Be still my beating heart.
And there’s more to the GA-K8NSNXP-939 than two of everything. The board also sports NVIDIA’s nForce3 Ultra chipset, a feature-rich BIOS, and a handy external Serial ATA adapter. What’s not to like? A couple of things, actually. Read on to see where the GA-K8NSNXP-939 shines and where it stumbles.
Before running the awkwardly-named GA-K8NSNXP-939 through the gauntlet, let’s quickly thumb through the board’s spec sheet.
|CPU support||Socket 939-based Athlon 64 processors|
|Chipset||NVIDIA nForce3 Ultra|
|North bridge||NVIDIA nForce3 Ultra|
|South bridge||NVIDIA nForce3 Ultra|
|PCI slots||5 32-bit/33MHz|
|AGP slots||1 AGP 4X/8X (1.5V only)|
|Memory||4 184-pin DIMM sockets
Maximum of 4GB of DDR266/333/400 SDRAM
|Storage I/O||Floppy disk
2 channels ATA/133
2 channels Serial ATA 150 via nForce3 Ultra with RAID 0,1 support
2 channels Serial ATA 150 via Silicon Image Sil3512 with RAID 0,1 support
|Audio||8-channel audio via nForce3 Ultra integrated audio and ALC850 codec|
|Ports||1 PS/2 keyboard
1 PS/2 mouse
4 USB 2.0 with headers for 4 more
Headers for 3 Firewire via Texas Instruments TSB82AA2
1 RJ45 10/100/1000 Gigabit Ethernet via Marvel 88E8001
1 RJ45 10/100 Fast Ethernet via nForce3 Ultra
1 analog front out
|Bus speeds||CPU: 200-455MHz in 1MHz increments
AGP: 66-100MHz in 1MHz increments
|Bus dividers||CPU:HT: 1:1, 1:2, 1:3, 1:4, 1:5|
|Voltages||CPU: 0.8-1.7V in 0.025V increments
AGP: default + 0.1-0.3V in 0.1V increments
DDR: default + 0.1-0.2V in 0.1V increments
HT: default + 0.1-0.3V in 0.1V increments
|Monitoring||Voltage, fan status, and temperature monitoring|
A couple of oddities jump off the GA-K8NSNXP-939’s spec sheet list off the bat. First, it doesn’t advertise any capability for ATA/133 RAID, despite the fact that the nForce3 Ultra chipset supports RAID 0, 1, 0+1, and JBOD arrays for ATA devices. The nForce3 Ultra even supports arrays that span both ATA and SATA drives, so I’m not sure why Gigabyte chose to leave this functionality off the board’s spec sheet. ATA RAID definitely works on the GA-K8NSNXP-939, though.
Another spec sheet curiosity pertains to the board’s dual Ethernet controllers, more specifically, the fact that the nForce3 Ultra Ethernet is only listed as 10/100 Fast Ethernet. The nForce3 Ultra has an integrated Gigabit Ethernet MAC, but Gigabyte is only implementing the on-chip Ethernet at a tenth of that speed.
Apart from those curiosities, which we’ll explore in more depth later in this review, the GA-K8NSNXP-939 looks pretty loaded.
The GA-K8NSNXP-939’s aesthetic is anything but understated. With a bright blue board, near-fluorescent ports, slots, and headers, and enough bling to catch 50-Cent’s gaze, the GA-K8NSNXP-939 is about as garish as they come. If you want a flashy aesthetic, this board definitely delivers. For the rest of us, the pimpin’ look will be hidden away inside a case anyway.
As far as layout goes, Gigabyte has done a pretty good job populating the board with chips, ports, and slots. Power connectors are located on the edge of the board, near the top, which nicely cleans up cable routing around the CPU socket. The layout isn’t all roses, though.
For starters, there are a couple of Serial ATA connectors sandwiched between the processor socket and the AGP slot. There’s still enough room for stock Athlon 64 heat sinks, but it gets a little tight with a graphics card installed and a couple of Serial ATA cables plugged into the board. Having Serial ATA ports located so close to the AGP slot can also interfere with double-wide passive GPU coolers like Zalman’s ZM80.
In Gigabyte’s defense, the Serial ATA port placement is identical to NVIDIA’s nForce3 250Gb reference board, so NVIDIA’s reference design shares some of the blame. This would have been a perfect opportunity for Gigabyte to deviate from the reference design.
There’s another layout problem around the AGP slot, but thankfully, DIMM slot clearance isn’t it. Instead, it’s the proximity of the board’s nForce3 Ultra chip. The chip is so close to the AGP slot that Gigabyte is forced to use a low-profile cooler. This cooler has hardly any surface area, so it’s largely reliant on a tiny fan to keep the nForce3 Ultra cool. Unfortunately, the cooler’s fan guard is going to restrict air flow right off the bat.
I’m worried about fan noise, too. Although the chipset fan was nearly silent during testing, the tiny fans used in chipset coolers tend to develop an annoying whine over time. Normally, it would be easy to replace a noisy cooler with a larger passive heat sink, but the chipset’s close proximity to the AGP slot could make it difficult to find a replacement cooler that fits.
On a more positive note, the GA-K8NSNXP-939 DIMM slots are color-coded in dual-channel pairs. Under the DIMM slots, you’ll find the board’s ATA/133 and floppy ports, which are conveniently located near the edge of the board.
The location of the board’s third and fourth Serial ATA ports isn’t quite as convenient, though. Those ports are buried at the bottom of the board with the USB and Firewire headers.
As I mentioned earlier, the GA-K8NSNXP-939 employs a six-phase “Dual Power System” (DPS). Gigabyte is quick to point out DPS’s support for Intel’s latest Prescott Pentium 4 processors (on Gigabyte’s P4 boards), so DPS should have no problem delivering clean power to an Athlon 64. The DPS also incorporates a funky VRM heat sink, complete with a fan, to help keep the board’s power circuitry cool.
Around the back, the GA-K8NSNXP-939’s port cluster is loaded with goodies, including two Ethernet jacks, four USB ports, and a trio of audio jacks.
The board also comes with expansion brackets for two Firewire ports, an additional four USB ports (two not pictured), three more analog audio jacks, and RCA and TOS-Link digital S/PDIF audio outputs. The board also has on-board headers for a game port and an extra Firewire port, although Gigabyte doesn’t include brackets to take advantage of those headers.
And there’s more. The GA-K8NSNXP-939 also comes with a neat external Serial ATA adapter. This bracket extends two Serial ATA ports and one four-pin power connector to the rear of the case, allowing Serial ATA drives to be connected externally.
A smattering of chips
Although the nForce3 Ultra is a single-chip design, the GA-K8NSNXP-939 is also peppered with a wide assortment of auxiliary chips to power the board’s integrated peripherals.
The nForce3 Ultra is the brains behind this operation, though. With a 1GHz/16-bit HyperTransport processor link, AGP/PCI lock, and support for both Serial and “parallel” ATA RAID, the nForce3 Ultra chipset is packed to the gills with goodies. Sadly, though, the GA-K8NSNXP-939 doesn’t take advantage of a couple of the nForce3 Ultra’s more notable features.
First, Gigabyte has gone with a PCI-based Silicon Image Sil 3512 Serial ATA RAID controller to augment the GA-K8NSNXP-939’s SATA RAID capabilities instead of milking all four of the nForce3 Ultra’s Serial ATA ports. Tapping all four of the nForce3 Ultra’s SATA ports requires a couple of PHY chips, but users would be able to span RAID arrays across up to four Serial ATA drives. Given how well NVIDIA’s RAID implementation scales up to four drives, it’s really a shame that Gigabyte isn’t taking full advantage of the nForce3 Ultra’s SATA RAID potential.
Not that the Sil3512 is a horrible SATA RAID chip; it just can’t do four-drive arrays on the GA-K8NSNXP-939. The Sil3512 also sits on the PCI bus, so it has to share 133MB/sec of bandwidth with other PCI devices, including the board’s Firewire and Gigabit Ethernet chips.
That’s right, this board has PCI-bound Gigabit Ethernet. Half of the GA-K8NSNXP-939’s networking duties are handled by Marvell’s 88E8001 GigE chip. The other half are tackled by the nForce3 Ultra’s integrated Gigabit MAC, which would be great were it not for Gigabyte’s puzzling decision to pair the nForce3 Ultra’s Gigabit MAC with a 10/100 Fast Ethernet PHY. The move caps the nForce3 Ultra’s otherwise speedy integrated GigE controller at 100Mbpsan order of magnitude slower than it should be.
Handicapping the nForce3 Ultra’s integrated GigE controller is bizarre, but at least it doesn’t cripple NVIDIA’s firewall. You can read more about the firewall, which is bundled with NVIDIA’s ForceWare drivers, here.
Moving to audio, it’s NVIDIA’s turn to take a little heat. Because NVIDIA left the SoundStorm APU out of its nForce3 chipsets, a decision that drew the ire of many enthusiasts, the GA-K8NSNXP-939 is stuck without hardware-accelerated 3D audio solution. Realtek’s ALC850 codec handles digital-to-analog signal conversions, serving up eight channels of audio at resolutions up to 16 bits and sampling rates up to 48kHz.
Hardly anyone integrates Firewire capabilities into chipsets these days, so Gigabyte uses Texas Instruments’ TSB82AA2 Firewire controller to serve the board’s three Firewire ports. The TSB82AA2 joins the Silicon Image RAID controller and Marvell GigE chip on what amounts to a very crowded PCI bus.
Two chips that won’t crowd the GA-K8NSNXP-939’s PCI bus are the board’s dual BIOS chips.
DualBIOS keeps a handy backup BIOS waiting in the wings should the board’s primary BIOS fail to flash properly or otherwise become corrupted. If the primary BIOS fails a checksum on boot, the board will boot from the backup BIOS and you can fix things from there. Nifty, huh? Gigabyte also supplies software to allow users to easily flash the board’s BIOS from Windows, which will be handy for less savvy users.
What about the BIOS itself?
It’s actually pretty loaded, although you have to press Ctrl+F1 to unlock most of the tweaking options. Once you do that, you’ll be treated to an array of memory timings, a full set of AGP tweaking options, and even HyperTransport frequency dividers that should keep the HT bus in spec when overclocking.
The GA-K8NSNXP-939 supports CPU bus speeds up to a whopping 445MHz and locked AGP bus speeds between 66 and 100MHz, both in 1MHz increments. Multipliers in 0.5x increments are available, as are CPU voltages up to 1.7V in 0.025V increments. The CPU voltage options are a little disappointing considering the board’s swanky six-phase power circuit, though. A CPU voltage ceiling of 1.7V is only 0.2V above stockhardly enough headroom for extreme overclocking.
For the paranoid like me, the GA-K8NSNXP-939’s BIOS serves up warning options that can be triggered by fan failures or CPU temperatures. The BIOS doesn’t go so far as to offer temperature- or fan failure-triggered shutdown conditions, but it’s halfway there with warning alarms.
To keep noise levels to a minimum, the GA-K8NSNXP-939’s BIOS also has a smart CPU fan that ramps up RPMs as CPU temperatures rise. The smart fan has four different fan speed settings, but the temperature triggers are hard-coded into the BIOS, so you can’t fiddle with them yourself.
Finally, the GA-K8NSNXP-939’s BIOS offers an ambiguously-named “Top Performance” option, which produces the following warning:
As far as I can tell, the Top Performance setting does little more than overclock the HyperTransport bus by 8MHz. That’ll be good for a small performance boost, but Gigabyte should indicate that the setting will run a system processor out of spec. After all, running a processor at faster than stock speeds will void its warranty.
For those who are timid about tweaking motherboard settings from the BIOS, NVIDIA offers a system utility application for Windows. The system utility works hand-in-hand with the board’s BIOS, so it’s up to mobo manufacturers to expose timing, tweaking, and overclocking options to the application.
With the GA-K8NSNXP-939, Gigabyte has exposed a handful of memory timing options and access to the board’s CPU and AGP bus. That’s a good start, but for more involved fiddling, it would be nice to have control over CPU multipliers and voltages.
Our testing methods
All tests were run three times, and their results were averaged, using the following test systems.
|Processor||Athlon 64 3500+ 2.2GHz|
|Front-side bus||HT 16-bit/1GHz downstream
HT 16-bit/1GHz upstream
|Motherboard||Abit AV8||Asus A8V Deluxe||Gigabyte GA-K8NSNXP-939|
|BIOS revision||Version 12||1006||F3|
|North bridge||VIA K8T800 Pro||NVIDIA nForce3 Ultra|
|South bridge||VIA VT8237|
|Chipset drivers||Hyperion 4.53||ForceWare 4.27|
|Memory size||1GB (2 DIMMs)|
|Memory type||OCZ PC3200 EL Platinum Rev 2 DDR SDRAM at 400MHz|
|RAS to CAS delay||2|
|Hard drives||Western Digital Raptor WD360GD 37GB SATA
Maxtor DiamondMax Plus D740X 40GB ATA/133
|Graphics||ATI Radeon 9600 XT|
|Graphics driver||CATALYST 4.8|
|OS||Microsoft Windows XP Professional|
|OS updates||Service Pack 2, DirectX 9.0c|
Thanks to OCZ for providing us with memory for our testing. If you’re looking to tweak out your system to the max and maybe overclock it a little, OCZ’s RAM is definitely worth considering.
We used the following versions of our test applications:
- SiSoft Sandra Standard 2004
- ZD Media Business Winstone 2004 1.0.1
- ZD Media Multimedia Content Creation Winstone 2004 1.0.1
- TCD Labs HD Tach v3.01
- Futuremark 3DMark03 Patch 340
- DOOM 3
- Far Cry v1.2
- Unreal Tournament 2004 v3270
- RightMark Audio Analyzer 5.3
- RightMark 3D Sound 1.02
- Cinebench 2003
- Sphinx 3.3
The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at a 75Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests. Most of the 3D gaming tests used the Medium detail image quality settings, with the exception that the resolution was set to 640×480 in 32-bit color.
All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.
Although the boards are locked in a virtual tie in Sandra, the GA-K8NSNXP-939 falls behind in Cachemem’s memory bandwidth test. The board’s write bandwidth is particularly disappointing, especially since all three platforms are sharing the same on-chip Athlon 64 memory controller.
The GA-K8NSNXP-939 scores first and last in our Winstone tests, but its margin of victory in the Multimedia Content Creation test is more than double its margin of defeat in the Business test.
The GA-K8NSNXP-939 wins all but one of our gaming tests, but never by a significant margin.
Scores remain tight in Cinebench.
Sphinx speech recognition
And in Sphinx.
The GA-K8NSNXP-939’s audio performance is nothing to write home about; it’s right up there with other software-accelerated solutions.
For RightMark’s audio quality tests, I used a Terratec DMX 6fire 24/96 for recording. Analog output ports were used on all systems. To keep things simple, I’ve translated RightMark’s word-based quality scale to numbers. Higher scores reflect better audio quality, and the scale tops out at 6, which corresponds to an “Excellent” rating in RightMark.
The GA-K8NSNXP-939’s performance in our audio quality tests is middle-of-the-pack, too.
Disk controller performance
ATA performance was tested with a Maxtor 740X-6L ATA/133 hard drive using HD Tach 3.01’s 8MB zone setting.
The GA-K8NSNXP-939 scores a bit of a breakout performance in our ATA write speed tests, where it outdistances the competition by more than 2MB/sec. The board also registers the lowest CPU utilization of the lot during PATA transfers.
Moving to Serial ATA, we tested performance with a Western Digital Raptor WD360GD SATA hard drive. Again, we used HD Tach 3.01’s 8MB zone test.
The GA-K8NSNXP-939’s performance with Serial ATA drives looks pretty good, too. Both the board’s SATA controllers perform well, especially in the write speed test where the nForce3 Ultra leads the pack again. The nForce3 Ultra SATA controller also has the lowest SATA CPU utilization of the lot.
Our USB transfer speed tests were conducted with a USB 2.0/Firewire external hard drive enclosure connected to a 7200RPM Maxtor 740X-6L hard drive. We tested with HD Tach 3.01’s 8MB zone setting.
The GA-K8NSNXP-939 performs well in our USB tests, although it’s not the fastest board on the block when it comes to read speeds.
Our Firewire transfer speed tests were conducted with the same external enclosure and hard drive as our USB transfer speed tests.
Firewire is a whole other story, though. The GA-K8NSNXP-939’s Firewire transfer speeds are embarrassingly low, suggesting that something is very wrong. Tests were conducted using Windows XP’s standard Firewire drivers on all boards, and since the GA-K8NSNXP-939’s driver CD doesn’t have anything for the board’s Firewire chip, I suspect this could be a hardware problem.
We evaluated Ethernet performance using the NTttcp tool from the Microsoft’s Windows DDK. The docs say this program “provides the customer with a multi-threaded, asynchronous performance benchmark for measuring achievable data transfer rate”.
We used the following command line options on the server machine:
ntttcps -m 4,0,192.168.1.25 -a
..and the same basic thing on each of our test systems acting as clients:
ntttcpr -m 4,0,192.168.1.25 -a
Our server was a Windows XP Pro system based on Chaintech’s Zenith 9CJS motherboard with a Pentium 4 2.4GHz (800MHz front-side bus, Hyper-Threading enabled) and CSA-attached Gigabit Ethernet. A crossover CAT6 cable was used to connect the server to each system.
Although the GA-K8NSNXP-939’s PCI-bound Marvell GigE controller performs reasonably well in our Ethernet throughput test, the nForce3 Ultra’s crippled GigE controller is embarrassingly slow. 10/100 Fast Ethernet doesn’t cut it, and that stings all the more considering the nForce3 Ultra’s untapped GigE potential.
Update 6/13/2005 We recently discovered that the ntttcp CPU utilization results included in this review were incorrect. The CPU utilization results have been removed, but they didn’t factor prominently into our overall conclusion, so that remains unchanged. A full explanation can be found here.
For our overclocking tests, I swapped out my test system’s OCZ PC3200 2-2-2-5 memory in favor of some of the company’s PC4400 sticks, which are rated for 2.5-4-4-8 timings at 550MHz. These DIMMs nicely remove memory speed as a possible overclocking bottleneck. The PC4400 sticks carry higher latencies that could hinder performance at stock speeds, though. I’ve provided scores for the AV8 system running at stock speeds with 2-2-2-5 timings, 2.5-4-4-8 timings, and overclocked speeds with the 2.5-4-4-8 sticks.
In testing, I was able to get our GA-K8NSNXP-939 sample stable with a system bus speed of 245MHz. I used 9x CPU and 4x HyperTransport multipliers to isolate the motherboard rather than the CPU as the limiting variable. The board wasn’t stable in DOOM 3 or Sphinx with faster bus speeds, and in some cases, it wouldn’t even post.
As is always the case with overclocking, your mileage may vary.
Sphinx sees a big performance boost going from a system bus of 200MHz to 245MHz, at least when we look at the 2.5-4-4-8 scores. DOOM 3 appears to be indifferent to faster bus or memory speeds. Given the high frame rates and low graphics quality settings, I suspect DOOM 3 may be CPU limited in this case.
I’ve had boards like the Abit AV8 stable with bus speeds as high as 260MHz, so the GA-K8NSNXP-939 is by no means the best overclocking board I’ve had in the Benchmarking Sweatshop. Still, the board fared better than the Asus A8V Deluxe, whose initial release lacked a working AGP/PCI lock.
At first glance, the GA-K8NSNXP-939 looks like an attractive Socket 939 board. With a capable nForce3 Ultra chipset, mostly great performance, every integrated peripheral one could ask for, a well-equipped BIOS, and neat extras like DualBIOS and an external SATA adapter, there’s certainly a lot to like. However, there’s also a lot of untapped potential lurking just under the surface, potential that Gigabyte could have tapped to create a much better board.
For starters, Gigabyte’s use of a 10/100 Fast Ethernet PHY cripples the nForce3 Ultra’s integrated Gigabit Ethernet controller, reducing potential peak throughput by a factor of 10. That’s inexcusable, and PCI-bound GigE doesn’t do much to ease the pain.
A slightly less offensive instance of untapped potential is Gigabyte’s use of a PCI-bound Serial ATA controller instead of all four of the nForce3 Ultra’s Serial ATA RAID ports. Not only does the PCI-bound SATA controller have to compete with other devices for bandwidth, it also prevents users from creating RAID arrays that span four Serial ATA drives. Having two Serial ATA RAID controllers might look better on paper, but a full four-port nForce3 Ultra SATA RAID implementation would have offered superior functionality and potentially better performance.
And then there’s the board’s horrendous Firewire performance. While few may actually need or use the board’s Firewire port, I expect better from a product that sells for $209 online. That price tag is really the key. Were the GA-K8NSNXP-939 more affordable, I could excuse a poor performance here or there, or even some untapped potential. However, on a board that costs almost twice as much as the Abit AV8, the GA-K8NSNXP-939’s flaws are too glaring to ignore.