Categories
Hardware Technology

Foxconn A79A-S Final verdict: Unfit for use

I must admit that I forsook my vow never to build a PC again, and as a result am now more than convinced of its validity than ever…

Read on for a final look at the Foxconn’s flagship A79A-S AM2+ motherboard, and how it is eclipsed in almost every way by the Gigabyte GA-MA790FXT-UD5P board with which I have now replaced it.

I don’t want to dwell on this for too long, so here’s the shortened version of why I’m more convinced than ever that if you want a PC then you should probably be buying one off-the-shelf:

After the instability of the A79A-S I wanted something provably correct, and saw that the Phenom II CPUs have native ECC support in their on-chip memory controllers. This seems to be a pretty rare requirement, since ECC support doesn’t seem to be documented anywhere – so I contacted both Gigabyte Technical Support (who confirmed that ECC memory was supported, and even showed me the part number of the ECC DIMM qualified against this board) and Crucial.com Technical Support (who, after first insisting that the chosen replacement board was actually a DDR2 board – it isn’t – confirmed that, in their opinion, the board should support ECC DIMMs but that they hadn’t tested this combination). Predictably, it turns out that ECC modules do not work, and the board fails even to POST. After going backwards and forwards or at least two weeks confirming what the problem might be, and Crucial allowing me to place an order for out-of-stock modules with next-day delivery without telling me that they’re out of stock, the replacement non-ECC modules arrived today – at half the capacity I ordered. Since building this new system, the 955 CPU I ordered has been superseded by the new 965 – that’s how long it’s taken! At least it’s fairly stress-free now – the worrying time was when, not having another system to test DDR3 memory in, it was unclear which component was faulty. I’m now running on 2GB memory until Crucial can fulfill my original 4GB order, which has been delayed because they have a DIMM shortage!

It’s probably worth a recap of the (many) problems of the Foxconn A79A-S:

  • it always had problems with USB devices (about 1/3 of the time failing to initialise mice or keyboards on boot and frequently failing to boot at all with a USB Bluetooth adapter inserted – this latter problem inexplicably causing a CMOS reset);
  • for much of its life it was unable to boot with DDR2-1066 memory set to run at more than 400MHz (x2), and then even after a BIOS update resolved this issue the board would require a CMOS reset if you told it to do anything as radical as using the DIMM’s SPD data for memory tunings;
  • various PCI-Express BIOS options would cause the machine to hard-lock either immediately or only when the internal codec is used to play sound (good luck on diagnosing that one…);
  • setting any PATA/SATA interface to “Legacy IDE” mode would cause the board to freeze before POSTing, requiring a CMOS reset to recover;
  • SATA ports are labelled backwards, with the BIOS inserting connected eSATA drives before on-board drives – meaning that a linux OS booted without an eSATA device connected sees its root partition on /dev/sda, but with an eSATA drive present instead sees its root partiion on /dev/sdb;

…and finally,

  • pressing the reset button briefly would cause the system to fail to restart until powered off – it has to be held down for several seconds.

All in all, one of the worst pieces of engineering I’m had the misfortune to come across. To rub salt into the wound, this was not a budget option – is cost just a shade under £200 when new, and is still about that price from certain retailers. To add insult to injury, the PCIe slots were of non-standard spacing – so that if you did ever want to run a Crossfire/SLI setup in any of the four 16-lane PCIe slots fitted, you’d have to order (at great expense, no doubt) a custom longer bridge-board from Foxconn in order to connect the cards together.

I was never sure, and still am not, quite what proportion of the problems with the A79A-S were down to the board itself, and what proportion were down to the AMI BIOS used on it. As mentioned before, recent Linux kernels have specific checks for AMI BIOS due to its tendency to corrupt the host machine’s low memory, especially after resuming from sleep.

In any case, I shelled out again for a Gigabyte MA-790FXT-UD5P, and so far I couldn’t be happier.

On the one hand, it’s certainly not as tweakable as the A79A-S – there are no BIOS settings to manually adjust the PCIe timings and configuration, for example. Since most of these resulted in a corrupt CMOS on the Foxconn board, though, this is probably no great loss. With non-ECC memory, the board worked first time – although it does seem strange that memory timings can be set to Auto (e.g. slow) or Manual, where you have to specify every setting yourself. SPD data is displayed alongside, but there’s no option to use use the SPD data. Very odd design choice…

Annoyingly, Gigabyte have removed the option to enable or disable Numlock, and have hardwired it to be on. This is exactly what you don’t want if you have a numeric keypad-less IBM Model M Space-saver keyboard, and want to be able to type anything other than numbers with the right-hand keys. The inexplicable lack of any indicator lights on this keyboard doesn’t help matters either.

The major chance between the two boards, though, is software – and specifically Linux – support. The A79A-S threw up SATA ‘softreset failed’ errors, stated that memory was being lost due to incorrect BIOS setup, output pages of complaints about malformed and incorrect (Microsoft-compiled) ACPI data, and frequently failed to resume from any form of sleep, requiring a power-cycle to continue. The Gigabyte, on the other hand, is much improved: Linux boots and suspends and resumes (from hibernation) without a line of complaint – from the same kernel! The ACPI tables are marked as ‘GBT’, suggesting that Gigabyte care enough to build their own tables – which, given that they appear to work, can be no bad thing.

The A79A-S had lofty ambitions and, on the face of it, an excellent hardware specification. However, this was marred by inexplicable design choices (such as the non-standard PCIe slot spacings), broken hardware support (for things as standard, these days, as USB), and general instability under Linux and Windows. The whole package felt unfinished even after BIOS updates, and reeks of throwing in all of the best toys and then shipping without taking any time or care to check that the combination actually works.

Even if the A79A-S were a budget board, then this would still be indefensible – but at almost £200 it is a shocking travesty. I honestly feel that this board is unfit for its intended use under the Trade Descriptions Act (1968) – and I for one will never again make the mistake of again buying a Foxconn-branded product(*).

The Gigabyte board, on the other hand, is the polar opposite: a well designed and informative, if not overly feature-laden, BIOS sits before an able and high-quality mainboard which exudes capability and reliability in a way that the A79A-S failed to from day one.

(*) Yes, I realise that they manufacture most of Apple’s designs. Apple, presumably, have some concept of fitness and quality control which Foxconn themselves appear to be sadly lacking.

5 replies on “Foxconn A79A-S Final verdict: Unfit for use”

In all fairness, I may have spoken too soon. Whilst I still believe that the Gigabyte is a better board, it too has some issues:

  • If MCE-on-ECC-errors is enabled, then Windows 7 will fail to resume from hibernation;
  • Holding down a key (such as F12 to go to the boot-menu) during POST will result in the BIOS freezing and a hard-reset being needed to continue;
  • Sometimes, when shutting down or restarting from Windows 7 or from Linux, the machine will power-off and not power back on again until power is physically removed for at least 30 seconds.

    Two different processor/motherboard combinations (I had both of the initial items replaced at the same time due to my assumption that the power-off problem was a hardware fault) and three different BIOS versions have shared these problems.

    I’d suggest that manufacturers just aren’t paying enough attention to non-Intel boards these days, but then we hear that the “Intel P55 Express Chipset” Core i5/i7 chipset also has it’s own troubles… so perhaps non-OEM motherboards, lacking a big company behind them to insist on quality control and correct operation, are suffering at a time when all are trying to reign-in costs. Perhaps the burden of having to remain compatible with the vast amount of PC-compatible hardware that has appeared over the years is now too great. In any case, quality is definitely suffering and these expensive items simply don’t work reliably 100% of the time. I still have some old mainboards at home from computers which are now laughably old and slow (a 350MHz K6-3 and a 1.5GHz Athlon, IIRC)… but the one thing that they share in common is that they worked perfectly first-time and were absolutely stable and reliable. It is worth noting that it is in the intervening years between these products being new and now that there has been a massive consolidation in the PC motherboard market, with most companies going out of business or being bought up – so that now there are only a small handful of different manufacturers, releasing hardware under several more different brands. The drop in quality does seem to have accompanied the drop in competition.

    Plus ça change, plus c’est la même chose

I applied every firmware update released between when I purchased the board and when I finally gave up over nine months later – just as described above, in fact. It took several months and a couple of BIOS updates before the board even supported running DDR2 memory above 800MHz without locking-up.

And, if you look, the most recent update for the A79A-S was on the 17th March – well before the 14th August when this summary post was written, so all of these comments apply to the latest (and almost certainly final) BIOS.

Yes, firmware updates eventually improved the A79A-S beyond its shockingly bad initial state – but even with these, the board simply doesn’t work properly.

I’ve looked at those carefully – especially the Tyan boards which have a great reputation and look to be of excellent quality. However the problem is that (when I last looked, at least) most of their boards are either AGP-era, or are based on nVidia chipsets – which I wouldn’t touch with a barge-pole. The relatively few HT1000/HT2000 boards either lack PCI-Express slots, or have the slots but note that they can’t be used for Graphics boards. Whether this is a hardware/chipset limitation, a BIOS/firmware issue, or whether everything would work perfectly fine but just hasn’t been qualified to do so, I don’t know… and I’m not especially keen on spending that sort of money to find out!

The LS21 blades in our IBM eServer Bladecenters are based around the Broadcom/Serverworks HT2000 chipset and, save for an annoying IBM firmware bug which means that enabling HPETs prevents Linux from booting, their Linux performance and compatibility has been, as you’d hope, excellent.

Leave a Reply