solid board with a solid chipset.
Oh, and it worked with a Phenom II X4 955, but only at 800 MHz. This is because the board has single plane power and while the board is rated to 125 watts, AMD's AM3 processors require dual power planes at 125 watts and will clock down to 800 MHz if they detect a board without. By hammering the FSB (HT clock) I was able to get a working quad core 1,400 MHz system. Not too shabby, but soon replaced. The motherboard is one of many which have run the server hosting this website, and it was the incumbent from 2012 to 2015. It was sold soon after it was replaced by a HP Microserver.
Asus Crosshair III Formula
This socket AM3 motherboard was paired up with a Phenom II X4 955 Black Edition and immediately something was not quite right. The system would power on, but then not proceed to past the POST sequence. Flexing the motherboard by pushing the large CPU heatsink would resolve this, but when running it was never quite stable.
I eventually attributed it to shipping damage, it was likely that the AM3 socket had a flaky connection somewhere.
When running, it was a truly magnificent sight to behold with everything and more (including LEDs everywhere, which can mercifully be turned off). AMD's 790FX chipset, heatpipe motherboard cooling, five SATA slots
and a "SupremeFX X-FI" card, actually an Analog Devices AD1989B relabeled as AD2000B and slapped onto a PCIe 1x card when it would have been better off on the motherboard proper. And likely better off as a Realtek, which enjoy much better driver support as well as superior HD audio characteristics. Creative's bundled "SB X-FI Mobo" software added some rubbish to the control panel and an emulated EAX/OpenAL. Joy.
On the back we're treated to a button to clear CMOS (!), six fitted USB2.0 and six more able to be added to standard headers (backplanes supplied), a single IEEE 1394a port and a header for another.
There are buttons everywhere. Aside from the rear-mounted CMOS reset, you also get buttons to reset and power on the motherboard, mounted right on the motherboard. Vaguely handy, I guess.
Asus' "MemOK!" (another button) is intended to smooth over RAM incompatibility problems by essentially brute-forcing RAM timings until one is found which works. Normally, this is the slowest. It can do all kinds of other weird overclocking things, but the only one really worth pointing out is that it can save multiple copies of the entire BIOS set up as profiles, so set it up and save your profile. If you change stuff and it goes south, you can reload a profile and everyone's happy.
It also features AMD's "Advanced Clock Calibration", essentially a core unlocker.
Through overclocking, steady DDR3 speeds of 1600 MHz are possible and this is clocked cleanly, so doesn't really count as overclocking except the CPUs don't officially support it.
On to the ugly. AMD's SB750 has perhaps the world's worst SATA implementation. When the CPU is clocked down due to little load, disk performance tanks. As CPUs are normally under very little load when accessing the disk (as the task is loading, rather than running!) this drags down the entire system's performance. There is actually a noticable difference between the default 800 MHz minimum power state and using Phenom MSR Tweaker to prevent it ever going below 2 GHz. Power draw is higher, of course, but the system is also much more responsive.
Asus' chipset cooler, while it looks impressive, actually has a very poor connection to the AMD 790FX northbridge, resulting in temperatures surpassing 90C under load. This caused me a brief "eureka" moment when trying to stabilise the system, only to discover that even keeping the chipset below 50C didn't help matters.
This motherboard can do practically everything except, in my case, work properly.
Gigabyte GA-990X-UD3 Rev 1.0
AMD's 800 and 900-series chipsets were not really new chipsets per-se, they were iterations on the 700-series. They followed ATI's naming convention, so the 990X was the RD980 chipset. RD990 was 990FX, and RX980 was the 970. The 990X chipset drops sixteen PCIe 2.0 lanes, but still allows multiple GPUs by using a x8/x8 configuration. 990X also drops four of the "general purpose port" (GPP) PCIe lanes, presenting only 6 of the total 10. The two GPP clusters, six lanes in GPPa and four lanes in GPPb support any mixture of slots from GPPa, such as x4/x2, x4/x1/x1, x2/x2/x2 or even six x1s. The GPPb only allows a single device, so must be connected in x4 configuration. GPPb is, of course, not present in the 990X chipset.
On to the board. The layout of the board is excellent, but for the silly ATX power connector. Like many boards, it blocks drives in cases which aren't as deep as they could be. The PCIe layout of this board was bloody weird for a 990X board. It uses PCIe switches (the four high density chips just below the top PCIe x16 port). The upper port was full x16, but became x8 if the middle x16 port was in use, this is electrically x8. Look carefully, and you see the pins don't reach to the end of the slot. The lower port is wired from, it seems the second block of 8 lanes from the PCIe GFX of the 990X chipset. They go via the switches to the middle port and, also, the lower port, which is a very interesting port. Looking at the picture, you can clearly see the right half of the slot is blank, so it's a x8 slot at x16 size. This is okay, it allows a video card to clip in securely. However, half the contact housings inside the slot are not populated, so the x8 slot in x16 size is wired up to be only an x4 slot! All three x16 physical slots share the same 16 lanes on the 990X. The extra six lanes are wired to, respectively:
Furthermore, the board's PCI bus (hosted by the SB950 southbridge) has two slots, a SuperIO controller (left of the lower PCI slot) and a VIA VT6308P IEEE-1394 controller. As IEEE-1394 is licensed per port, the board only comes with one port, but has provision for two more.
- Upper PCIe x1 slot
- Lower PCIe x1 slot
- EtronTech USB3.0/3.1 Gen 1 controller (lower right)
- EtronTech USB3.0/3.1 Gen 1 controller (just below VRM heatsink)
- RealTek gigabit ethernet controller
Gigabyte is proud of the 108db SNR audio on this, and it aced Rightmark Audio Analyzer. The ALC889 is serious kit, featuring 10 DACs, 6 ADCs, 8 remappable ports, the ability to do full 7.1 and run headphones concurrently with different sources and can handle S/PDIF input and output. It'll also do Dolby DTS, SRS TruSurround XTTM and is cerfified by Dolby PCEE Program for studio mastering. It's also capable of bit-exact S/PDIF pass through to software.
In a double blind test at Tom's Hardware, the ALC889 was not distinguishable from a $2,000 Benchmark Media DAC2 HGC. My experience with some now-very-old Sennheiser HD-202 headphones mirrored this: The ALC889 was transparent. It added and took away nothing from the sound. Oddly enough, I also discovered that the very cheap Sennheiser HD-210 cans do an excellent job for their price.
This board was paired with a Phenom II X4 955 Black Edition chip, and tasked to run it as fast as it could. The Deneb C2 stepping ran a little hotter and clocked a little lower than the C3, so I wasn't expecting miracles. I got it stable at 3.72 GHz and managed to boot at 4.0 GHz and an ungodly 1.65V core voltage, which even the Corsair H80 liquid cooler couldn't keep under control. Load temperatures rapidly rose about 90C. At 1.45V and 3.8 GHz, it was just barely able to function, with quite frequent crashes under load. 3.72 GHz was stress test limit for this chip.