Review: DFI NB72-SR

Introduction

It seems that the i845 is a popular chipset to base a Pentium 4 motherboard around these days. With my review of the Asus P4B out of the way, the next board on my pile was DFI's NB72-SR. DFI are one of those companies that I had heard of but hadn't seen much from them in the way of products, especially to buy here in the UK.

It's always nice to feature a new manufacturer since a new manufacturer equals a fresh perspective on component design and more choice for us, the consumer.

Like most motherboard manufacturers, DFI are a Taiwanese company doing motherboards and also graphics cards. They are one of the few companies to take up a new Radeon license from ATi so we may see a DFI Radeon for review at Hexus at some point.

You can find them over at DFI.com and they have an impressive array of boards to choose from.

Rather than regurgitate information on the Intel i845 chipset itself which powers the NB72-SR , I'll point you to the hot off the press Asus P4B review where we had a closer look at the chipset itself.

DFI sell two versions of this board, the SC and the SR. The SR version that we take a look at today features on-board Promise IDE RAID while the SC is RAID-less. Other than that, the boards are identical so watch out for that when you are buying.

Lets take a look at DFI's take on what makes a good i845 board.

DFI's take on the Intel i845 chipset

Chipset
Intel 845 /ICH2

CPU Socket
Socket 478

CPU Supported
Intel Pentium 4 processor (Supports 400MHz system data bus)

Memory
3 x DIMM sockets. Max 3GB
Supports PC133 SDRAM

PCI IDE Dual PIO mode 4 EIDE channels up to 4 IDE devices
UltraDMA/100 transfer rate up to 100MB/sec

Super I/O 2 x NS16C550A compatible UARTs
1 x SPP/ECP/EPP parallel port

External Connectors
2 x USB, 2 x DB-9, 1 x DB-25, 1 x mouse,1 x keyboard, 1 x game/MIDI, 3 x Audio jacks

Internal Connectors
1 x connector for 2 additional external USB, 1x IrDA, 2 x IDE, 2 x RAID IDE, 1x Floppy, 3 x ATX power, 3 x fan, 1 x WOL,1 x WOR, 2 x internal audio connectors

Power Management
ACPI and OS direct power management
Wake-on event: RTC/USB Keyboard/Modem/LAN/Keyboard/Mouse

Hardware Monitor
System, processor temperature, voltage and fan speed

Expansion Slots
1 AGP slot (Supports 1.5V AGP 4x)
5 PCI slots
1 CNR slot

Audio On Board
AC`97 Audio codec

Other features
Promise IDE RAID 0 or 1, STR

BIOS
Award /2Mbit

Form Factors
ATX, 4 layers
30.5cm x 24.4cm
12.05 inch. x 9.6inch.

Like the Asus, DFI have included the headers and provided access to the AC'97 audio on the ICH2. With the headers being so cheap, it makes sense to include on board audio if a southbridge supports it since it it very easily disabled via the BIOS or jumper if you wish to use a PCI audio solution.

The DFI goes one step further and provided hardware IDE raid via the popular Promise controller. It supports RAID 0 and 1 in hardware but you can't create seperate arrays using both RAID types. If you have more than one array built then they both must use the same RAID type.

You lose a PCI slot on the DFI with only 5 slots but given the inclusion of hardware RAID and onboard sound, this isn't a shortcoming.

Size wise, the board is bigger than the P4B and uses the full 9 hole mount that the ATX spec gives us. It's not unwieldy but watch out if you have a mid or small sized case. Other than not supporting the Asus proprietary features like the smart card reader and iPanel the two boards are similar in feature set.

Layout and Installation



As mentioned above, watch out for the size of the board in some cases. It's a full 9-hole mount on the DFI.

Layout is very standard with everything in sensible places. Starting at the top left we have the Socket 478 enclosure. The DIMM slots feature next and are further away from the socket enclosure than on the Asus. Mounting the heatsink assembly can easily be done with memory modules installed since there is plenty of room.

Alongside DIMM3 we have the main ATX connector, vertically aligned with the AUX+12 socket alongside, flush with the right hand vertical edge of the PCB. The AUX+12 on the DFI is the standard style connector, very much like one half of the old AT style split connector. AUX+12 and the ATX12 connectors are mandatory on P4 board to provide the power that the new processor requires.

Underneath ATX and AUX+12 are the 2 ICH2 IDE ports and the floppy port, all vertically aligned. It's an orientation I prefer and especially so on the floppy connector which in this case is also up high on the board. While it's a personal preference of mine and this will vary from person to person and from case to case, I give DFI points for a sensible layout with regards to those 3 ports.

To the left, underneath the socket enclosure, we have a group of internal audio connectors for CD-ROM etc and the ATX12 connector. Being a P4 proprietary connector it doesn't feature on all power supplies, especially if your PSU is more than a year or so old. The Enermax 431W supply in the test machine is suitably equipped but do check your PSU if you are uncertain.

The Promise IDE RAID ports are in the right hand bottom corner and horizontally aligned. On the DFI there isn't space for the connectors to be vertically aligned without moving some components that live around that area such as the USB header and the Promise controller itself.

The slot layout is standard as always with everything moved down one backplane slot compared to the Asus. This means your AGP card will occupy a different backplane slot to most motherboards so if this is a problem for whatever reason, think ahead. Cooling solutions installed in a case are the main culprits for needing exact card placement so do check.

Finally you get 3 fan headers, one up high to the right of the socket enclosure, one to the top right of the DIMM's and one down with the ATX case connectors at the bottom right. Overall, a standard, solid layout. Nothing out of place, amiss or plain weird. Thumbs up.

BIOS, Manual and Bundle

The BIOS is an Award ACPI BIOS that provides CPU adjustment and everything you are used to in a P4 BIOS. Being an SDR board you can adjust SDR RAM timings to some extent with CAS delay adjust. You are free to choose either S1 or S3 for your ACPI implementation which correspond to Power on Suspend (POS) and Suspend to RAM (STR) respectively, something not always available on all boards.

One thing not available from BIOS or jumpers on the board is any kind of voltage adjustment for CPU, I/O or memory. While the Asus didn't provide I/O adjust, it gave us CPU adjust up to 1.85V and memory adjust up to 3.6V. While CPU voltage might be auto adjustable to 1.85V should the CPU need it, you can't adjust the setting manually. This hindered my overclocking attempts on the board as you'll see later. Not a big deal but likely to put the enthusiast off.

The manual is one of the more impressive that I've seen. It's a large, clear and easy to read manual covering everything needed to setup the board properly. The BIOS especially is covered very clearly. No complaints as far as the manual is concerned and it covers both versions of the board with a section on the Promise controller.

The bundle is just average. No external USB ports were supplied and while it's not always obvious to tell with review samples, it did arrives in the full retail box so I'll hazard a guess it'll be the same when buying this board off the shelf. The box is an attractive design.

Performance

We'll do exactly the same marks as we did with the P4B, a pair of 2Ghz runs at 20 x 100 and 15 x 133. DFI make no explicit claims in their literature about the NB72-SR supporting the new Northwood CPU's or 133Mhz FSB. However given that the board has a 100/133 jumper we presume it does and we'll test out 133Mhz FSB. We'll also test the overclockability of the board and see how we get on without any voltage adjustment.

Test System

• DFI NB72-SR Intel i845 Socket 478 Motherboard
• Intel 2.0GHz ES Processor
• 1 x 512Mb CAS2 PC133 Module
• MSI G3Ti500 Pro GeForce3 Ti500 Graphics Card
• Adaptec 39160 PCI SCSI Dual Channel U160 controller
• 2 x 73Gb Seagate Cheetah U160 10,000rpm SCSI disks
• Plextor 12/10/32S SCSI CDRW
• Creative 12x IDE DVD

• Windows XP Professional Build 2600.xpclient.010817-1148
• DetonatorXP 22.40 NVIDIA drivers
• Aquamark v2.3
• Quake3 v1.30
• POVRay v3.1g.msvc.unofficial-win32 dated 28 August 2001
• 3DMark 2001 Professional
• SiSoftware Sandra v8.53.1.11

Again, just like the Asus, WindowsXP needed no INF update. i845 support is build right into WindowsXP including support for all ICH2 functions like the audio. Only the Ti500 and Promise needed drivers to get themselves up and running.

Performance at 20 x 100

As usual, really just to verify that the CPU works correctly on the host chipset and that memory settings are configured correctly, we run Sandra's 3 most interesting benchmarks, CPU Arithmetic that tests raw ALU and FPU performance, CPU Multimedia that tests the SIMD instruction sets on the host CPU and the Memory Bandwidth test that tests, you guessed it, memory bandwidth. They give us visual indications that the core elements of a motherboard are working correctly.

CPU Arithmetic Benchmark


CPU Multimedia Benchmark


Memory Bandwidth Benchmark


The 2 CPU benchmarks give near identical results to the Asus P4B. Everything is ship shape as far as the CPU is concerned. The memory benchmark has an identical ALU score and a few points higher on the FPU. Identical performance from identical chipsets running the memory at identical settings. Nothing to choose between them here.

A quick run of POVRay usually confirms the findings from the 2 CPU benchmarks we run in Sandra.

P3 Binary


P4 Binary


Usual results from the two binaries are ~55s and ~41s respectively. While the P4 binary result is in our expected range, try as I might I couldn't get the P3 binary result anywhere near 55s. I ran it 5 times getting 64s each time, rebooted and ran it 5 times again, getting 64 each and every time. The reboot was just incase something had gone amiss after running 3DMark which I ran right before doing the POVRay runs. However the P3 binary results are erroneous for some reason. Very strange and not something I can readily explain.

Lets see if our 3DMark run is also out of the ordinary.



The Asus scored 43 marks more. Asus sneakily runs its FSB on the P4B ever so slightly above spec at 100Mhz. WCPUID confirms an FSB of 100.9/101 depending on the refresh. This small increase and changes in testing conditions are what we can attribute to the loss of marks on the DFI.

Aquamark should also confirm this theory and show ever so slightly slower numbers on the DFI.

AquaNox - AquaMark V2.3 2001/11/20 - (c) Massive Development
------------------------------------------------------------

RESOLUTION : 1024x768x32
FSAA       : NONE
PIXELSHADER: YES
TEXTURE    : 24MB
SECONDS    : 50.27 
......
MIN FPS    : 31.0
MAX FPS    : 63.8
AVG FPS    : 44.9

0.2fps is too close for comfort. Testing conditions only as the reason for this I'm afraid. The DFI is doing everything right here. All benchmarks bar the weird POVRay P3 binary result are within the smallest margin when comparing the Asus and the DFI.

Performance at 2Ghz (15 x 133)

The DFI uses a jumper system to switch between 100Mhz and 133Mhz FSB. Setting for 15 x 133 was a case of dropping to 15x multi at 100Mhz FSB, booting, turning off and moving the jumper to the 133Mhz position. Upon rebooting I had 15 x 133 and a successful boot into the OS. Running the Sandra memory benchmark revealed some slightly bizarre results. Check the following screenshot.



As you can see, something is slightly wrong. At 133Mhz memory clock, the maximum bandwidth available from an SDR system is 1.06Gb/sec. We are clearly exceeding that here. For the reason why, take a closer look at the screenshot, especially the memory bank information. Incase our resized screenshot it too unclear, here's a blowup of the section.



The memory stick is running at 177Mhz, giving an estimated 1.416Gb/sec maximum memory bandwidth according to Sandra. What's happened is that the DFI has decided to set the memory bank to HOSTCLK + PCICLK. PCI at 133Mhz on the DFI board is using a 1/3 divider for 44Mhz PCI. So the board is running quite a fair bit out of spec at 133Mhz FSB with the memory at 177Mhz (very lucky that the 512Mb PC133 stick did 177Mhz at stock memory voltage, CAS2-3-3!) and PCI running at 44Mhz.

Doing a double check in the BIOS and making sure the board was set to auto-detect the PCI speed and that I wasn't setting any funny memory settings except Auto didn't change anything. The board steadfastly ran at 1/3 PCI divider and HOSTCLK + PCICLK for the memory speed.

We can assume from this that the board doesn't explicitly support 133Mhz FSB due to the out of spec PCI and the addition of the PCI clock to the FSB for the memory speed. We can therefore assume that explicit support for Northwood CPU's is out too. This is the sort of thing that can be fixed in the BIOS however since we know the Asus does the correct speeds. Hopefully DFI will sort a BIOS update for proper 133Mhz support.

We'll skip over the CPU benchmarks for Sandra since they are unremarkable. Check the original 20 x 100 results to see what the screenshots would have looked like.

POVRay next.

P3 Binary


P4 Binary


We're still getting our strange P3 Binary result and an identical P4 Binary result. Again, the P4 Binary result is fine and I can't explain the P3 Binary result at all! A run of around 61 seconds is typical of a P4 processor at 1.8Ghz.

3DMark is the next of our benchmarks at 133fsb. Being at 177Mhz memory clock it turned out to be impossible to get a clean run of 3DMark without some textures being corrupted either mid test or before the test even ran. 3DMark would either quit to desktop or quit the running scene with an error regarding texture corruption. Reducing the latencies to 3-3-3 made no difference. It seems that at 177Mhz, the memory wasn't happy being a source of texture data in 3DMark.

Aquamark however ran without a problem.

AquaNox - AquaMark V2.3 2001/11/20 - (c) Massive Development
------------------------------------------------------------

RESOLUTION : 1024x768x32
FSAA       : NONE
PIXELSHADER: YES
TEXTURE    : 24MB
SECONDS    : 44.71 
.....
MIN FPS    : 32.4
MAX FPS    : 67.3
AVG FPS    : 50.5

A result of 47.3 on the Asus shows the effect that our 177Mhz memory speed has on the results. The P4 processor finally has a few more Mb/sec from the memory subsystem which it craves and uses to good effect here. 50.5 is the quickest Aquamark score yet seen in a Hexus review, all from a lowly SDR chipset board.

Overclocked performance

Initial thoughts on overclocking the DFI were good. Before opening the box I was hoping to match the Asus's overclocking exploits and give it a good run for it's money. However, 2 things stopped me from doing any benchmarking at plus 133Mhz FSB. Firstly, the lack of memory voltage adjustment and secondly the fact that memory speed at 133Mhz FSB was a rather high 177Mhz. A big hand to the memory stick for managing to boot at 177Mhz with no extra volts since 134Mhz was a no go. We can tell at 177Mhz the memory is right on the limit due to it not running 3DMark successfully.

However I was able to run a few benchmarks at 16 x 133 (2133Mhz). Remember stock CPU and memory volts here so it's a decent overclock.

Sandra CPU Arithmetic Benchmark


Sandra CPU Multimedia Benchmark


Sandra Memory Bandwidth Benchmark


Nothing exciting except again being able to run off the memory bench at 177Mhz. Some increase on the memory benchmark that we can attribute to the 133Mhz in CPU speed and the 2 CPU benchmarks run as we'd expect giving normal scores for a 2.133Ghz Pentium 4 CPU.

The P3 binary of POVRay would crash at 27% into the render everytime at 2133Mhz. The crash was graceful with WindowsXP catching it and offering to send a bug report to Microsoft HQ. The P4 binary ran without a hitch.



39 seconds for the run slots in nicely between the 2Ghz result and the 36 second run by the P4B at 2271Mhz. The successful run of the P4 binary shows that the CPU is on the edge too. More Vcore is needed for total stability at this speed but the DFI can't provide it since you can't adjust Vcore in the BIOS or via jumper.

Aquamark also ran without a problem, texturing fine from system memory.

AquaNox - AquaMark V2.3 2001/11/20 - (c) Massive Development
------------------------------------------------------------

RESOLUTION : 1024x768x32
FSAA       : NONE
PIXELSHADER: YES
TEXTURE    : 24MB
SECONDS    : 44.14 
.....
MIN FPS    : 34.5
MAX FPS    : 67.5
AVG FPS    : 51.1

A 0.6fps increase isn't earth shattering and suggests that Aquamark responds quite well to card clocks. This makes sense being heavy on the pixel and vertex shaders on full DX8-class hardware. Increasing card clocks, especially the GPU core which increases the speed of the nFiniteFX engine which implements the shader architecture will show a healthy increase in Aquamark performance.

Finally, after giving up on getting 3DMark to run, I went back to it just before finishing off this review. Here's the result.



I wasn't able to run 3DMark on the Asus at 2271Mhz however the score should be similar to this one. The extra clock speed on the Asus should have balanced out the higher memory bandwidth the DFI has at 177Mhz. Not bad from an SDR P4 system however the need for a ~£600+ 2Ghz ES processor, themselves very rare, and a state of the art graphics card means that if you can afford that kind of money to get those items, you can afford i850 + RAMBUS or one of the P4 DDR boards. This score will not be typical on such a motherboard.

Overall, not an overclockers board at all due to lack of any voltage adjustment and the PCI and memory speed issues. 133Mhz is very rare on locked P4's due to the large increases in CPU speed you get at the high multipliers on the P4's. I've heard rumours of new stepping 1.5Ghz P4's that share the 2.0Ghz P4 stepping. This suggests nearly 2.0Ghz is obtainable at FSB's approaching 133Mhz. Something for a brave person to try perhaps.

Conclusion

On paper, the DFI is a similar beast to the Asus. Identical chipset and similar features with nothing much to seperate them. Support for extra proprietary options here, a PCI slot there. Too close to call. A closer look in our reviews shows a different story however. The DFI is a rock solid implementation of the i845 chipset. The layout is fine, performance is where it should be at stock clocks and I didn't experience a single BSOD during any of the testing. Overall a very pleasant board to use and test.

Performance is unremarkable due to the chosen chipset but again it achieves Intel's main goal of getting the P4 to the masses by means of low cost. DFI get to add a solid string to their bow with this board and here's hoping it makes its way into more than a few OEM systems.

The only criticisms you can legitimately level at the board are lack of voltage adjust and the clock issues when you increase the FSB. All of these might possibly be fixable with say an overclockers BIOS. It would be interesting to test the board with a BIOS that allows voltage adjust and fixes the divider and memory clock issues we saw.

However given DFI's target audience with this board we can argue that those things aren't needed. It's designed to go into a system, run at 100Mhz FSB with stock clocked memory and run solidly at those speeds. This is an OEM board at heart I feel.

At the moment in the UK Avanti are the sole distributor for DFI boards and distribute to a small number of places. GlobalDirect list the non-RAID version, the SC, at £109.99 which is cheap for a Socket 478 board. The RAID version shouldn't be much more. However no other site on the Avanti list stocks either version. DFI's penetration in the UK is tiny which is a shame. It's always nice to have more choice when looking for components. Hopefully something will be done about this in the future.

Overall the feature set is fine and I'm a fan of the ICH2 AC'97 audio provided by the board. Unremarkable performance but extremely solid are the keywords to describe things here. Hopefully we'll see more from DFI at Hexus in the future!