Author: Terren Tong
Editor: Howard Ha
Publish Date: Monday, March 21st, 2005
Originally Published on Neoseeker (http://www.neoseeker.com)
Article Link: http://www.neoseeker.com/Articles/Hardware/Reviews/dfinf4sli/
Copyright Neo Era Media, Inc. - please do not redistribute or use for commercial purposes.
Making a big transition is never easy. I don't know if DFI's move from being an OEM board manufacturer to a high end enthusiast company should even be classified as a transition as the chasm between the two spheres seems huge. The strategy and focus change seems to have worked for DFI though as they have literally exploded onto the enthusiast scene on the strength of a few very highly regarded products. While it is easy to want to change, the devil is in the details and in the case of motherboards, it is the implementation that matters the most.
Socket 939 and PCIe is an important one for a lot of mainboard manufacturers as the last eight months or so must have been a marketing nightmare for a lot of companies with the continual transitions, first the move to Socket 939 from 754 then the subsequent move from AGP to PCIe. Both were branded as must have features to be on the cutting edge so it must have been a hard sell to try to promote older technologies heavily. In the late fall, NVIDIA gave an even more compelling reason to wait: SLI. With the nForce 4 SLI chipset, power users would be able to harness the power of two identical GeForce 6 cards and almost double the graphical horsepower. The nForce 4 SLI chipset can only be described a perfect fit into DFI's mould - both DFI and NVIDIA are aiming for the high end, high performance, power user crowd. The only question is how good is DFI's implementation of the nForce 4 SLI and we will answer that question today with a look at the LanParty nF4 SLI-DR.
|
CPU
|
|
Chipset
|
|
Front Side Bus
|
|
Memory
|
|
BIOS
|
|
Power Management
|
|
Hardware Monitor
|
|
Audio
|
|
LAN
|
|
IDE
|
|
Serial ATA with RAID
|
|
IEEE 1394
|
|
Rear Panel I/O Ports
|
|
I/O Connectors
|
|
Expansion Slots
|
|
PCB
|
The nForce 4 chipset provides 4 SATA ports natively through the MCP along with the two IDE channels. The DR model adds an additional 4 SATA ports through a Silicon Image Sil 3114 chip like the Gigabyte K8NXP-SLI that we took a look at earlier. Each of the SATA solutions have their own advantages - the SI chip allows for software RAID5 support which provides some of the advantages of both RAID0 and RAID1 in a three drive setup. Each nForce 4 native SATA port allows for 3Gb/s of bandwidth, double that of strictly SATA 1.0 implementations. RAID partitions on the nForce 3/4 MCP can also span across both the IDE and SATA controllers for some monster drive configurations.
Audio for the nF4 SLI-DR is provided through Realtek's ALC850 codec but the implementation is a little different on the nF4 lineup. A quick glance at the back panel for the board will reveal a large empty spot where the usual audio jacks are located. A lot of enthusiasts usually scoff at onboard audio because of electrical noise introduced by other components on the board. DFI's response is the ALC850 chip being moved onto its own PCB and has dubbed this arrangement Karajan audio. There is a coaxial in and out located on the back of the motherboard. Interestingly enough, there is a jumper that lets the end user toggle between the coaxial connection being handled by the NVIDIA MCP unit or through the ALC850 codec.
The nF4 lineup from DFI shares the same PCB so the second PCIe x16 slot is found on both the nF4 Ultra and the nF4 SLI. In fact, both boards even share the same BIOS despite the fact that there is the SI controller found on the SLI-DR that was not present on the Ultra-D. Unlike the hoops that have to be jumped through on the Ultra-D board to get SLI working, the infamous SLI bridge piece is included with the SLI board and no modding is needed to get SLI to function correctly on the DFI board.
There is no doubt that a mainboard is coming with a lot of extra stuff than the usual cable or two to placate the end user when the box is double the usual size. Opening up the main box is like taking apart a Russian nesting doll - inside is a collection of additional boxes housing the various components and extras.
DFI has done their homework in choosing the items included in the full blown LanParty kit - UV reactive cable sleeves, front panel connectors with the FrontX kit, a PC transport kit and of course stickers. Rumor has it that stickers add a few extra megahertz to any overclock when applied correctly to the outside of the case. When applied on a car or bike it adds more horsepower. Kidding aside, it is a nice touch and I'm sure that there will be a few cases that will be sporting the sticker at the next LAN I attend.
FrontX is a cool concept - it sits in a 5.25" drive bay and allows for various types of connections to be moved to the front of the case. The bay is modular so that it can be customized so it should prove useful for the person who has found the ultimate case as far as construction goes but may be missing a firewire port or whatever. The choice of connectors in the LanParty package is a bit strange though and I think I would have picked a different set of components if it were up to me. Included is a SATA connector, an optical out connection, a 4-pin firewire connector and a LED diagnostic panel. The LED diagnostic panel on the front is a great idea. If for whatever reason the board does not boot up, there is no need to tear open the case to diagnose the board. The other three choices I'm less sure about. The SATA connector is a good idea since it is hot swappable but there is no way to power it with the included components. Optical out on the front is always a peeve of mine especially when the bay cover is stealthed. My perspective (as I have voiced many times in SFF reviews) is that when someone uses optical out, it is not an one shot deal but likely to be a connector that is somewhat permanent which makes little sense to put it up front. Firewire is a good choice is to include as a front panel connector but make it the 6-pin connector. Two of the firewire devices I use are the iPod and a DV camcorder both of which are 6-pin. I'm not too sure of the popularity of 4-pin devices. I do understand that there is already a 6-pin connector on the back but I still think that an additional 6-pin connector is more useful than a 4.
 |  |
The PC Transport strap is a cool thing especially for people who do frequent LAN parties with a tower. Someone here at the office has used it to drag a system home on public transit before. It works!
Like the plastic components on the board, and the sheathing the included cables are all UV reactive so those with a blacklight inside the case should be in for a treat.
Aesthetically the nF4 SLI-DR board looks real great. The color scheme is loud but is not overdone. Two orange DIMM slots complement the heavy use of yellow plastics for the rest of the connectors. Both contrast very well with the black PCB and makes for a sharp looking board. Although color scheme is highly subjective this is one that definitely ranks near the top of my list.
The board layout on the nF4 is very well thought out for the most part. The DIMM slots are aligned at the top of the board like the MSI K8N Neo board we took a look at last year. This is an important move for this board in particular with the ability to pump a lot of voltage into the DIMMs. Since there is generally more airflow in this area of the case from both the ATX exhaust and hopefully fans near the CPU cavity, it should allay some memory cooling concerns for those who push the limit a bit harder.
The CPU socket is located right dab in the middle of the board as opposed to being closer to the top edge of the board. This is another good design decision especially when monster coolers like this Zalman unit are used. One of the concerns that have been brought up to us after the review is the difficulty of mounting that particular unit on certain motherboards not because of a lack of space around the CPU cavity but because of the design of a lot cases nowadays that have the powersupply very close to the edge of the motherboard. Because of the width of the Zalman, CPU sockets near the edge of the board may be incompatible even though Zalman tries very hard to avoid any problems with capacitors around the socket area with an aggressive tapering design. With the socket so close to the center of the board however, DFI has also eliminated the problem even though this problem does not necessarily fall into the domain of board makers.
Both of the basic power cables, the ATX connector and the 12V P4 rail are located right beside each other on the right side of the board. We like this a lot as it helps with cable cleanliness unlike designs that have them on opposite side of the boards. Note that the ATX connector is a 24-pin and not a 20 pin design. We've heard various accounts of instability with people running 20->24pin connector converters so it may be a good idea to drop down the extra cash for a new PSU.
There are two additional power connectors on the nF4 however - a molex and a floppy connector to provide additional power to the board. The DFI documentation states that both of these are optional but will help system stability when running a SLI setup. When we were doing some reading, there were claims that the molex is actually for the memory when running at 3Vs +. While the molex connector does not sit near the main ATX connector is close to the IDE and FDD connectors so that those cables can probably be bound together. The floppy connector is in a more awkward place as it sits underneath the processor socket.
There are a total of five fan headers spread around the board and this should be ample room for throwing on additional cooling. Diagnostic LEDs reside in the lower left corner of the board. As mentioned before, the LEDs can be pushed to the front with the FrontX kit.
Because enthusiasts like their fans, DFI has included not one but five fan headers scattered around the board. LED diagnostic readouts are available in the corner of the board. While they do their job adequately, I've really grown to like the numbered readout found on the Soltek FGR and EpoX boards. One awesome feature is the built in power button and reset switches located by the front panel connectors. For the tinkerer it sure beats shorting out the front panel leads. The DR versions of the nF4 boards also come with an extra four SATA ports courtesy of a Sil3114 controller. The markings and chip position are outlined on the board but none of the connectors are present on the Ultra-D revision.
The MCP fan unit is also of note. DFI specifically mentioned this in their PR material. The fan is neither a sleeve or ball bearing fan like most but is a maglev fan. DFI makes several claims about this fan
The longevity issue is something of note. Too often do manufacturers put junk fans on motherboards that break within months and make a huge racket. It would be interesting to visit this issue again a few months down the road and see how well it has held up. For those that are looking for aftermarket MCP coolers, note that the back of the graphics card will always hang on top of the MCP so the profile must be quite low.
The nF4 BIOS is pretty complex and we've done our best to summarize the different settings available with most of the enthusiast settings. We will go into some commentary about the BIOS on the following page but here is summary of settings available on the nF4 SLI-DR (most of which relate to memory).
| FSB Bus Speed | 200-456 Mhz in 1 Mhz increments |
| HyperTransport Ratio | 1-3 in 0.5 steppings 3-5 in integer steppings |
| HyperTransport Bandwith | 8/8, 16/8, 8/16, 16/16 |
| CPU Multiplier Ratios | 4 to CPU Max in 0.5 steppings |
| PCI Express Frequency | 100-145Mhz in 1 Mhz steppings |
| CPU Voltage | 0.800-1.550 in 0.0125v steppings; voltage multiplier settings 1.00, 1.04, 1.10, 1.13, 1.23, 1.26, 1.33, 1.36 |
| HyperTransport Voltage | 1.20v to 1.50v in 0.1v steppings |
| Chipset Voltage | 1.5v to 1.8v in 0.1v steppings |
| DRAM Voltage | 2.5v to 4.9v in 0.1v steppings |
| Memory Divider | 1:2, 3:5, 2:3, 7:10, 3:4, 5:6, 9:10, 1:1 |
| Command Per Clock | On/Off |
| CAS Latency Control (TCL) | 1.0-4.5 in 0.5 steppings |
| RAS Latency Control (TCL) | 0-7 Bus Clocks |
| Min RAS active time (Tras) | 0-15 Bus Clocks |
| Row Precharge Time (Trp) | 0-7 Bus Clocks |
| le time (Trc) | 7-22 Bus Clocks |
| Row Refresh Cycle Time (Trfc) | 9-24 Bus Clocks |
| Row to Row Delay (Trrd) | 0-7 Bus Clocks |
| Write Recover Time (Twr) | 2-3 Bus Clocks |
| Write to Read Delay (Twtr) | 1-2 Bus Clocks |
| Read to Write Delay (Trwt) | 1-8 Bus Clocks |
| Refresh Period (Tref) | 16-4708 Cycles; varied stepping size |
| Write CAS Latency (Twcl) | 1-8 Bus Clocks |
| DRAM Bank Interleave | On/Off |
| DQS Skew Control | Increase/Decrease Skew |
| DQS Skew Value | 0-255 |
| DRAM Drive Strength | Level 2-8 |
| DRAM Data Drive Strength | Level 1-4 |
| Max Async Latency | 1-15ns |
| Read Preamble Time | 2-9.5ns in 0.5 ns steppings |
| Idle Cycle Limit | 0-256 Cycles; stepping increments are double the previous value |
| Dynamic Counter | Enable/Disable |
| R/W Queue Bypass | 0,4,8,16 |
| Bypass Max | 0-7 |
| 32 Byte Granularity | 4 or 8 Bursts |
Since we're using the same BIOS as for the Ultra-D, there are very few changes to the SLI-DR's BIOS.
Like one of the myriad of custom chopper or custom car shows, DFI takes a basic Phoenix AwardBIOS and adds a lot of extra functionality to it DFI style. A couple menu options of note are the Genie BIOS and the CMOS Reloaded options which have been in several generations of DFI's products so far.
The standard CMOS Features menu is pretty boring. One thing of note is that it shows the onboard IDE/SATA controllers only while it is in non-RAID mode. Devices attached to the Silicon Image controller will not be listed on this particular screen.
 |  |
There are probably several things that end users will want to change in the Advanced BIOS menu. First is disk boot priority. No recommendations here, you should know better than I do what you want to boot off of. The Hard Disk Boot Priority menu helps differentiate between which hard drive gets to boot first which is useful for those who decide to load up the drives on the SATA and IDE controllers. Scrolling down a bit is two options that probably need to be changed; generally we like to keep S.M.A.R.T. on even though its probably too late by the time you get the warning. Second is the Full Screen LOGO Show. DFI has a cool looking boot screen but those that are tweaking and setting up initially will likely prefer the familiar BIOS screen listing drives, some memory settings and what processor is being used.
 |  |
DFI has a more extensive Integrated Peripherals menu than most though the default options should be fine for most people. Those that have a fancy sound card and are sure that they do not need to use the Karajan audio module can turn off AC97 audio to disable onboard sound. On a non-BIOS related note, there is a jumper on the board that can change the SPDIF source from the ALC850 controller to the nForce 4 MCP. The Karajan module does not need to be installed for the MCP to drive the SPDIF. We were unable to get any further clarifications between quality or perfomrance differences. Strangely enough the Silicon Image controller is not listed under here. There are some options to change the way that data is read over USB here. Vivian over at DFI did try to get some answers for us regarding the USB options -
"Isochronous transfers, periodic, continuous communication between host and device typically used for time relevant information. This transfer type also preserves the concept of time encapsulated in the data. This does not imply, however, that the delivery needs of such data are always time critical."
Changing this in BIOS will affect the reserved debug register in CK804 which will change certain transfer type formats , but NVIDIA's documentation does not give more details regarding effects with setting ISO or Non-ISO. Users should leave USB settings at the default value unless there is some compatibility issue in the USB device, as this is reserved for debug purposes."
Time sensitive devices can include things like USB headsets and TV Tuners. It does not sound like anything will get broken so it might be interesting to fool around with these settings especially for some of the more bandwidth hungry operations with USB HDs.
Power Management isn't my bag baby. No matter what the cover of the book says. My computer is generally always on, I don't care much for wake on LAN, sleeping hard drives or anything else but the options are here for those that do.
The only thing I ended up changing in PNP/PCI Configurations is setting the Init Display First option to PCIex.
The PC Health Status shows various statistics on what voltages are being fed to your board and lots of other good stuff including temperatures and such. Be sure to check out a recent article of ours that looks at the accuracy of BIOS reported voltages as compared to what a multimeter sees. For those that obsess about noise as I do, there are fan adjustment options although it would be great if there were more options than off / throttled and full blast.
All of the overclocking settings are stored under the Genie BIOS menu. There are a lot of things that stand out but several in particular will be highlighted. The front side bus is adjustable from 200-465. While a lot of manufacturers have absurdly high FSB values, there are few that can claim that the majority of the FSB range is usable. DFI will be one of those that can make such a claim as will be seen later on.
 |  |
The Genie BIOS Settings menu is where DFI's magic really happens as this where all the overclocking voodoo is located. The front side bus is adjustable from 200-465Mhz - while a lot of manufactuers have a wide range of FSB values, most of them cannot claim that most are usable. DFI can as will be seen later on. The only change we would suggest is the ability to type in a value instead of forcing the user to scroll through 150+ values. Voltage adjustments for the CPU on the nF4 is pretty unique. Besides the standard 0.0125 voltage stepping, DFI also has something an additional fine tuner if you will - a multiplier can be applied to the base voltage to change the actual voltage fed to the CPU with even smaller increments. The ranges are listed in the table and with the highest multiplier of 1.36, a whopping 2.108v can be applied to the processor. Voltages are shown on the Genie BIOS page which is good but the functionality could be improved - DFI shows the current voltage setting that was saved. It might be more useful to show the default voltage instead since the set voltage is already shown through the voltage selection menu.
Memory voltage can be cranked up to 4.0v. Because the potential for damage is there, DFI limits the range by default from 2.5v to 3.2v. A jumper on the motherboard has to be set in order to unlock the 3.2-4.0v range.
The large set of memory dividers is also very welcome as this will allow again for better fine tuning of the memory when the FSB starts hitting the stratosphere. To help test for memory stability, DFI also has MemTest86+ built in as a BIOS option. With this enabled, after the post, MemTest86+ will launch. No CD required, all of this is stored on one of the onboard flash memory chips. How cool is that?
The only new option in the Genie BIOS that is not present on the nF4 Ultra-D is SiI3114 item that allows for some adjustments to the SiI controller including whether it should be simply a SATA controller or a RAID controller. Generally RAID1 arrays cannot be transported between different computers but RAID0 on the other hand is controller specific but there is option called SATA RAID BY OLD MB - this allows for RAID arrays built on the SiI 3114 controller on a non-RAID5 revision to be accessed and dumped to another disk.
That said there are a few options that might be better off in the Integrated Peripherals menu including the options for the SiI, the SATA ports from the MCP, the Firewire controller and finally the Marvell ethernet controller.
In addition to the very impressive voltage range and options, there is also the DFI exclusive, CMOS reloaded. After the initial post screen and boot detection, settings that are bootable are saved to the SEEPROM. The user then has the option of naming and saving this, or any other configuration with the CMOS reloaded configuration. Four save banks are available and at the whim of the user, each of these sets can be loaded. A cool future option may be a save to floppy disk option. I noticed the documentation mentioned that this would be useful for system integrators but I'm thinking that being able to trade configurations between friends with the same board might be kind of neat also.
The nForce 4 Family
For further details about the nForce 4 family and features, we encourage readers to check out our preview of the nForce 4 platform where we tackle some of the improvements from the nForce 3 in some more detail. Some of the topics include nTune, the improvements to the Firewall and to the drive controllers.
Hardware & Benchmark Setup
Benchmark Setup
Our benchmark system consisted of the following:
NVIDIA's Forceware 66.93 was used for the 6600GT across all platforms. The DFI nF4 SLI-DR was tested with the Forceware 6.39 driver package. The nF4 Ultra-D was tested with the ForceWare 6.37 drivers. VIA's Hyperion 4.55vp1 and the driver revisions for the Gigabyte nForce 4 SLI board can be found here.
A list of benchmarks performed follows:
The spread is quite close - the nF4 Ultra-D ends up in the middle of the pack while the SLI-DR takes a small hit and rounds up the rear.
Multimedia Winstone 2004
The DFI and Gigabyte boards throw out some pretty consistent numbers at 29.4
SiSoft Sandra Memory Test
Surprisingly enough it is the VIA K8T890 board that takes top honors with regards to memory bandwidth. The two DFI nF4 boards are right up there though. Surprisingly enough, both the Gigabyte boards seem to be a bit hampered on the memory bandwidth side. Seems like they may want to tweak their BIOS a little bit.
Three different drives from three different manufacturers are used in our HD I/O tests so scores are relative between boards, not between drive interfaces.
IDE HDTach
There is very little variation in IDE drive scores. The small variations in the Burst Rate and the processor usage can be simply chalked up to testing ambiguity.
SATA HDTach
Average read rate remains the same between all the drives but there is definitely some variation with the Burst rate - the difference between the fastest controller and the slowest more than 7%. Both the Gigabyte boards and both the DFI boards seem to perform similiarly and DFI seems to have their implementation tweaked a bit better.
USB2 Throughput
USB2 throughput on the nForce 4 boards is consistent if nothing else. The K8T890 implementation does come out on top by a fair margin though. It looks like there are some occasional hiccups as the processor usage did vary a bit between the four boards.
LAN Testing**
Processor usage for the GbE connection provided by the nForce 4 MCP is around the 35% range - this is not correct as with the ActiveArmor, the processor usage here should be in the 20% range. We cover the issue in more detail here and after a couple weeks of sleuth work and a lot of phone calls NVIDIA has resolved the issue. We are still waiting for updated drivers to test this but CeBIT has been complicating things a bit.
Throughput as measured by NTTTCP is amazing on the nF4 boards - they actually break the 1Gbps barrier. The Gigabyte boards are no slouches but they do not manage to keep up. The VIA K8T890 solution trails behind significantly. The second GbE controller is a PCI Marvell based unit. Hampered by the PCI bus, throughput is only roughly 751 Mbps while processor usage was a bit lower at around the 33% mark.
There are some major variations with the processor usage as measured by the Rightmark Audio Utility. The Gigabyte GA-K8N Ultra-SLI board cleans house here finishing significantly ahead of the rest of the pack. The nF4 SLI-DR does relatively well in comparison to the other boards featured.
Call of Duty paints a slightly different picture. The Soltek SL-K890Pro finishes on top of the heap while the synthetic sound benchmark leader, the GA-K8N-SLI falls to last place. The nF4 SLI-DR board falls a couple notches too and ends up being a frame or so slower than the Ultra-D.
MP3 Encoding
MP3 encoding is pretty consistent across all the boards as they finish within seconds of each other over a span of 15 minutes.
Xvid Encoding
There is quite a bit more variation with Xvid Encoding done through AutoGK. The two Gigabyte boards are quite a bit slower than the nF4 boards - the gap between the fastest board, the nF4 SLI-DR and the Ultra-SLI is a not so insignificant 40 seconds.
Comanche 4
Comanche 4 is a good indication of triangle throughput - the DFI boards take the one two by a smidgen over the rest of the pack.
Halo
Again the gap between first and last is large - there is a 9% difference between the Soltek board when compared to the GA-K8N Ultra-SLI. The nF4 SLI-DR captures second place with the Ultra-D right behind it. Both are clear of the Gigabyte boards.
X2: Rolling Demo
The K8N Ultra-SLI has a resurgence here and takes the top spot in Egosoft's X2: Rolling Demo. The nF4 SLI-DR is middle of the pack but is a few frames off the pace of the leaders.
Call of Duty
The Soltek takes the top spot with Call of Duty. Right behind are the two DFI boards. There is a fairly significant gap between the top three boards and the two Gigabyte SLI boards - they do have some trouble keeping up and there is the lingering question of if this is related to the comparatively lower memory benchmark scores seen with Sandra.
UT2k4
The story here is similar to Call of Duty except the DFI pair manages to take 1, 2.
Half-Life 2
Some very nice numbers from the two DFI boards in Half-Life 2. The SLI-DR takes top honors followed by the Ultra-D. The Soltek board is not too far behind, but once again we see the Gigabyte boards round up the rear.
DOOM3
The nF4 SLI-DR once tops out the boards in DOOM 3 by a slight margin. There K8NXP-SLI has some trouble keeping up though the K8N Ultra-SLI puts in a good performance.
SLI Testing
With SLI, we felt a need to go a bit beyond the usual motherboard benchmarks and take a look at how a board performs in a SLI configuration and also to get a taste of what to expect while overclocking. DFI has chosen to implement a jumper system to toggle between SLI and single card mode. The jumper system also allows for more lanes to be routed to the PCIe x4 slot from the default of x1 bandwidth to x4. The jumper process is more time consuming than card method seen on most of the other boards but I think there is arguably less ambiguity with this method. Upon the arrival of our first Gigabyte SLI board, we had a small debate as to whether it was in SLI or Normal mode.
There are a couple of quirks with with SLI mode enabled - unless two video cards are plugged in while it is in the SLI configuration, the board will not POST with a single video card plugged into either slot. This was not the case when testing the Gigabyte boards. DFI says that this is planned functionality. A second quirk was the way that the video cards were initialized before SLI was enabled through the NVIDIA drivers - on the first video card, there would be output during the boot process, but once the video card drivers were initiliazed by Windows, the display would go blank. On the second card, there would be no output during the boot up sequence but then after the drivers were initialized by Windows, the display would appear. After enabling SLI through the driver control panel, the video output on the first card worked correctly for both the boot process and Windows.
In the charts below are four different configurations -
DFI LanParty nF4 SLI - this is the default configuration used in the motherboard benchmarks. Games are run witih low quality settings where indicated. Clock rate on the processor is 200x9 which gives 1800 Mhz
Single 6600GT @ 640 - this is an OVERCLOCKED configuration. Game settings are the same ones used as the default motherboard configuration setup. Clock rate on the processor is 250x9 giving 2250 Mhz. Memory timings remain unchanged and are at a 1:1 ratio. This is strictly a 25% increase in clockrate.
Single 6600GT @ 1600 - this is an OVERCLOCKED configuration. Clock rate on the processor is 250x9 giving 2250 Mhz. Memory timings remain unchanged and are at a 1:1 ratio. Graphics settings ARE changed. Resolution is changed to 1600x1200 with the game settings maxed out. AA/AF is not applied, but details, texture quality and everything else is maxed out. This will be a typical graphical configuration when playing games.
SLI 6600GT @ 1600 - this is an OVERCLOCKED configuration. Clock rate on the processor is 250x9 giving 2250 Mhz. Memory timings remain unchanged and are at a 1:1 ratio. Graphics settings ARE changed. Resolution is changed to 1600x1200 with the game settings maxed out. AA/AF is not applied, but details, texture quality and everything else is maxed out. This will be a typical graphical configuration when playing games. We throw in the second 6600GT card and enabled SLI to see the performance benefits that should be expected from a handful of games.
A 25% in clockrate gives a 24% boost in framerates. DOOM 3 at 640 is very processor dependent. When the resolution is raised and the graphic options that make DOOM 3 so pretty is turned out, the framerate drops to 46.2 FPS. Dropping the second 6600GT card in there raises the framerate 66%, up to 76.6 FPS
Again at 640x480 we get nearly a 25% bump in the framerate - the clockspeed and framerate are linked linearly. Raising the resolution to 1600 drops us to 41.476.40 FPS. Adding the second 6600GT results in a 85% boost in framerates at 1600.
Once again the framerate increase from the processor overclock is roughly 24%. Raising the bar up to 1600x1200 and maxing out the graphical features does not hurt the framerate nearly as much as DOOM3 or Halo. In SLI mode, the increase in framerate is merely 15%.
With X2, we have do not have the strict CPU bound that was seen with the previous three games. This time, the overclock results in only a 13% jump in framerates. Turning on shadows and raising the resolution to 1600 put a big strain on the 6600GT. With SLI enabled there is a 36% increase in framerates.
Overclocking on the DFI nF4 SLI-DR still comes in at a blistering rate. In SLI mode, the board stayed stable at 359, which is a bit shy of the 365 that was achieved with the Ultra-D board. Despite bumping up the LDT and Chipset Core voltage, the board would lock in Prime95 at 360. In single card mode we did not have a lot more luck - 360 stayed stable for 22+ hours but beyond that Prime95 would tend to lock up. Again, additional voltage to both the LDT (HyperTransport) bus and the chipset did not seem to help with stability. While the SLI-DR did not set any new lab records, it shows that the Ultra-D is not a random fluke. A couple things of note - with some of the failed overclocks, a reboot put the BIOS back into default settings. For those that have spent hours tweaking the memory just right, it might be an idea to take advantage of the CMOS Reloaded feature so settings do not get wiped out accidentally for the one time the FSB gets pushed a bit too high.
Say cheese! The nF4 SLI-DR stayed composed during 22+ hours of Prime95 with a 360Mhz FSB before we stopped testing
Like the nF4 Ultra-D, DFI has once again, unsurprisingly put together a very solid board with the SLI-DR variant and has shown that the overclocking abilities of the Ultra-D are not an one off fluke. With the full blown LanParty package, DFI has included a lot of items that should be of interest especially for those who are getting active in the modding and LAN party scene. Some of the nice value-added items include a case carrying strap, UV cable wrap and the FrontX kit (although the choice of some of the FrontX components are a little questionable from my perspective). Aesthetically the look of the board is top notch; the large aluminum mosfet coolers sitting between the bright UV reactive plastic stands out but is not overwhelming. Overall, the design of the board is well thought out. A couple minor gripes include the placement of the USB headers, optional molex and floppy power connectors which we would like to see closer to the edge of the board. Conversely, having the DIMM modules above the processor should work better in cooling high voltage memory modules. The processor socket is located in a very central location on the board which means that large cooler units like the Zalman will not be impeded by the power supply or anything of the sort. DFI accordingly pairs a premium audio solution with their Karajan audio module as opposed on an arguably premium motherboard instead of plunking down the ALC850 codec directly onto an electrically noisy mainboard.
The DFI BIOS is very robust and in particular the memory adjustment options are second to none. If the answers we got from the question about the use of the USB transfers options are any indication, DFI is providing a very direct interface to the workings of the nForce 4 chipset, and throwing caution completely to the wind so that there are few limits for the savvy end user as it seems like anything that can be possibly tweaked with the nForce 4 chipset is fair game. Some of the other BIOS highlights are the extreme voltage options, a built in version of MemTest+ and the CMOS Reloaded feature which allows the end user to save different configurations. Another bonus is support for single slot SLI cards; right now it looks like the only one will be the Gigabyte 3D1 but ASUS does have a 3D1 clone and a 6800Ultra SLI monster on a single board also coming out and I have no doubts that DFI will end up supporting both of these as well.
The benchmarks are generally very good for the nF4 SLI-DR as it was near the top of the pack for the most part and never really ran into any serious drops in performance unlike the two Gigabyte SLI boards that were looked at. While the integrated memory controller of the Athlon 64 lineup has sometimes limited performance differences, there are still gaps of about 10% between the fastest and slowest boards.
Overclocking results were excellent - the FSB stayed stable at 359 in SLI mode and 360 in single card mode but it is a few megahertz short of what we were able to get out of the Ultra-D. However this is by far, a lot higher than what we have been able to squeeze out of other Socket 939 boards thus far.
There is not a whole lot more to say about the DFI LanParty nF4 SLI-DR that we have not already said about the Ultra-D. This should be on the short list of boards to get for anyone that has the slightest inclination to tinker and is also looking for a SLI board. This is a board that will let power users squeeze as much performance out of their 939 chip as possible. It is not often that it can be claimed that one particular part will never hamper a spirited overclocking attempt but the nF4 SLI-DR is one of those rare cases where there should be no doubts about its pedigree.
Special thanks to NCIX for providing us with the board sample!
Please do not redistribute or use this article in whole, or in part, for commercial purposes.