Bulldozer mobos from Asus and MSI: Sabertooth 990FX & 990FXA-GD80

Bulldozer is coming. AMD lifted the lid on its next-generation processor architecture in August of last year, and PC enthusiasts have been eagerly awaiting its arrival ever since. As one might expect from something named after a multi-ton piece of machinery pulled from the Tonka catalog, Bulldozer’s approach has been a slow one. Some nine months after AMD’s big reveal of the underlying architecture, the first Bulldozer-based Zambezi CPUs have yet to roll onto store shelves.

AMD’s new baby may require a little more gestation before being released into the wild, but Taiwanese motherboard makers seem uninterested in waiting for its official arrival. I can’t say I blame them. The Computex trade show takes over Taipei this week, giving mobo manufacturers the opportunity to show off their wares on home turf. Even if it tried, AMD probably couldn’t have stopped Bulldozer boards from lining the show floor.

Besides, this incoming wave of Socket AM3+ motherboards is perfectly capable of getting by without Zambezi-based CPUs. The AM3+ socket is backward compatible with standard Socket AM3 processors, so any old Athlon II or Phenom II will do. Impatient fanboys looking to get a head start on their Bulldozer builds will have no shortage of options, and we’ve rounded up a couple of interesting ones from Asus and MSI.

In Asus’ corner, the TUF series, of which we’ve grown fond, finally adds an AMD-compatible model in the Sabertooth 990FX. Meanwhile, MSI looks to continue its recent upward trajectory with the 990FXA-GD80. Both boards are lavished with all of the luxuries one might expect from high-end enthusiast models, such as military-class electrical components, UEFI replacements for legacy BIOSes, plentiful 6Gbps Serial ATA ports, loads of USB 3.0 connectivity, and enough PCI Express x16 slots for three-way CrossFire or SLI graphics configs. There are important differences between how Asus and MSI have approached this next-generation platform, though. We’ll explore those differences over the following pages as we see how the Sabertooth 990FX and 990FXA-GD80 stack up against the latest and greatest Sandy Bridge motherboards.

New socket, same silicon

As far as Zambezi is concerned, the most important element of these new motherboards is the AM3+ socket. Ignore the AM3b label in the picture above. Despite what component maker Lotes would have you believe, the official name for AMD’s new socket is AM3+.

At first glance, Socket AM3+ appears to be little more than an AM3 socket done up like Robert Downey Jr. in Tropic Thunder. Black is in, but it’s not the only change. If you take the time to count, you’ll notice that the AM3+ socket has 942 individual pins—one more than AM3.

Intel made the transition to an LGA socket all the way back in the Pentium 4 days, but AMD continues to put the pins on the processor rather than the motherboard. As someone who handles both components on a daily basis, I have to admit that I prefer the LGA approach. Processors tend to be more expensive than motherboards, so the cost of a bent pin is likely to be higher with a conventional socket. The fact that CPUs are much smaller and easier to mishandle also makes them more prone to accidental pin damage. To be fair, though, realigning a processor’s bent pins can be easier than performing similar surgery on an LGA socket.

Sticking with a traditional socket allows AMD to maintain its long-standing tradition of offering some degree of backward compatibility when introducing new CPUs and sockets. As a result, Socket AM3+ motherboards will happily accept existing AM3 CPUs. The picture isn’t quite as clear for folks hoping to pair upcoming AM3+ CPUs with older motherboards. Asus and MSI have announced BIOS updates for select Socket AM3 boards that promise compatibility with these next-gen processors. However, neither company has been able to confirm that those updates will allow boards to fully exploit the clock-scaling and power-saving features built into Zambezi. Asus expects everything to work properly, but it won’t know for sure until the final CPU silicon is ready. With AMD intent on keeping the details of Zambezi’s socket-specific mojo to itself until the CPU’s official unveiling, there isn’t much more we can add on the subject.

We can tell you that some motherboard makers have introduced Socket AM3+ models based on older 8-series chipsets, and that the presence of 942-pin sockets should guarantee Zambezi will properly. However, with only a + separating these boards from older AM3 models, there’s plenty of potential for confusion. To provide some clarity, AMD is rolling out 9-series chipsets that will be available exclusively on Socket AM3+ motherboards.

Led by the 990FX, this core-logic family is new in name only. The 990FX consists of the same north- and south-bridge components as its 890FX predecessor, and those chips continue to be manufactured by TSMC using a 65-nm fabrication process. AMD didn’t have a block diagram for the 990FX when we asked, but it only took a couple of minutes in Photoshop to put one together using the block diagram from our 890FX review.

My changes are not-so-subtly highlighted in pink, and there really isn’t much to see. The old 890FX block diagram shows a HyperTransport 3.0 processor link with 5.2 GT/s of peak interconnect bandwidth. With the 990FX, that link gets upgraded to version 3.1 and 6.4 GT/s of bandwidth. Because the north-bridge silicon is unchanged, the fatter pipe is likely tied to the socket rather than the chipset. I’d expect AM3+ boards with 8-series chipsets to offer the same boost in HT link speed.

Despite building blocks that are more than a year old, the 8990FX looks very well equipped. The north bridge is essentially a massive PCI Express 2.0 switch loaded with 42 lanes of connectivity. 32 of those lanes can be split evenly between a pair of x16 slots, and the resulting x16 links can be cleaved in two to enable all kinds of multi-GPU excess. Speaking of multi-GPU configs, I probably should have added a couple of SLI logos to the block diagram. After years of preventing multiple GeForce graphics cards from sharing rendering duties on AMD chipsets, Nvidia has a new graphics driver that enables SLI support on 990FX motherboards.

At the south bridge, the SB950 offers a couple of additional PCI Express lanes alongside the usual assortment of I/O blocks. All six of the Serial ATA ports boast 6Gbps connectivity, but the USB controller is so last-generation. To be fair, Intel also stuck with USB 2.0 for its Sandy Bridge chipsets, which have just two 6Gbps SATA ports. Both firms have been slow to embrace the new USB standard, even though it’s an arguably more necessary step up in performance than the transition from 3Gbps to 6Gbps SATA. Just about any external hard drive should be fast enough to saturate a USB 2.0 connection, but you’ll need a high-end SSD to saturate the bandwidth offered by a 3Gbps SATA port.

And then there were two

As we do before most major chipset launches, we pinged The Big Three motherboard makers to see what they had to offer. Only Asus and MSI were able to provide boards before my flight for Computex left on Saturday morning, so sadly, Gigabyte will have to sit this one out.

  Asus Sabertooth 990FX MSI 990FX-GD80
CPU power 8+2 8+2
Expansion slots 3 PCIe x16 (x16/x16, x16/x8/x8)

1 PCIe x16 (x4)

1 PCIe x1

1 PCI

3 PCIe x16 (x16/x16, x16/x8/x8)

1 PCIe x16 (x4)

2 PCIe x1

1 PCI

Gigabit Ethernet Realtek RTL8111E Realtek RTL8111E
Auxiliary SATA 2 x JMicron JMB362 JMicron JMB362
USB 3.0 2 x ASMedia ASM1042 2 x NEC D720200
Audio Realtek ALC892 Realtek ALC892
FireWire VIA VT6308P VIA VT6315N
Warranty length Five years Three years
Price $209 $200

At least on the surface, the Sabertooth 990FX and 990FXA-GD80 are quite similar. The boards sport the same number of power phases and PCI Express x16 slots, and they rely on identical controllers to handle audio, Gigabit Ethernet, and auxiliary Serial ATA duties. There’s some variety on the USB 3.0 front, with MSI opting for NEC’s near-ubiquitous controller, while Asus uses a new chip from subsidiary ASMedia.

One of the most notable differences between these two motherboards is the Sabertooth’s five-year warranty, which has yet to be matched by desktop boards outside Asus’ TUF series. The additional coverage is a nice bonus for folks who upgrade infrequently or migrate older hardware to secondary systems like HTPCs, closet file servers, and PCs for the rest of the family. You’ll pay a little extra for the privilege, though.

Asus’ Sabertooth 990FX motherboard

For years, it feels like motherboard makers have been compelled to differentiate high-end designs by piling on integrated peripherals and adding more extremeness to overclocking features. The thing is, modern mid-range boards offer more than enough expansion slots and ports to meet the needs of the typical enthusiast. The very same boards also tend to have enough overclocking headroom to cover air-cooled exploits, leaving most of us with little incentive to upgrade to something fancier.

Asus took note, and last year, it introduced a new family of TUF-series motherboards bearing the Sabertooth name. Designed for professionals, this line eschews the usual excesses to focus on component quality and thermal management. After watching Intel’s X58, P55, and P67 chipsets get the TUF treatment, AMD fans finally have one of their own: the Sabertooth 990FX.

Like the rest of the family, the 990FX is draped in earthy tones that wouldn’t look out of place in a camouflage print. The army aesthetic walks a fine line between being distinctive and perhaps a little bit dull, but it plays nicely with the chunky heatsinks and really suits the board’s utilitarian personality.

As you can see, Asus hasn’t bothered with the Thermal Armor cooling shroud that appeared on the Sabertooth P67. Good riddance. The armor is supposed to direct airflow to cool board-level components, but we found that it can have the exact opposite effect with typical system configurations. To help users fine-tune their own cooling solutions, the Sabertooth 990FX is equipped with no fewer than 10 temperature sensors covering everything from the CPU to the power regulation circuitry and USB 3.0 controllers.

While this slew of temperature sensors is a relatively new addition to the TUF arsenal, Sabertooth mobos have always had a penchant for exotic electrical components. Most high-end motherboards do these days. Unless you’re familiar with the intricacies of capacitor chemistry, it can be difficult to tell which motherboard is really using the best parts. Asus makes its case with a laundry list of military specifications met by the Sabertooth’s capacitors, chokes, and MOSFETs. Some of those standards are more applicable than others to life inside an enthusiast’s PC, so it’s hard to get too excited about the fact that Asus claims compliance with more military specs than its competitors. I’m far more convinced by the five-year warranty, which suggests a level of confidence in the board’s long-term reliability that’s unmatched in the industry.

The Sabertooth’s socket looks a little crowded thanks to the close proximity of VRM and north-bridge heatsinks. Fortunately, the heatsinks stand just 29 mm tall, so they shouldn’t conflict with too many aftermarket CPU coolers. You may not be as lucky with taller memory modules. The DIMM slots sit just 20 mm from the socket, and AMD’s retention bracket doesn’t allow coolers to be rotated in 90-degree increments to avoid clearance conflicts.

Drifting south gives us a good look at the Sabertooth’s internal expansion options. The light brown PCIe slots run in a dual-x16 configuration, and the dark brown slot can be added to the mix in an x16/x8/x8 setup for three-way CrossFire or SLI. With only four lanes of bandwidth, the black PCIe x16 slot can’t participate in any of that multi-GPU madness. There wouldn’t be enough room for a fourth double-wide graphics card, anyway.

To the right of the expansion slots sit two rows of Serial ATA ports. The ports don’t go right to the edge, providing a little more clearance for cramped enclosures that put the hard drive cage right next to the motherboard tray. As you might have guessed, the brown ports are connected to the chipset, while the black ones are linked to an auxiliary JMicron controller.

The JMicron controller is a 3Gbps model and nothing special. It’s one of a pair, with the second chip feeding a couple of eSATA ports in the rear cluster. One of those ports has integrated USB power, while the other must go without. You’re probably more interested in the USB 3.0 ports, though. There are two SuperSpeed ports at the rear, plus an onboard header for two more.

Like just about every other high-end motherboard, the Sabertooth has integrated audio fed by a Realtek ALC892 codec chip. Asus also relies on Realtek to provide surround-sound virtualization for stereo speakers and headphones, which can be enabled through the audio control panel. Surround virtualization is a nice little extra, but we’d rather have support for real-time Dolby Digital Live or DTS encoding. Such a feature would allow multi-channel game audio to be passed to a compatible receiver or speakers over a pristine digital connection that completely bypasses the motherboard’s DAC.

The legacy BIOS we’ve been living with for ages is being phased out in favor of UEFI. This next-gen firmware interface is a huge improvement over its predecessor, offering native support for hard drives larger than 2.2TB alongside perks like mouse input and fancy graphics. Of all the UEFI implementations we’ve seen thus far, Asus’ is easily the best. The EZ interface has some graphical flair but offers far too little control for real tweaking. However, the advanced mode, pictured above, layers just enough eye candy over a familiar layout that’s very responsive and easy to navigate. The mouse integration works well, and you can even use the wheel to scroll through the various options.

Beneath its slick GUI, the Asus firmware has all the options you’ll need to adjust clock speeds, to unlock disabled cores, to tweak memory timings, and to adjust voltages. Those features are old hat by now, but good fan speed controls are still hard to find. Asus does a good job on that front, serving up easily configurable, temperature-based fan speed controls for the processor and system fan headers.

If you’d rather tune your system from Windows, Asus’ AI Suite software has you covered. Of particular interest is the Thermal Radar app, which is unique to the Sabertooth family and adds and extra layer of temperature monitoring and fan speed control. My old-school sensibilities still prefer such things to be handled independent of the operating system, but it’s nice to be able to tweak various settings without rebooting into the UEFI.

MSI’s 990FXA-GD80 motherboard

Judged on specifications alone, MSI’s 990FXA-GD80 looks a lot like the Sabertooth 990FX. That’s what happens when you’ve got two boards based on the same core-logic chipset drawing auxiliary peripherals from an increasingly shallow pool of options. Spend a little time with the GD80, though, and you’ll discover that it’s an altogether different animal in many ways.

The GD80 obviously has a different look, but only when it’s compared to the Sabertooth. No fewer than four different motherboard makers are using similar black-and-blue color schemes on their enthusiast-oriented models, and the sheer sameness of it all is enough to make me long for the days when each mobo maker drew from a garish—but unique—palette of neon hues. To MSI’s credit, all the colors match, the heatsinks follow a cohesive design language, and nothing stands out as an eyesore. That’s more than can be said of some of the boards riding the black-and-blue bandwagon.

Speaking of bandwagons, the GD80 has all the swanky electrical components you’d expect from a high-end mobo. MSI claims to have been somewhat of a trendsetter in this arena, having been the first to use driver MOSFETs, digital VRMs, and tantalum-core capacitors. I don’t remember seeing much talk about military specifications before Military Class logos started showing up on the company’s boards, either.

Perhaps I’m getting jaded in my old age, but all this talk about superior component quality rings a little hollow without something to back it up. Claims of improved power delivery and efficiency are difficult to test at the component level, and so are promises of better stability when overclocking. The fact that MSI covers the GD80 with the same three-year warranty as lesser models suggests the high-end board is no more reliable than standard fare.

After measuring clearances for our last few motherboard reviews, I’ve noticed something interesting. On the boards in our Z68 Express round-up, the DIMM slots sit between 25.5 and 30 mm from the socket. With the two 990FX models we’re looking at today, that gap shrinks to only 20-22 mm. The tighter spacing shouldn’t be a problem for standard-height DIMMs, but taller modules may not be so lucky if you’ve got a heatsink tower that branches out toward them.

To MSI’s credit, there’s more room around the GD80’s socket than there is around the Sabertooth’s. The north-bridge cooler is barely taller than some of the surface-mounted components, so it definitely won’t get in the way. I do have some reservations about how the low-profile cooler might handle the heat piped over by the VRM heatsink. The Sabertooth board has a similar heatpipe, but its north-bridge heatsink has a lot more surface area to aid heat dissipation.

With a second PCI Express x1 slot squeezed into the top of the stack, the GD80 offers one more expansion slot than the Sabertooth. The lane assignments for the PCIe x16 slots are pretty much identical, though. The first and third x16 slots can pull off a full-bandwidth dually, while the top three can be arranged in an x16/x8/x8 config. As with the Asus board, the fourth x16 slot is limited to four lanes of electrical connectivity.

The GD80’s SATA ports are pushed right up to the edge of the board along with the front-panel USB 3.0 connector. Given how popular edge-mounted SATA ports have become in recent years, I hope case makers design their drive cages accordingly. You won’t want longer graphics cards bumping up against a drive cage, either.

Before moving on, I’ve gotta give MSI props for including a two-digit POST code display to aid troubleshooting. I’m also going to add a shout out for the CMOS reset button that’s been integrated into the port cluster.

The cluster looks pretty crowded, and it’s nice to get USB power for both the eSATA ports. Surprisingly, there are only four USB 2.0 ports at the rear. Internal headers add another four ports, bringing the grand total to just eight. Even with two ports consumed by the hybrid eSATA connectors, four of the SB950’s USB ports are left unused. There’s room in the cluster for more ports and space on the board for additional headers, making the decision to skimp even more puzzling.

At least MSI brings more to the table on the audio front. In addition to a second S/PDIF output, the board will ship with THX TrueStudio PRO software that provides surround-sound virtualization and a smattering of other effects. We haven’t spent enough time with the various speaker virtualization schemes to know how the THX software compares to what’s available in Realtek’s drivers, but you have the option of using either with the GD80. Alas, neither of those solutions offers real-time Dolby Digital Live or DTS encoding.

Believe it or not, the GD80 has a UEFI rather than an old-school BIOS. The only thing that gives it away is the mouse cursor hovering near the center of the picture above. The GUI we saw on MSI’s first P67 motherboard has been scrapped, and the company is working on a new graphical interface to take its place. In the meantime, the GD80 is stuck with one of the most frustrating firmware interfaces I’ve used so far.

Believe it or not, the problems have nothing to do with how the UEFI looks and everything to do with how it reacts. Let’s start with the mouse cursor, which flickers constantly and moves with less urgency than the tail end of a pack of mall-walking senior citizens. Even the keyboard response is sluggish, and it takes a few seconds to trigger rapid scrolling when holding down an arrow key. That’s a bigger problem than one might expect, because multipliers and voltages must be selected from long lists of available options—they can’t be keyed in directly. MSI’s old-school BIOSes were much more responsive and easier to use.

The GD80’s UEFI might be a pain to interact with, but at least it’s loaded with all the overclocking and tweaking options one would expect from an AMD motherboard targeted at PC enthusiasts. Like the Sabertooth’s UEFI, there’s a core unlocker, an integrated flashing utility, and support for multiple configuration profiles. MSI’s fan speed controls could use some attention, though. Users can set a target temperature and a minimum fan speed for the CPU fan, but there’s no temperature-based control for the system headers, which can only be toggled between three static speed settings.

MSI’s Control Center software doesn’t improve on those fan speed controls, but it does bring most of the tweaking and overclocking options into Windows, where your mouse will actually work properly. I’d be curious to see some statistics on the number of enthusiasts who prefer tweaking with a BIOS or UEFI as opposed to using Windows software. Perhaps that would explain why some motherboard makers seem to have devoted far more resources to developing Windows software than transitioning to UEFI.

Digging into the details

Warning: unless you’re intently curious about specific UEFI settings, port counts, or details surrounding the systems we used for testing, you should probably skip ahead to our performance results. This page is littered with overstuffed tables that hardly make for engaging reading, but we know that some folks eat this stuff up. Weirdos.

  Asus Sabertooth 990FX MSI 990FX-GD80
Clock speeds Base: 100-600MHz

DRAM: 800-1600MHz

PCIe: 100-150MHz

NB: 1400-3200MHz

HT: 800-2000MHz

Base: 190-690MHz

DRAM: 800-1600MHz

PCIe: 90-190MHz

Multipliers CPU: 4-35X

Turbo: 4-35X

CPU: 4-32.5X

Turbo: 4-32.5X

CPU NB: 4-20X

HT: 1-13X

Voltages CPU: 0-2.1V

CPU NB: 0-1.875V

DRAM: 1.2-2.5V

VDDA: 2.2-2.8V

NB: 0.8-1.5125V

NB HT: 0.8-1.5125V

NB 1.8: 1.8-2.8V

SB: 1.1-1.8V

PCIe: 1.1-2.2V

VDDR: 1.2-1.8V

CPU: 1.014-2.0115V

CPU NB: 0.796934-1.8451V

CPU PLL: 2.04765-3.127993V

DRAM: 1.442-1.906V

NB: 0.964-1.303V

SB: 0.6245-1.5755V

HT: 0.9636-1.157233V

DDR VREF: 0.519034-1.164266V

DDR PHY: 0.868667-1.617480V

NB PCIe: 1.1316-2.09536V

DDR VTT: 0.509334-1.166266

Fan control Temp-based CPU, system Temp-based CPU

Manual system 1-4

Glancing at the chart above might give you the impression that the Sabertooth 990FX and 990FXA-GD80 have comparable UEFI implementations. A similar mix of options might be available with each board, but the UEFIs couldn’t be more different otherwise.

  Asus Sabertooth 990FX MSI 990FX-GD80
CPU power 8+2 8+2
DIMM slots 4 DDR3-1333 4 DDR3-1333
Expansion slots 3 PCIe x16 (x16/x16, x16/x8/x8)

1 PCIe x16 (x4)

1 PCIe x1

1 PCI

3 PCIe x16 (x16/x16, x16/x8/x8)

1 PCIe x16 (x4)

2 PCIe x1

1 PCI

Storage I/O 6 6Gbps SATA RAID

2 3Gbps SATA

6 6Gbps SATA RAID
Audio 8-channel HD 8-channel HD
Ports 1 PS/2 keyboard/mouse
2 USB 3.0 w/ 2 headers
10 USB 2.0 w/ 4 headers

1 eSATA/USB
1 eSATA
1 FireWire w/ 1 header
1 RJ45

1 analog front out

1 analog bass/center out

1 analog rear out

1 analog surround out

1 analog line in

1 analog mic in

1 optical S/PDIF out

1 PS/2 keyboard

1 PS/2 mouse

2 USB 3.0
4 USB 2.0 w/ 4 headers

2 eSATA/USB

1 FireWire w/ 1 header

1 RJ45

1 analog front out

1 analog bass/center out

1 analog rear out

1 analog surround out

1 analog line in

1 analog mic in

1 optical S/PDIF out

1 coaxial S/PDIF out

If you paid attention over the first few pages, the spec sheet will contain no surprises.

Our testing methods

The 990FX boards were waiting on UEFI updates until the middle of last week, so there wasn’t much time for testing before I had to be on a plane to Computex. Nevertheless, I managed to run the Sabertooth 990FX and 990FXA-GD80 through our full motherboard suite, allowing us to compare their performance to that of the collection of mobos we rounded up for our recent Z68 coverage. The Intel boards were tested with a Core i7-2600K that costs quite a bit more than the Phenom II X6 1090T, so the results won’t be directly comparable. All the Intel boards have been greyed out in the graphs to avoid confusion.

Sandy Bridge motherboards typically crank the Core i7-2600K’s multiplier up to its 38X Turbo peak with one-core loads. When all four physical cores are busy, the Turbo multiplier is generally capped at 35X as long as the motherboard can keep the CPU supplied with sufficient power. The MSI and Gigabyte Z68 boards follow this behavior, and so did Asus’ P8P67 PRO back in January.

With its latest BIOS, however, the P8P67 PRO uses a 38X Turbo multiplier when under four-core loads. The same is true for the Sabertooth P67 and the P8Z68-V PRO. This only happens if you set the memory multiplier manually, which we do with all our test configurations. Leave the memory multiplier on auto, and the max Turbo multiplier for four-core loads will stay at 35X.

Asus has confirmed this behavior, but the company has yet to explain why it’s changing the Turbo multiplier when users adjust the memory multiplier. The two aren’t connected, and other boards allow adjustment of one without changing the other. As a result, our performance results are somewhat tainted. They are representative of what end users will see with these boards if manual memory multipliers are used, though.

We tested the Z68 boards with the Sandy Bridge IGP serving as the primary display adapter and the Radeon HD 5870 sitting behind Lucid’s Virtu software. We used the following system setups for testing. With few exceptions, all tests were run at least three times, and we reported the median of the scores produced.

Processor AMD Phenom II X6 1090T 3.2GHz Intel Core i7-2600K 3.4GHz
Motherboard Asus Sabertooth 990FX MSI 990FXA-GD80 Asus P8P67 PRO Asus Sabertooth P67 Asus P8Z68-V PRO Gigabyte Z68X-UD3H-B3 MSI Z68A-GD80
Bios revision 0138 E7640AMS.B0I 1502 1502 8801 F2d E7672IMS V17.0B17
Platform hub AMD 990FX/SB950 AMD 990FX/SB950 Intel P67 Express Intel P67 Express Intel Z68 Express Intel Z68 Express Intel Z68 Express
Chipset drivers Catalyst 11.5 Catalyst 11.5 Chipset: 9.2.0.1025

RST: 10.1

Chipset: 9.2.0.1025

RST: 10.1

Chipset: 9.2.0.1025

RST: 10.5

Chipset: 9.2.0.1025

RST: 10.5

Chipset: 9.2.0.1025

RST: 10.5

Memory size 8GB (2 DIMMs) 8GB (2 DIMMs) 8GB (2 DIMMs) 8GB (2 DIMMs) 8GB (2 DIMMs) 8GB (2 DIMMs) 8GB (2 DIMMs)
Memory type Corsair Vengeance DDR3 SDRAM at 1333MHz Corsair Vengeance DDR3 SDRAM at 1333MHz Corsair Vengeance DDR3 SDRAM at 1333MHz Corsair Vengeance DDR3 SDRAM at 1333MHz Corsair Vengeance DDR3 SDRAM at 1333MHz Corsair Vengeance DDR3 SDRAM at 1333MHz Corsair Vengeance DDR3 SDRAM at 1333MHz
Memory timings 9-9-9-24-1T 9-9-9-24-1T 9-9-9-24-1T 9-9-9-24-1T 9-9-9-24-1T 9-9-9-24-1T 9-9-9-24-1T
Audio Realtek ALC892 with 2.59 drivers Realtek ALC892 with 2.59 drivers Realtek ALC892 with 2.59 drivers Realtek ALC892 with 2.59 drivers Realtek ALC892 with 2.59 drivers Realtek ALC892 with 2.59 drivers Realtek ALC892 with 2.59 drivers
Graphics Asus EAH5870 1GB with Catalyst 11.3 drivers
Hard drive Raptor WD1500ADFD 150GB
Power Supply PC Power & Cooling Silencer 750W
OS Microsoft Windows 7 Ultimate x64

We’d like to thank Asus, Corsair, PC Power & Cooling, and Western Digital for helping to outfit our test rigs with some of the finest hardware available. Thanks to each of the motherboard makers for supplying their boards, too, and to Intel and AMD for providing the CPUs.

We used the following versions of our test applications:

The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at a 60Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests.

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.

SLI support… on an AMD chipset

Before getting into our usual mix of motherboard tests, let’s indulge in a bit of a tangent. SLI support is one of the few new wrinkles associated with the 990FX chipset, and it’s kind of a big deal because Nvidia has always barred SLI graphics configurations from working with AMD chipsets. In the beginning, Nvidia restricted SLI support to its own core-logic offerings. As nForce chipsets faded into obscurity, select Intel motherboards were graced with SLI certification. Now, with Zambezi approaching, Nvidia has gone all the way over to the dark side and enabled SLI support on the 990FX.

The 990FX is a particularly interesting candidate for SLI, because it has enough PCI Express 2.0 lanes to provide a pair of graphics cards with matching full-bandwidth x16 slots. Even the most expensive Sandy Bridge motherboards are hampered by the CPU’s limited PCIe connectivity, which at best can offer eight lanes of bandwidth per card. The obvious question, then, is whether the additional bandwidth provided by the 990FX’s dual-x16 config makes a difference.

Nvidia didn’t come up with a working SLI driver until Thursday evening, leaving far too little time to reach a definitive conclusion on the subject. However, we did manage to run a few quick tests using a pair of GeForce GTX 560 graphics cards provided by MSI. The N560GTX Twin Frozr II has 870MHz core and 1020MHz (4.08 GT/s) memory clocks, putting it roughly in the middle of the clock-speed range offered by GTX 560s currently on the market. If the card looks familiar, that’s because we reviewed it just a couple of weeks ago.

You’ll be hard-pressed to find a modern game that won’t run smoothly on just one of these cards with all the eye candy turned up at a 1080p resolution. MSI has done a good job with the Twin Frozr II cooler, too. The dual-fan unit packs a mass of cooling fins and four heatpipes, allowing the card to maintain low GPU temperatures without making a racket. If you want one (or two) of your own, Newegg has ’em for $217.

In an attempt to give the 990FX a chance to flex its ample PCIe bandwidth, we’ll start with a couple of tests designed to bombard the cards with gobs of geometry data. The geometry-intensive Unigine Heaven and TessMark benchmarks were run with their tessellation detail turned all the way up at a relatively modest 1080p display resolution. Performance was tested on the 990FXA-GD80 and Asus’ Z68-based P8Z68-V PRO motherboard using Nvidia’s 275.36 graphics drivers. The Z68 board was equipped with a Core i5-2500K processor that’s a little more comparable to the Phenom II X6 1090T we used with the 990FX.

Neither benchmark shows a preference for the 990FX’s greater PCI Express bandwidth. We’ve lopped off the decimal places to simplify the graphs, so don’t worry about the fact that bars with slightly different lengths have the same score.

Next, let’s tackle some real games. My original intent was to test with a 3 x 1080p Surround Gaming config spread across the monitors sitting on my desktop. One of the screens has a 1080p panel, while the other two scale up to 1920×1200. I’ve configured these displays in a widescreen trio using AMD’s Eyefinity scheme, but Surround Gaming wasn’t nearly as cooperative. While the screens were detected by Nvidia’s drivers, enabling Surround Gaming caused all kinds of problems, from blanking screens to full-on system crashes. With the clock ticking and my will already broken by a couple of days of frantic testing, I bailed on the triple-display config and settled for a handful of gaming tests on a single 1080p monitor. Totally lame, I know.

Interestingly, it looks like the 990FX board’s Phenom II CPU is the limiting factor in the Lost Planet 2 benchmark and Civilization V‘s late-game-view test. Sandy Bridge’s gaming prowess is no secret, and the 990FX’s additional PCIe bandwidth can’t make up the difference. In the other tests, frame rates are all but identical.

I wouldn’t be surprised if some games, benchmarks, or specific graphics configurations could benefit from the additional PCIe lanes offered by the 990FX. However, we haven’t encountered much difference between the performance of dual-x8 and dual-x16 graphics configurations when we’ve tested them in the past, leaving me a little dubious that things have changed with this latest generation of games and hardware.

Memory performance

All the boards used the same memory modules running at 1333MHz with identical 9-9-9-24-1T timings. Any differences in memory subsystem performance come down to the memory controllers on the CPUs and how each motherboard maker has chosen to tune them.

The Sabertooth 990FX comes out a little bit ahead of the 990FXA-GD80 in our memory bandwidth and latency tests, but the results are really very close. Well, they are if you ignore the Sandy Bridge systems, which squeeze more bandwidth and lower latency from the very same DIMMs.

Application performance

Score a few more for the Sabertooth. Asus’ spin on the 990FX is marginally faster than MSI’s through most of our application tests.

The Z68 boards are only slower in Metro 2033 because they were run with Lucid’s Virtu software hiding the discrete graphics card behind a virtualization layer that renders game-specific optimizations useless. Lucid is working on an alternative d-Mode that virtualizes the Sandy Bridge IGP instead, but it’s not quite polished enough for mass consumption.

Power consumption

We measured system power consumption, sans monitor and speakers, at the wall outlet using a Watts Up Pro power meter. Readings were taken at idle and under a load consisting of a Cinebench 11.5 render alongside the rthdribl HDR lighting demo. We tested with Windows 7’s High Performance and Balanced power plans.

The Asus and MSI mobos have handy BIOS switches (dubbed EPU and APS, respectively) that enable advanced power-saving features. Gigabyte has similar power-saving mojo, but it requires a separate Windows application (DES). We tested each board with its extra power-saving capabilities enabled and disabled.

The 990FX systems draw more power than our collection of Sandy Bridge competitors both at idle and under load. Between the two AMD mobos, the Sabertooth has a substantial power-efficiency advantage. The Asus consumes about 20W less than the MSI at idle, and the delta between the two is even larger under load. Switch both into high-performance mode, and the GD80 really starts sucking power at idle. So much for the supposedly superior power efficiency of MSI’s fancy-pants electrical components.

Interestingly, the APS and EPU power-saving features have a similar effect on the power drawn by the 990FX systems. In both cases, power consumption falls by modest but measurable margins at idle and under load.

Overclocking

With the Sabertooth 990FX and 990FX-GD80, processors can be overclocked by raising the motherboard’s base clock speed, increasing the multiplier on unlocked Black Edition CPUs, or combining both approaches. Alternatively, you can leave the whole process in the hands of the auto-overclocking intelligence Asus and MSI have imparted on their respective boards. To keep things focused on the motherboards, we let each one try its hand at overclocking our Phenom II X6 1090T. Next, we dialed back the CPU and memory multipliers to see how high the motherboard’s base clock would go with the processor and DIMMs taken out of the equation.

After hitting the handy OC Genie button on the 990FX-GD80, the board booted up at 3.44GHz with a 16X CPU multiplier and a 215MHz base clock speed. CPU-Z listed a CPU voltage of 1.42V, and the system didn’t so much as hiccup under a six-way Prime95 load combined with the rthdribl HDR lighting demo. Of course, it was running only 240MHz faster than stock.

Our manual run at the base clock squeezed 255MHz from the board without any extra voltage. CPU-Z reported a CPU voltage of 1.35V this time around, but no amount of additional juice would coax the GD80 to POST with a 260MHz base clock.

The Sabertooth’s built-in overclocker proved to be much more aggressive than the GD80’s. This so-called OC Tuner settled on a 232MHz base clock speed that pushed the CPU to over 3.7GHz with its default multiplier. According to CPU-Z, the processor needed just 1.37V to maintain this speed, which it did with the same stress test we used on the MSI board.

When we took matters into our own hands, the base clock sailed up to 280MHz with ease. We didn’t have to touch the CPU voltage, which CPU-Z reported as 1.3V. The system made it all the way into Windows with a 285MHz base clock. However, Prime95 consistently spit out errors on at least one of the CPU cores, and bumping various voltages didn’t alleviate the issue.

Motherboard peripheral performance

The last stop on today’s journey brings us to the exciting world of peripheral performance. I hope you saved room for more tables.

  HD Tach USB 3.0 performance
  Read burst

speed (MB/s)

Average read

speed (MB/s)

Average write

speed (MB/s)

CPU utilization

(%)

Asus P8P67 PRO 220.9 176.6 57.9 2.0
Asus Sabertooth P67 221.2 177.0 58.3 2.0
Asus P8Z68-V PRO 198.0 173.0 61.6 2.0
Gigabyte Z68X-UD3H-B3 167.3 166.1 62.9 3.0
MSI Z68A-GD80 174.1 161.5 55.1 2.0
Asus Sabertooth 990FX 187.7 159.5 62.1 5.0
MSI 990FX-GD80 188.1 155.7 56.5 7.0

With the Super Talent thumb drive we used for testing, the Sabertooth 990FX offered faster USB 3.0 read and write speeds than the GD80. The Asus board’s CPU utilization is also lower, completing the sweep.

  HD Tach USB 2.0 performance
  Read burst

speed (MB/s)

Average read

speed (MB/s)

Average write

speed (MB/s)

CPU utilization

(%)

Asus P8P67 PRO 35.1 35.0 25.2 2.0
Asus Sabertooth P67 35.1 35.0 25.2 2.0
Asus P8Z68-V PRO 36.4 34.2 24.1 2.0
Gigabyte Z68X-UD3H-B3 37.5 34.8 24.8 2.0
MSI Z68A-GD80 36.3 34.9 23.3 1.0
Asus Sabertooth 990FX 35.1 30.9 25.8 4.0
MSI 990FX-GD80 31.1 30.8 24.1 7.0

Transfer rates are much slower when we switch to a USB 2.0 port, but the Sabertooth maintains its slim advantage over the GD80. Once again, the Sabertooth complements better performance with lower CPU utilization.

  HD Tune Serial ATA performance – VelociRaptor
  Read Write
  Burst (MB/s) Average (MB/s) Random 4KB (ms) Burst (MB/s) Average (MB/s) Random 4KB (ms)
Asus P8P67 PRO 292.1 129.9 7.0 292.3 125.9 2.7
Asus P8P67 PRO (Marvell) 235.6 129.9 7.2 238.9 114.8 2.6
Asus Sabertooth P67 294.2 129.9 7.0 294.1 125.8 2.7
Asus Sabertooth P67 (Marvell) 235.4 129.9 7.2 234.4 127.2 2.7
Asus P8Z68-V PRO 288.3 129.6 7.2 280.1 122.9 2.7
Asus P8Z68-V PRO (Marvell) 203.3 129.7 7.2 203.9 123.3 2.6
Gigabyte Z68X-UD3H-B3 276.0 129.6 7.2 284.0 123.8 2.7
Gigabyte Z68X-UD3H-B3 (GSATA) 177.4 129.8 7.2 178.8 121.7 2.5
MSI Z68A-GD80 234.7 129.8 7.2 264.3 123.6 2.6
MSI Z68A-GD80 (Marvell) 195.0 129.0 7.2 197.5 85.5 2.6
Asus Sabertooth 990FX 275.0 129.9 7.2 264.6 125.5 2.9
Asus Sabertooth 990FX (JMicron) 132.3 117.9 7.1 115.6 89.7 2.6
MSI 990FX-GD80 247.6 129.9 7.2 247.4 123.6 2.6

Score another one for the Sabertooth. The Asus 990FX board transfers data to and from our VelociRaptor hard drive a little bit faster than the MSI. Do your best to avoid the Sabertooth’s auxiliary SATA ports, though. The JMicron controller’s burst performance harkens back to the days of 1.5Gbps SATA interfaces, and sustained transfer rates are slower with both reads and writes.

  HD Tune Serial ATA performance – Vertex 3
  Read Write
  Burst (MB/s) Average (MB/s) Random 4KB (ms) Burst (MB/s) Average (MB/s) Random 4KB (ms)
Asus P8P67 PRO 387.8 383.1 0.05 348.1 279.6 0.06
Asus P8P67 PRO (Marvell) 263.0 261.3 0.07 241.8 130.6 0.09
Asus Sabertooth P67 388.7 383.7 0.07 346.4 278.5 0.07
Asus Sabertooth P67 (Marvell) 261.3 258.8 0.08 238.1 167.9 0.10
Asus P8Z68-V PRO 381.3 375.0 0.06 340.3 252.0 0.06
Asus P8Z68-V PRO (Marvell) 232.2 243.5 0.09 210.5 152.5 0.11
Gigabyte Z68X-UD3H-B3 383.9 378.5 0.06 325.5 210.3 0.06
Gigabyte Z68X-UD3H-B3 (GSATA) 193.9 194.5 0.06 175.2 136.5 0.08
MSI Z68A-GD80 367.2 362.1 0.06 321.9 234.1 0.07
MSI Z68A-GD80 (Marvell) 208.4 213.5 0.11 200.8 97.9 0.14
Asus Sabertooth 990FX 369.6 339.5 0.05 305.0 238.9 0.05
Asus Sabertooth 990FX (JMicron) 129.0 130.9 0.08 127.1 93.8 0.14
MSI 990FX-GD80 316.3 318.1 0.07 274.1 235.5 0.08

The JMicron controller looks even worse when paired with a cutting-edge Vertex 3 SSD. Thankfully, the Sabertooth 990FX’s 6Gbps SATA ports are plenty quick. The GD80 has to settle for second place once more, and that’s without taking the Intel boards into account.

I don’t want to make too much of the burst results because dynamic clock scaling can have a big impact on performance in that kind of test. However, it’s worth noting the 990FX boards have slower sustained read and write speeds than all of the Sandy Bridge mobos. Intel’s 6-series chipsets may have only two 6Gbps SATA ports, but those ports appear to be faster than the six AMD stuffed into the SB950.

  NTttcp Ethernet performance
  Throughput (Mbps) CPU utilization (%)
Asus P8P67 PRO 934.6 1.8
Asus Sabertooth P67 938.3 1.8
Asus P8Z68-V PRO 940.6 1.9
Gigabyte Z68X-UD3H-B3 944.5 3.9
MSI Z68A-GD80 (1) 943.9 3.7
MSI Z68A-GD80 (2) 937.1 3.4
Asus Sabertooth 990FX 947.7 9.8
MSI 990FX-GD80 939.7 9.9

The 990FX boards both use the same Realtek Gigabit Ethernet controller, so it’s no surprise to see similar throughput and CPU utilization numbers.

  RightMark Audio Analyzer audio quality
  Frequency response Noise level Dynamic range THD THD + Noise IMD + Noise Stereo Crosstalk IMD at 10kHz Overall score
Asus P8P67 PRO 5 4 4 5 3 5 5 5 4
Asus Sabertooth P67 5 4 4 5 3 5 5 5 4
Asus P8Z68-V PRO 5 4 5 3 5 5 5 5 4
Gigabyte Z68X-UD3H-B3 5 5 5 5 3 5 5 5 5
MSI Z68A-GD80 5 4 4 5 3 5 5 5 4
Asus Sabertooth 990FX 5 4 4 5 3 4 5 4 4
MSI 990FX-GD80 4 4 4 5 3 4 5 5 4

Although they post different scores in a few of RightMark Audio Analyzer’s tests of analog signal quality, the 990FX boards both end up with the same overall result. With everyone using the same Realtek audio codec, I’m not surprised. If you’re serious enough about audio quality to be looking at this chart, do yourself a favor and buy a real sound card. Your ears will thank us, and with the excellent Xonar DG available for only $35, so will your wallet.

Conclusions

In some ways, the Sabertooth 990FX and 990FXA-GD80 are very similar boards. But there are also stark differences between them, making it easy to pick a favorite. If you’ve been paying attention over the preceding pages, you probably have a pretty good idea of which board we prefer. Here’s a hint: it’s not the GD80.

Although MSI’s first take on the 990FX has a solid feature set, competitive performance, and thoughtful little touches like the debug display and cluster-mounted CMOS reset switch, several problems are particularly glaring. The first is the UEFI, whose flaky mouse support and cumbersome menus make navigation and tweaking more difficult than with the company’s old BIOSes. Then there’s the board’s power consumption, which is considerably higher than that of the Sabertooth. Denying users access to some of the USB 2.0 ports available in the chipset doesn’t make much sense to me, either.

Don’t get me wrong—the 990FXA-GD80 isn’t a bad board. It’s just not as good as the Sabertooth, which has the best UEFI implementation around and slightly better performance overall. The Sabertooth is also loaded with temperature sensors and flexible fan speed controls, making it a much better option for folks looking to strike the perfect balance between strong cooling performance and low noise levels. Throw in two more years of warranty coverage, and the $10 premium is well justified.

If you’re in the market for an AMD motherboard and have your eye on a Bulldozer upgrade down the road, the Sabertooth 990FX looks pretty sweet. However, I’m having trouble getting excited about the prospect of actually buying one. Maybe it’s the fact that we don’t yet know when the first Bulldozer-based Zambezi CPUs will arrive or how they’ll perform. There’s also a slight twinge of disappointment knowing that AMD’s new hotness will be paired with an older chipset that lacks native USB 3.0 support and can’t quite keep up with the 6Gbps SATA performance of Sandy Bridge motherboards. Should Zambezi turn out to be as impressive as many of us hope, the Sabertooth 990FX looks to be the perfect sidekick. I just wouldn’t pull the trigger on any Socket AM3+ board while we’re still in the dark on what to expect from CPUs destined for that socket.

Comments closed
    • Arclight
    • 8 years ago

    The design looks pretty solid….it will all come down to the performance of the CPUs so we’re still in the dark on that department. Also i heared the release of Bulldozer was delayed…..One thing i like about Intel mobos above AMD mobos is the installation of the CPU and CPU cooler. Intel’s design seems better on that front and i thought AMD would have improved things with this new CPUs, but it seems the mechanism remains the same.

    Also something i noticed: they moved the installation arrow for the CPU from the socket itself to the PCB of the mobo, on this board being painted with white colour.

    • Bensam123
    • 8 years ago

    “I hope you saved room for more tables.”

    Such a kidder.

    Sort of odd why the Asus board doesn’t put the extra USB 3 ports on the backport instead of on a header… You’d think people would want to use those over USB 2 ports…

    • derFunkenstein
    • 8 years ago

    Good gravy – if you’re going to put firewire on a motherboard, at least spring for the Texas Instruments chipsets. The VIA chipsets have latency issues with alot of firewire audio hardware, and that probably translates into less-than-idea performance with other firewire devices.

    • Duck
    • 8 years ago

    42 PCIe lanes? Meh. Any with less than that? 16 will do from the northbridge. mATX would be better too. And 6x 6Gbps SATA ports? Meh, rather have 2 of Intel’s ports.

      • JustAnEngineer
      • 8 years ago

      990FX/890FX has more PCIe lanes than 870 has.

      • Kurotetsu
      • 8 years ago

      [quote=”Duck”<]Meh, I'd rather have a 990GX/890GX motherboard.[/quote<] Fixed that for you.

    • Prion
    • 8 years ago

    The Supermicro P5MMS (complete with an AWESOME 512kB of cache right on the board) that I built an AMD K6-based system on back in 1997 had a GUI interface for the BIOS complete with mouse control and I’m pretty sure it wasn’t UEFI.

      • Palek
      • 8 years ago

      Now that you mention it, I’m pretty sure one of my PCs waaay back also had a GUI BIOS. Cannot remember what CPU & chipset it was, though.

        • ew
        • 8 years ago

        I also had a Supermicro board like this. If I recall correctly the mouse pointer was even animated!

      • Starfalcon
      • 8 years ago

      I have an old compaq 486 system with a GUI mouse driven bios also, although the actual GUI part is stored on the hard drive in a special hidden partition…so if the hard drive dies the board will be hosed. Fortunately the GUI has a nice utility you can save on a floppy to set up a partition on another hard drive. Although it would suck to find that out after a hard drive failure…joys of vintage computers.

      • ronch
      • 8 years ago

      I think I also saw one of these BIOSes in one of our old systems. Not sure when, but I’ve seen it, maybe in a classic Pentium or a Cyrix 6x86MX or something. It’s like, they already did it, nobody took notice, they scrapped it, and now they’re bringing it back, with extra bells and whistles, of course.

      In the end, it’s up to Award/Phenix and AMI (and Insyde?), since they’re the main BIOS providers for mobo makers.

    • WaltC
    • 8 years ago

    Enjoyed the article, but did think of a few nitpicks to mention…

    [quote<]The Intel boards were tested with a Core i7-2600K that costs quite a bit more than the Phenom II X6 1090T, so the results won't be directly comparable. All the Intel boards have been greyed out in the graphs to avoid confusion.[/quote<] In that case, wouldn't the logical solution have been to simply [i<]leave out[/i<] all but the least expensive of those other systems? I see no value in including all of them with no hint as to how the economics break down. It would be nice to know how many 1080T's you could buy for the cost of a single i7 2600K, though....;) I mean, I really don't see the point in saying that none of the Sandy Bridge systems are directly comparable to the 1080T you used in this [b<]Bulldozert[/b<] motherboard--and then doing nothing except [i<]making a direct comparison[/i<] of them for the length of the article. I realize the thrust of the article was core-logic chip set performance, but as I point out below, often these things are somewhat indivisible from the performance of the cpu installed into the system. [quote<]Intel made the transition to an LGA socket all the way back in the Pentium 4 days...To be fair, though, realigning a processor's bent pins can be easier than performing similar surgery on an LGA socket.[/quote<] Certainly true, and it is one hell of a lot easier to change out a mauled cpu than to have to change out a mauled motherboard--at least, *I* think so. (Obviously, I see LGA as the inferior technology and particularly applicable to the P4 era. But then I also recall Intel's whiz-bang cpu-slot-for-cpus-on-cards ideas, too...;) ) Although you didn't specifically mention it, do the Sandy Bridge chip set/mobo combos you tested allow for the installation of a pre-Sandy Bridge cpus as AMD's chip set/socket scheme allows here? (No? Hmmmmm....That strikes me as a distinct AMD advantage you neglected to point out for some reason--you mentioned it was possible with the 990FX--you just failed to mention that the same latitude of utility wasn't possible with Sandy Bridge core logic/systems.) I mean, while we are at it, let's just briefly review what "cpu dependency" means. It doesn't mean "nothing" in terms of device or drive performance. To the degree that a device is cpu dependent, say a SATA or other kind of drive, even a 3d card's drivers, a faster cpu will clock faster performance on that device. You might have mentioned that, as well. And ironically, the more cpu-dependent a given device or drive is, the more the processing speed of the installed cpu will affect its performance numbers. Even a 2% cpu dependency benchmark number will materially benefit from a faster cpu. But a 7% cpu dependency, say, will *really* benefit from a faster cpu as well as be more penalized in performance when running with a slower cpu. You also might have mentioned a bit more about the cpu memory controllers employed here as well, as the Sandy Bridge mem controller is clearly *expected* to be better than both AMD's pre-Bulldozer controllers and Intel's pre-Sandy Bridge controllers, too. [quote<]The Sabertooth 990FX comes out a little bit ahead of the 990FXA-GD80 in our memory bandwidth and latency tests, but the results are really very close. Well, they are if you ignore the Sandy Bridge systems, which squeeze more bandwidth and lower latency from the very same DIMMs.[/quote<] This would have been the perfect place to mention the difference in cpu memory controllers as the reason for being able to "squeeze more bandwidth and lower latency from the very same DIMMs." Eh? WAIT... [quote<]Any differences in memory subsystem performance come down to the memory controllers on the CPUs and how each motherboard maker has chosen to tune them.[/quote<] Perfect time to mention that comparing a Phenom II mem controller to a Sandy Bridge mem controller is comparing apples to oranges, regardless of how they are "tuned," right? The Bulldozer mem controller will provide Sandy Bridge with an apples-to-apples comparison here. Since you knew that already, what was the purpose in a direct comparison between cpu mem controllers here? Obviously, that has nothing to do with the comparison of core logics, and is you making direct comparisons between cpus where you said that we should infer none. Last on the list of critical comparative omissions here is, I think, UEFI updates. There will be things particularly addressing Bobcat that simply aren't going to be there with current Phenom II cpus, even the 1080T, and whereas Sandy Bridge chipset platforms have been shipping long enough to allow for a relative maturity in UEFI code, the same is certainly not true for either of the 990FX mobos tested here with a pre-Bulldozer AMD cpu. Hopefully, when Bulldozer ships TR can install one, along with its specific Win7 drivers and any applicable UEFI updates and revisit these issues and tests. I'll wager most of them will be somewhat different in terms both of results and individual motherboards. [quote<]I'd be curious to see some statistics on the number of enthusiasts who prefer tweaking with a BIOS or UEFI as opposed to using Windows software. Perhaps that would explain why some motherboard makers seem to have devoted far more resources to developing Windows software than transitioning to UEFI.[/quote<] Well, OK. If the Windows control software allows me to do everything I can do in the UEFI, then there is simply no reason for me to have to double-boot my system to make changes as is required currently with a BIOS--that is, if the UEFI Windows software allows me to make changes in real-time as software like AMD's Overdrive currently does with a BIOS. You didn't mention if the real-time aspect applied here to the UEFI Windows software you looked at. Wouldn't that suffice as the reasonable answer to the question you raise above as to why they are devoting resources to the Windows control software? I should think this would be a very big advantage for UEFI over a standard BIOS--that it might be, or should be, far more responsive to OS control than a BIOS. [quote<]My original intent was to test with a 3 x 1080p Surround Gaming config spread across the monitors sitting on my desktop. One of the screens has a 1080p panel, while the other two scale up to 1920x1200. I've configured these displays in a widescreen trio using AMD's Eyefinity scheme, but Surround Gaming wasn't nearly as cooperative. While the screens were detected by Nvidia's drivers, enabling Surround Gaming caused all kinds of problems, from blanking screens to full-on system crashes. With the clock ticking and my will already broken by a couple of days of frantic testing, I bailed on the triple-display config and settled for a handful of gaming tests on a single 1080p monitor. Totally lame, I know.[/quote<] Well, it isn't your fault, of course...;) But I think it does point up the fact that it is most likely nVidia's SLI/drivers to blame for any paucity of performance perceived at this time, as opposed to a singular 990FX dearth of bandwidth. Overall, I think this comparison is a bit misleading since it doesn't really address the issue of how much the addition of Bulldozer is likely to change the picture. These motherboards, to borrow your word, seem "tuned" to Bulldozer with the added advantage of simply supporting older AMD cpus as well. I doubt very much that either of these companies would be selling Bulldozer boards without having obtained the necessary support information from AMD. Edit: Yea, thanks, riquee, for the correction. Hope I got them all. Getting my bc's and bd's confused is just what I deserve for going on such a nitpicking rampage...:)

      • riquee
      • 8 years ago

      BULLDOZER not bobcat.

    • flip-mode
    • 8 years ago

    Same silicon? Can we make an exception to the “family friendly” language rule for a moment, because I need to employ some vulgar language to properly address this situation.

    This chipset does not excite, and that’s about the kindest thing I have to say about it. It’s the same chipset that’s been available for over a year.

    Too bad Nvidia stopped making AMD chipsets.

      • shank15217
      • 8 years ago

      Apart from usb 3.0 what else did you expect? PCI 3.0 is about a year off and frankly you don’t know what this chipset enables for bulldozer.

        • flip-mode
        • 8 years ago

        With a whole new chipset series number, I expected it not to be identical silicon. Most of all, I expected improved SATA. Hell, improved SATA was the fundamental expectation – better performance with lower CPU utilization. And, yeah, USB 3.

        Asking for better SATA isn’t asking for the moon and stars. I can’t believe anyone would be OK with zero improvement in this area.

        Taking year-old silicon and slapping a new model number on it is a total joke and it’s sleazy. I find it a sleazy thing to do with chipsets just as I find it sleazy to do with GPUs.

          • Goty
          • 8 years ago

          It’s either rename the chipset for the inclusion of a new socket to make CPU compatibility more clear, or leave the average, none-too-bright pseudo-enthusiasts to clog up every message board on the internet with message about how his 890FX motherboard won’t work with his new processor for some reason.

          • shank15217
          • 8 years ago

          The south bridge is not the same as the north bridge, the sata perfomance may have improved, seriously you are over-reacting. These chipsets may be newer silicon with more undocumented features and compatibility fixes, I really don’t think its a re-badge, its all speculation anyways.

            • flip-mode
            • 8 years ago

            We’re going to have to come to an understanding here:

            1. If we’re going to have a reasonable discussion about this then we’re both going to have have to read the review – maybe not every single word of the review, but at least read the first page of the review.

            2. We’re going to have to agree to pretty much believe what TR says. If you’re not willing to do that then you really should go hang out somewhere else where you believe what they say, but when TR says “same silicon” I’m going to go out on a limb and believe them, so when you say “These chipsets may be newer silicon…” I’m going to start accusing you of not reading the article. When you start saying things like “its all speculation anyways” I’m going to point you out to the fact that this is an actual product launch we’re dealing with here, not a distant rumor, so this is the time for AMD to tell us anything we don’t know, and if they don’t tell us now, then “undisclosed features” mean squat until they’re disclosed. You want me to give bonus points for hypothetical undisclosed features? Is it also overreacting to call that stupid? LOL.

            You can call it overreacting, but I’m pissed about the lack of improvement in the SATA department. Now even the Brazos platform – you know, AMD’s low performance platform – has better SATA performance than can be had on AMD’s “new” FX high-end chipset.

            • shank15217
            • 8 years ago

            What are you talking about? New silicon can have tweaks or enabled features, there is no way TR or anybody would know the features these chipsets have without the platform to test. Now the article I read wasn’t a review of the motherboards it was a preview of the board with yet unknown features. Unknown because we haven’t seen why AMD insists that the 990FX is required for Bulldozer if indeed its the same silicon as the 890FX then AMD has been making things up all along. I don’t draw conclusions based on a preview, I don’t care if its TRs preview or not. AMD has paired newer north bridges with older southbridges. I don’t have any idea what the SB950 brings to the table, if it fixes the SATA performance issue or not. Far as I could tell, the SB850 had decent SATA performance and SB950 should be decent as well. Also SB950 isn’t 990FX, its just the new southbridge that would be pared with all 990XX chipsets.

            • flip-mode
            • 8 years ago

            I can see a rational conversation is out of the question. Let’s just talk about tooth fairies and stuff. Speaking of which, have you heard about those undocumented P67 features? Yeah, like, SATA 10.0 and USB 4 and Cross-SLI and stuff – things Intel never said didn’t exist and stuff.

            • shank15217
            • 8 years ago

            Lol whatever dude, you’re calling out all kinds of crap on a chipset that has barely been demoed without it’s intended cpu and all you can say is that it’s the same as the one before it. Well if 890FX and 990FX are the same then whats the difference between AM3 and AM3+?

            • flip-mode
            • 8 years ago

            AM3+ uses one more pin than AM3 if I remember correctly. I don’t know if the pin actually does anything or if it is just to prevent putting an AM3+ cpu in an AM3 socket.

            Anyway, here’s some quotes for you:

            [quote=”HardOCP”<]The 990FX chipset is simply a rebranded / rebadged 890FX chipset.[/quote<] [quote="TechReport"<]Led by the 990FX, this core-logic family is new in name only. The 990FX consists of the same north- and south-bridge components as its 890FX predecessor, and those chips continue to be manufactured by TSMC using a 65-nm fabrication process.[/quote<] [quote="tomshardware"<]The 990FX chipset employs the same silicon as 890FX.[/quote<]

            • shank15217
            • 8 years ago

            This is the first review I could find with benchmarks, SATA performance hasn’t improved over SB850, other performance improvements may just be tweaks by the manufacturers.

            [url<]http://www.legitreviews.com/article/1626/1/[/url<]

      • Krogoth
      • 8 years ago

      Chipset platforms don’t change that often. AMD and Intel have been practicing the whole rebranding the same silicon for “years”.

      955X => 975X (The only difference is that 975X had “official CF” support)

      P35 => P45 (Die-shrink)

      X38 => X48 (The latter had official 400FSB support)

      • maxxcool
      • 8 years ago

      Since the only changes needed to keep it modern are frequency changes for pins, and adding lanes to allow for more gfx-cards there isnt a reason to engineer a new chipset. Intel’s chipsets are no different. unless there is a new uber feature that affects the ‘core’ of the chipset there’s no need to do anything but add that feature off the side of the die. seriously? why would anyone need to re-engineer a entire chipset to add support for usb3, or lightpeak.. and thats all thats changed in the pc world… we now have usb and light peak…. wooo 😛

      the current chipset is actually a spin on the 7xx series, the north bridge has remained the same for 3 generations now in terms of “core” functionality. its a hub for christ’s sake… a hub. do you buy a new hub every 1 year for your local lan? is there a reason to re-engineer hubs ? nope

      also, the chipsets now, with the northbridge on die… isnt really a north bridge… again its a hub. its where stuff goes to get directed elsewhere. the southbridge does most of the lifting. and as long as amd doesn’t make it worse… then its fine.

      until there is a new bus for intel and amd they are not going to change anything major… period. theres no reason to.

        • flip-mode
        • 8 years ago

        SATA needs more cow bell. If that can be done with drivers, fine, but after this long I’m doubting it. If it needs a new chip, do it. Brazos platform has better SATA performance than SB850 / SB950, so we know it can be done. Better performance, less CPU. I feel like I’m repeating myself – probably because I am:

        [quote=”flip-mode”<]With a whole new chipset series number, I expected it not to be identical silicon. Most of all, I expected improved SATA. Hell, improved SATA was the fundamental expectation - better performance with lower CPU utilization. And, yeah, USB 3.[/quote<] Beyond that, there is the pedantic desire I have not to see chips rebranded. Can I get over that one? Yes, I can.

        • MrDigi
        • 8 years ago

        If nothing has changed why force your customers to buy a new motherboard. If 8 series work the same with a one pin bigger socket attached then AM3+ is a joke on customers. Come on not even a process/shrink upgrade.

      • deruberhanyok
      • 8 years ago

      flip-mode is on point. 🙂

      I’m disappointed to see the same southbridge in use. AGAIN. How many southbridges have they gone through trying to get the SATA performance straigtened out now? Six? Eight?

      I was really hoping for a one-chip approach, but even with sticking to two, and even without updating the northbridge, I really don’t see how they can be content with the SATA performance. It was a problem several years ago and now it’s just kind of a joke. But it’s a bad enough joke to keep me from buying into an AMD based platform.

        • flip-mode
        • 8 years ago

        Yes, the fact that AMD is still using a two chip solution is indeed curious. I don’t know if that makes board manufacture easier and cheaper or harder and more costly, but I’d guess the latter.

          • Palek
          • 8 years ago

          Like you said, the latter is much more likely.

          The key issues are almost certainly board design challenges and price – which are of course linked. I could not find detailed specifications anywhere for AMD’s chip-sets (they are probably available to customers only) so I could not verify, but I would venture a guess that the total pin count of a typical chip-set would be quite substantial. Routing all the PCIe, SATA, USB etc signals to their respective connectors/components from a single location would be quite a bit more difficult compared to a 2-piece chip-set, where you have a nice split between CPU link + PCIe links versus peripheral signals. A single-chip chip…erm…set may necessitate the use of 8 or 10-layer PCBs rather than the typical 4 or 6-layer PCB in motherboards. Needless to say 8+ layer PCBs are much more expensive.

          (I found a Gigabyte page that has a well-written explanation on PCB layers. Here: [url<]http://www.gigabyte.com/webpage/16/article_12_ud3.htm[/url<] )

          • shank15217
          • 8 years ago

          They are rebranding their server chipsets where you may have more than one 890FX on a board from more than one socket. This is quite common with quad socket and higher designs. In this case, the south bridge functionality is pretty redundant and so not normally implemented. Going forward, AMD will push the llano platform we will move into single chip solutions for desktops.

      • BobbinThreadbare
      • 8 years ago

      It has a new name to make it clear it’s compatible with AM3+ CPUs.

      Which is a new feature.

        • MrDigi
        • 8 years ago

        Not when the 8-series is also compatible.

      • TheBowerbird
      • 8 years ago

      From legitreviews: “The socket change was needed since HyperTransport has been upgraded from version 3.0 to 3.1 as AMD needs to support 6.4GT/s for their new processor design. You always want to build on the platform that supports the next generation of products and this is it.” So it’s not entirely just a rebadge.

    • ronch
    • 8 years ago

    There you go.

    [url<]http://www.fudzilla.com/processors/item/22904-bulldozer-pushed-back-for-late-july[/url<] That's what happens when you miss your product and process refreshes and have to build a new architecture on a new process. AMD does have a tick-tock of sorts, introducing new processes every two years or so, and introducing new architecture refreshes (albeit not as revolutionary as Intel's) every two years or so as well. Except, AMD's schedules get skewed noticeably. And when these schedules collide, this is what happens.

      • Goty
      • 8 years ago

      I’m pretty sure the July launch date was a rumor to begin with, so it really can’t be “pushed back”. The only date that we’ve known for sure is the June unveil.

        • Game_boy
        • 8 years ago

        Launching for desktop in Q2 was all that was official. From the Analyst Day.

        • ronch
        • 8 years ago

        June 20 is what’s been said all along, but the late-July rumor came out just a couple of days ago, reported by Anandtech. Seems the B0 and B1 steppings weren’t reaching performance targets, and as someone already said, don’t expect miracles from a new stepping. All the silence around Bulldozer isn’t a good sign. I’ve been waiting for BD for so long and if AMD couldn’t get it right, and if this is another Barcelona (read: all hype, disappointing product), I’m heading to Ivy Bridge.

          • Goty
          • 8 years ago

          If the silence means anything like the silence before the Radeon 4000 series and later launches, I think we’ll be more than satisfied. If you’ll remember, the last launches where AMD made a lot of noise were the original Phenom launch and the 2900 series (though that could be attributed to ATI as a separate entity still).

            • ronch
            • 8 years ago

            The launch of Bulldozer is different from all those GPU launches. And it’s practically more momentous than the Barcelona launch in 2007. So much is riding on the success of Bulldozer. Having revealed the architecture back in August 2010, everyone is expecting AMD to at least show some wafers or demo some systems, even if they’re not running at full speed. As it is, mobo makers don’t even have final samples to test on their AM3+ products. Not a good sign at all. Mobo makers must be pretty pissed off right about now. The silence is deafening.

            • Palek
            • 8 years ago

            This sounds about right, as much as I hate to admit it… It would be nice for AMD to make a thunderous comeback, but the lack of early demos is not a good sign.

        • ronch
        • 8 years ago

        I hate to break it to ya, but…

        [url<]http://news.softpedia.com/news/AMD-Bulldozer-CPUs-Too-Slow-to-Compete-with-Core-i7-Chips-203267.shtml[/url<]

          • shank15217
          • 8 years ago

          This is pure FUD, bulldozer will be released Q2 as planned. There is a huge difference between September and 2012 and AT mentions July launch which also is 2 months before September.

            • ronch
            • 8 years ago

            Don’t believe it? How about this?

            [url<]http://www.anandtech.com/show/4393/computex-2011-amd-bulldozer-llano-trinity-new-vision-branding[/url<]

    • ronch
    • 8 years ago

    Looking at the chipset diagram made me notice a few things I never really gave much thought to before.

    1. Why do current AM3 motherboards offer 5.2GT/sec Hypertransport when the fastest desktop AMD processor, the Phenom II X6 1100T, only has 4.0GT/sec of Hypertransport bandwidth? Surely you’re not gonna plug in an Opteron (which probably doesn’t use that much bandwidth anyway)? And if one would think that extra 1.2GT/sec of bandwidth is reserved for future, more bandwidth-hungry SKUs, well, why wouldn’t AMD let today’s fastest CPUs use all that bandwidth [u<]now[/u<], power-consumption considerations aside? Some motherboards may support maxing out the HT speed, but going out-of-spec is beside the point here. 2. Even at 6.4GT/sec, HyperTransport 3.1 can only offer a max theoretical aggregate bandwidth of 25.6GB/sec, since HT on AMD boards use two 16-bit links, one up, one down. That makes it a 32-bit link with 12.8GB/sec in each direction. So, if the 890FX/990FX offers 42 PCIE lanes, this gives it a maximum bandwidth of 42GB/sec, since each PCIe lane has an aggregate throughput of 1GB/sec. AMD talks about bottlenecks and how their Direct Connect Architecture<tm> solves all that, and this bandwidth probably wouldn't be maxed out all the time, but the chipset diagram here makes it look like the PCIe lanes are outgunning the HT connection.

      • shank15217
      • 8 years ago

      Because the 990FX is just a new revision of AMDs server chipset.

      • mczak
      • 8 years ago

      Yes the “advantage” of 42 pcie lanes is quite theoretical. One 16x pcie 2.0 link offers 8GB/s bandwidth (each direction), just the same as what HT offers with current cpus. So there’s likely zero benefit of the “true” 32 lanes for graphics over the Sandy Bridge 16 ones.
      Though if we really see cpus with 3.2Ghz HT frequency instead of 2Ghz (and I’ve got some doubts they’ll increase it to that outside of server space) it COULD provide some (rather minimal in practice) advantage for SLI/CF – but that’ll only help if the cpu has at least similar performance…

    • Ushio01
    • 8 years ago

    If the southbridge has gigabit ethernet and hd audio why use realtek chips for both?

      • Palek
      • 8 years ago

      Board vendors know the Realtek chips inside and out (having used the for many many years); they can quickly design them onto any new board. The time and labour (design as well as evaluation) saved by cut&pasting existing designs probably outweigh the cost benefit of using the integrated units.

      Additionally, the south bridge only integrates the Ethernet MAC, so it would still need a PHY chip. This reduces the cost benefit of the integrated part.

      As for the audio, I’m not 100% certain, but I thought that the integrated audio still needed external CODEC (ADC and DAC) chips.

        • Duck
        • 8 years ago

        Good points.

        [quote<]As for the audio, I'm not 100% certain, but I thought that the integrated audio still needed external CODEC (ADC and DAC) chips.[/quote<] I don't know for sure, but I bet it does.

          • Palek
          • 8 years ago

          It does indeed:

          [url<]http://www.intel.com/standards/hdaudio/pdf/High_Definition_Audio_1_0a.pdf[/url<] Audio DACs and ADCs generally do not play nice with the high-speed digital and analogue circuitry present in south bridges. Attempting to put them on the same piece of silicon would pretty much guarantee noise-ridden audio, or present an incredibly difficult and costly design challenge. This is one of the main reasons why the AC'97 and HD Audio standards separate the DACs and ADCs.

    • Palek
    • 8 years ago

    [quote<]My changes are not-so-subtly highlighted in pink, and there really isn't much to see. The old 890FX block diagram shows a HyperTransport 3.0 processor link with 5.2 GT/s of peak interconnect bandwidth. With the 990FX, that link gets upgraded to version 3.1 and 6.4 GT/s of bandwidth. Because the north-bridge silicon is unchanged, the fatter pipe is likely tied to the socket rather than the chipset. I'd expect AM3+ boards with 8-series chipsets to offer the same boost in HT link speed.[/quote<] I would think the HT speed difference is simply the result of binning. The HT transmitters/receivers on some silicon will have better frequency margins, giving AMD the opportunity to actually differentiate between the 990FX and the 890FX. (Whether the HT speed increase will present a meaningful difference or not is another matter.) [EDIT]Of course the AM3+ CPUs will also have to support HT3.1. But I suspect that 8xx series chipsets will stick with HT3.0 even when paired with HT3.1-capable AM3+ CPUs.[/EDIT] [EDIT 2]Almost forgot, thanks for the article! The Asus board looks pretty good. It would have been nice to see an 890FX board thrown in for comparison, but we can't have it all, can we?[/EDIT]

Pin It on Pinterest

Share This