CPU/Mobo Dual Chipset AMD board tested(Foxconn NFPIK8AA-8EKRS)

Status
Not open for further replies.

dipdude

Forerunner
For the detailed review with benchies go to : Lostcircuits



An Opteron 875 will work in the NFPIK8AA-8EKRS as will any other Socket 940 processor, with or without dual core technology.

Summary

Foxconn is one of the world's largest OEM manufacturers of computer parts and only recently has started to enter the enthusiast market. Overall, the brand appears to be well-received, even though the naming conventions are somewhat counterintuitive. Enthusiast boards on one side, OEM boards on the other side, there is still ground to play and it appears as if the red tape was lifted at least for once to let the engineers and designers play their hearts out. The result is a true oddity in the conformistic world of motherboards - a dual chipset motherboard for a single - server class - CPU with two 16-lanes PCIe SLI slots that promise to deliver uncompromised bandwidth to the graphics adapters while in the background some super RAID configuration is on a rampage and the entire system talks to the intranet via dual Gigabit Ethernet ports. Needless to say that the latter are protected by nVidia's own Firewall.

All of this could be nice and dandy but often enough, we have been dealing with hardware that looks great at first sight and then turns out to be a dog. To make a long story short, such worries turned out to be unfounded, on the contrary, the NFPIK8AA-8EKRS we were dealing with had a few other surprises for us with a few huge D'UH effects that left us marvelling why we have not seen the same features elsewhere - so simple were they. There is only so much we can say in two short paragraphs, but it sure might be worth reading the rest of this review.

Final Thoughts

On the other side of the coin is the actual mainboard, one of the finest pieces of hardware we have come across in a long time. From a reviewer's or expert standpoint, the most amazing feature of the board is the overall love and attention to detail, everything is where it is supposed to be and where it makes most sense. This includes the jumpers that only raise the question why not everybody else does the same thing, after all, it is so simple to add the boot block jumper or that for the user defined BIOS table.

Needless to say that the infrastructure of the board is spectacular by itself. A dual chipset for a single CPU may appear overkill but just the possibility to stripe or mirror over two SATA controllers with a huge bandwidth headroom to accomodate up to eight drives in a single array is something we can only marvel about. And then there are the performance numbers we showed earlier. Arguably, nobody in his or her right mind will stripe over 6 or 8 drives, the failure rates would be unacceptable. However, as a demonstration of the possibilities of the system on the chipset integration level, this configuration served us just right and, if nothing else, underscored what the mainboard is capable of. .

Dual 16 lanes PCIe slots for SLI appear a bit overkill. Keep in mind that all data have to move through the memory and the hypertransport interface and regardless of how wide the pipes are on the SLI bus, if the backend cannot supply more than 5500 MB/sec then 5500 MB/sec is all that will ever fill the 32 GB/sec capacity of the interface. In reality, again, there is the limitation that the 32 GB/sec are for concurrent read and write transfers, which leaves 16 GB/sec overall unidirectional bandwidth that is shared between the two cards. For a single card, this results in 8 GB/sec and all of a sudden, the headroom is there but no longer appears astronomical considering the fact that the two SLI cards work in AFR mode. In other words, there is some sense behind the strategy, even if consumer applications like games are not able to take advantage of it. The story may be entirely different for professional graphics applications such as CAD or high-end OpenGL and we'll follow up on that matter shortly. Another situation where the extra lanes may become important is high level anti-aliasing in SLI mode where additional data transfer from one graphcis adapter to the other will cosume extra bandwidth.

Of course, there is always the issue of whether any esoteric design, regardless of its idiosyncrasies and whatever plethora of fantastic features will have any niche in today's marketplace. In other words, even the greatest design will not be a commercial success if there is no interest in its features and capabilities. In the case of the NFPIK8AA-8EKRS, hopefully, there shouldn't be any worries, since this board is ideal for the system integrator, for the small business in need of a classy file server, for the CAD designer with need for extreme reliability and as a data base server taking advantage of the Error Checking and Correction capabilities of the memory subsystem. In other words, as odd as the design appears at first glance, the NFPIK8AA-8EKRS is a mouth-watering dream of professional motherboard come true for any system integrator who's capabilities go beyond navigation of the system configuration on Dell's website.
 
Ok,

Well, first let me say that I've got two of these boards.

The thing that stands out about board is the build quality. However, there are a few things that are not as great.

They are:

1) FoxConn's lack of BIOS updates to ensure complete functionality with new OSes as they come out. (Not what you'd expect from someone trying to enter the server / workstation market...)

2) ACPI v1.0b BIOS compatibility, rather than v2.0 which every vendor has been including on their newer flagship products. (Which this one was when it came out... and it has been being sold through-out 2005/2006.)

3) Lack of comprehensive Opteron socket 940 series support... So this means that while it is claimed FoxConn developed the BIOS to support all Opterons, OSP variants are not supported. [More over, they refused to support them even though their documentation indicates ALL series 100 and 200 Opterons are supported.] (OSP variants are the Opterons used in servers by HP and IBM, and are lower power variants made from the best of the silicon yield... therefore needing less power to operate at high CPU frequencies.)

So, what all this means is that:

The FoxConn NFPIK8AA DOES NOT RUN >>LINUX<<

This is because the BIOS of the system will not enumerate IRQs on the PCI/PCIe bus, and this causes 2050 chipset NOT TO WORK AT ALL UNDER LINUX.

This means that ANYTHING attached to primary NForce Pro 2200 sorta works and anything attached to the 2050 "Companion Chipset" NForce Pro does not.

The big issue is that due to the IRQ conflicts as a result of the substandard ACPI BIOS used on a workstation/server class product, both chipsets fail to operate properly under one of the most popular operating systems on the planet.

---> I've installed 16 Linux distros so far..... and they all fail.

Windows on the other hand seems to work fine. Which I guess is a plus...

On the flip side, the SuperMicro and Tyan based NForce Pro 2200 / 2050 boards support ALL Opterons (200 series / 100 series / 800 series), and actually work with 2.4 and 2.6 Linux kernels.

So what I'm basically saying is: Great board from a hardware perspective. Horrible from a support perspective.

Gerhard
 
Status
Not open for further replies.