Gigabyte 8N-SLI Royal (nVidia nForce4) SLI Intel Edition

Gigabyte 8N-SLI
Chipset	nVidia nForce4 SLI Intel Edition
Processor	Intel Prescott/Gallatin/Smithfield (FSB 200/266MHz)
Memory	DDR2 400/533/667
HDD	2x UltraDMA/133(RAID) RAID - 1x UltraDMA/133 4x SerialATA II (RAID) 2x SerialATA II (RAID)
Additional	3 IEEE-1394b Audio AC'92 v2.3 10 USB 2.0 2x Gigabit Ethernet LAN
Price:	N/A

The first version of the nVidia nForce chipset was released far back in mid-2001. That time, released was the only version for AMD SocketA platform which at the consumer qualities was somehow inferior to the VIA KT333 chipset (appeared a bit later) but surpassed KT266A. At the same time, from the technological viewpoint the chipset offered a number of strong points. First, when overclocked, it blocked the PCI bus at the nominal frequency 33 MHz. Secondly, the memory controller was dual-channel, which allowed attaining a very high bandwidth = 6.4 GB/s. But that didn't give any essential advantages to the AMD Athlon architecture since the bandwidth of the processor bus was restricted to 3.2 GB/s. But to the Intel Pentium4 platform that would have made possible to achieve simply fantastic results in terms of performance. That time, Intel's best chipset was i845E that offered support for the DDR PC2100 memory (released late in 2001), and the bandwidth of the processor bus was not balanced by the memory bandwidth. This balance appeared only in 2003 when Intel released its E7205 Granite Bay chipset with the dual-channel memory controller.

Of course, Intel's management could not afford to let nVidia chipset hit the market versus which all then existing chipsets would have looked like toys. As a result, a license for the Pentium4 bus was denied to nVidia, so the company had to submit with that. Interestingly, VIA also was not given a license for the processor bus, but continued with "semi-bootleg" production of chipsets. It ended up in that sanctions were imposed upon the company so VIA's chipset business almost withered away. And in no earlier than 2004 Intel supported half-dead VIA through issuing the above license (which actually hasn't brought any benefit in the end so far).

As regards nVidia chipsets, the company had been building up its efforts which finally brought success. The nForce2 chipset proved to be the most popular platform among PC enthusiasts (despite the problems with reliable storage of BIOS settings during overclocking), and with the release of Athlon64 processors the company first produced nForce3 followed by nForce4 that offered support for the PCI Express and SLI.

In my view, it is just the SLI that was the last drop that compelled Intel to conclude a cross-licensing agreement with nVidia. Immediately afterwards, there appeared numerous rumors in the Internet about the forthcoming release of nForce chipset for Intel processors. De facto, the chipset was announced during the CeBIT in March. That time, nVidia held a press conference where a new chipset was presented, and the leading motherboard manufacturers displayed their produce. By the way, the chipset was dubbed nForce4 Intel Edition.

nForce4 Intel Edition Chipset

If we partition the nForce4 chipset for AMD processors into the north bridge and south bridge areas, we get an amusing picture: numerous features are fully gathered in the south bridge, whereas the north bridge offers merely support for 16 channels of the PCI Express. We won't find the memory controller at all, since it is embedded in the Athlon64 processor. Therefore, engineers at nVidia released the nForce4 as a single chip.

With the nForce4 Intel Edition version, that trick hasn't worked though: Pentium4 has no integrated memory controller, so engineers at nVidia had to release the chipset in its traditional make (north bridge + south bridge). Traditionally, they were dubbed SPP (System Platform Processor)

and MCP (Media and Communications Processor).

The bridges are interlinked with the 800 MHz HyperTransport bus.

The capabilities of the south bridge for AMD and Intel versions are fully coincident: support for the PCI Express x1 (4 channels), ParallelATA-3366100133 (2 channels), SerialATA II (4 channels), PCI Masters (5), USB 2.0 (10), AC'97 2.3, and Gigabit Ethernet.

Note that features like hardware Firewall as well as advanced capabilities at setting up RAID arrays (levels 0, 1, 5 and 0+1 for both SerialATA and ParallelATA, including the shared operation mode). Note the new operation mode, the RAID 5. It is some sort of Intel MatrixRAID analog that combines the benefits of both RAID 0 and RAID 1 modes. However, unlike the MatrixRAID, three (not two) hard disks are required to set up an array like that. At the same time, one third of each disk is allocated for the "system info" (i.e. unavailable to the user).

Both the nForce4 chipset for AMD version and that for Intel Edition offer two standalone controllers for disk storage units.

Based on that, the developers implemented features like RAID Morphing and Cross-Controller RAID. The former (RAID Morphing) substantially facilitates RAID array configuration. In particular, the user can add new disks into the RAID array "on the fly", and change the RAID level as well (e.g., convert RAID 1 into RAID 0).

The other feature, as is evident from the name (Cross-Controller RAID), allows merging both SerialATA and ParallelATA disks into a RAID array. By the way, a feature like that was already available in the nForce3 150 chipset (it even supported 3 channels of ParallelATA). On the whole, all the disk handling features are gathered in the common concept dubbed MediaShield. Also, remember that along with support for SerialATA II the nForce4 IE chipset supports NCQ (Native Command Queuing) hard disks, as well as the "hot swap" feature.

Personally, I did like it that nForce4 IE supports two ParallelATA channels versus one in the latest Intel chipsets with the south bridge ICH6(R). The thing is, two PATA hard disks and one PATA CD drive are used in the test station. Therefore, all these three devices can be plugged in to nForce4, whereas none to motherboards on the base of i925X(E) and i915P.

The north bridge of nForce4 Intel Edition supports the PCI Express (20 channels) which can be used for two video cards united by the SLI technology (16 channels are used). Besides, a dual-channel memory controller with support for DDR2-400,533,667 (bandwidth = 10.6 GB/s) is implemented in the north bridge. As regards support for the DDR-I memory, we still haven't got precise data yet. It is quite likely to appear in the cheaper versions of nForce4 IE (e.g. an analog to nForce4X).

On the whole, the flow block of nForce4 SLI Intel Edition chipset looks like this:

Let's look into the memory controller separately:

To reduce the costs of accessing the memory, engineers at nVidia implemented QuickSync and DASP 3.0 blocks. The DASP (Dynamic Adaptive Speculative Pre-Processor) appears to be a preprocessor with cache memory and offers quite powerful predictive features and data-prefetching from memory. This has resulted in reduced memory latency or time taken for accessing the required data.

The second block, QuickSync, is designed for memory optimization in asynchronous modes. Besides, the specific trait of the memory controller (they are two, actually :) in nForce4 chipset is that data and instructions are fed to each DIMM module via a dedicated line. Therefore, the chipset developers have gained more freedom at choosing the chipset operation algorithms. In particular, high stability at both high frequencies and in asynchronous modes was attained through enabling the "1T Command Rate" feature. According to some reports, boards on the base of nForce4 IE were able to operate with DDR-2 modules with absolutely incredible latency timings 3-2-8-2 (mentioned were memory modules made by Corsair).

All these improvements should result in that the overall performance of the nForce4 IE can be on par with high-end chipsets of other Intel chipset manufacturers. Moreover, two 64-bit memory controllers (running together in the 128-bit mode) provide more flexible operation with various combination of memory modules. In particular, it is quite possible that two 256 MB modules at one memory channel with one 512MB module on the other can be installed (once we get the chance, we'll certainly experiment with these modes). Note that something similar will be implemented in the new generation of Intel chipsets but their release is scheduled at a later time.

The maximum number of memory slots makes four. At the same time, the maximum memory capacity supported by the chipset is 16 GB! For reference - the maximum memory capacity in Intel 955X (not yet released officially) makes 8 GB.

Even having not started tests of nForce4 IE, we can make the first conclusions. At the expansion capabilities, the nVidia chipset is an absolute leader and no other chipsets are comparable to it. This situation will persist until the release of Intel's new-generation chipsets (i955X and i945P). But even after their release, nForce4 will have a powerful trump - support for the SLI technology. It is also quite probable that the SLI will be supported by Intel chipsets (as per the cross-licensing agreement Intel has acquired nVidia's developments in this field).

Now, they are talking about the consolidation of four chips in the SLI (which means four video cards or, more likely, two double video cards). As a result, in the nearest future a high-end computer may look like this: a dual-core Smithfield processor (with support for HyperThreading it is able executing four command instruction threads simultaneously) and four video cores consolidated within the SLI! And of course a 1kW PSU to feed the locomotive :)).

But let's turn back to nForce4 Intel Edition and briefly list its specifications:

• Support for Prescott processors (of all steppings) and dual-core Smithfield processors
• Support for DDR2-400, DDR2-533 and DDR2-667 (memory capacity = 16GB).
• Support for the HyperTransport bus (1616; 1000 MHz)
• Support for PCI Express x16 (in the SLI mode - 2 x 8 channels)
• Support for the PCI Express x1 (four channels)
• IDE Interfaces: ATA -33/66/100/133 (2 channels)
• Support for SerialATA II (4 channels)
• Support for RAID arrays (levels 0,1,5, and 0+1) both for SerialATA and ParallelATA
• Number of PCI Masters: 5
• USB 2.0 ports: 10
• Support for AC'97 2.3
• Gigabit Ethernet on a dedicated channel
• Hardware Firewall

Today, we are testing the very first motherboard on the nForce4 Intel Edition chipset. Meet the 8N-SLI model by Gigabyte. Note that the manufacturer is the very first to present the press the specimens of boards on the most recent chipsets (e.g., a review of Gigabyte K8NXP-9 (nForce4 Ultra) which was released last year). However, there is another side of the coin: we often get à preliminary revision of the board having some shortcomings in the design and BIOS which have got to be regarded with understanding. In particular, the number of revision of the board to be tested is 0.2, i.e. boards of official revision may be different at both the performance and BIOS features and overclocking results.

In conclusion, let me note support for the nVidia nTune Performance Application utility offered by the nForce4 IE chipset. The program is aimed at fine-turning the motherboard, monitoring, tuning the memory latency timings, and overclocking. However, the latest version of the utility did not work "to the full": some features were unavailable. Evidently, programmers at nVidia are partly to blame for that, and developers at Gigabyte (again, it is a preliminary revision of the board and the BIOS version).

Gigabyte 8N-SLI Specifications

We are not reviewing the package bundle, since it is still an early test specimen of the board, and the package design may still be not ready yet. Also, the retail price is still unknown, which makes it difficult to draw conclusions on the package bundle.

However, note there is a Bluetooth module in a separate box

Besides, there is a DPS module as a separate bundle item aimed at enhancing the power supply to the motherboard:

Also note an additional bracket of quite tricky shape aimed at firm attachment of the SLI connector which in its turn links two video cards:

And here is the already mentioned SLI connector:

and also a SLI-terminator C19+RD1:

The thing is, the onboard SLI terminator is aimed at regular nVidia cards. The additional terminator is designed for joint operation of the motherboard and a dual-core Gigabyte RD1 video card which has somehow different scheme of PCI Express channel distribution.

Finally, the package bundle includes a fan for the north bridge. This approach is traditional for Gigabyte, which allows the user to choose between the active and passive methods of cooling.

Gigabyte 8N-SLI Board Layout

There aren't any issues while assembling the computer: the PCB design is quite traditional. In particular, there is no problem with blocking the latches of DIMM slots - the PCI Express x16 slot is positioned at quite a distance. Secondly, the main power connector is positioned on the board's bottom end. But the additional connectors (a 4-pin and that for Molex type) are installed closer to the board's center, which makes cabling an issue. Also, connectors for plugging in IDE cables are positioned with gaps and are colored differently. The ParallelATA connector of the additional RAID controller is positioned in parallel to the board's plane.

If we look at the processor socket, we can see the lack of Foxconn marking. That means alternative suppliers of LGA775 sockets are finally there.

In my view, that won't make high-end board cost cheaper :)

To cool the chipset, a passive radiator is used

on top of which you can place a 40 mm fan:

The heat emission from the north bridge proved to be so high that the radiator is unable to cope with it on its own, so the system wouldn't start. Therefore, installation of an additional fan (available in the bundle) is a must. In my view, that is a shortcoming of the current stepping of the nForce4 chipset, not that of the PCB design.

The cooler is plugged in to the two-pin NB_FAN connector. Besides it, there are the following connectors for fans: the PWR_FAN near the main power connector, SYS_FAN near the battery, and the 4-pin CPU_FAN near the processor socket.

There are four DIMM slots on the board, and the maximum memory capacity makes up 16 GB.

The scheme of switching the modules to the dual-channel mode is standard: the first two slots relate to the first channel, the other two - to the second (i.e. memory modules have to be installed into slots of the same color).

As regards the SLI technology, there are two PCI Express x16 slots onboard (in the SLI mode, they run as x8, that is enable 8 channels each), with an SLI terminator between them. The latter is designed for switching between the single-channel and the SLI mode.

There is an additional Molex connector nearby. Note that Gigabyte K8NXP-SLI (AMD) lacked an additional connector. But the Prescott system consumes more power than the AMD platform. So, stability and quality is power supply are very important for the board built on the nForce4 Intel Edition chipset: that resulted in the emergence of an additional connector.

Nevertheless, we used two GeForce 6600GT video cards in the SLI mode without plugging in additional power supply connectors. There was nothing to complain about the stability during operation. But while operating high-end cards of the 6800GTUltra class, use of additional connectors is a must.

Also, there are two PCI Express x1 and two PCI slots onboard.

Expansion options

With the emergence of nVidia onto the market of chipsets for Intel processors, support for SerialATA-II appeared for the first time. Four respective connectors are installed in the bottom left corner of the board, are colored orange and offer quite tricky plastic limiters.

Two more SerialATA channels are implemented with a PDC20779 chip made by Promise. That same controller supports one more ParallelATA channel whose connector is positioned near the south bridge and directed in parallel to the board's plane.

Remember that nForce4 offers support for 2 ParallelATA channels, which makes it possible to plug in as many as 12 hard disks to Gigabyte 8N-SLI: 6 over the SerialATA (four of which over the SerialATA II), and 4 over the ParallelATA.

Then, there are two Gigabit LAN controllers. The first one is the 88E1111 chip made by Marvell (connected to the PCI bus):

The second one is the 88E8053 chip (connected to the PCI Express x1 bus) made by the same company.

Thus, we get two RJ45 connectors on the rear panel of the board at a time.

Besides the two LAN controllers, 8N-SLI uses two controllers to support the Firewire bus.

These are TSB82AA2 and TSB81BA3 chips made by Texas Instruments. Therefore, the board offers support for three IEEE1394b ports. All the ports are additional - i.e., they require that brackets be plugged in. Besides the Firewire, the board supports 10 USB 2.0 ports, four of which on the rear panel, with 6 more plugged in with brackets.

And the last thing to dwell on is the integrated audio. ALC850 chip is used as the codec, which supports output for 8 channels.

The board's rear panel looks like this:

Instead of one of the COM ports, there is an optical and coaxial SP-DIF output. The missing COM port is even not provided by the PCB design.

Note that Gigabyte 8N-SLI uses no jumpers. To reset the BIOS settings, close the CLR_CMOS connector near the battery.

Now on to the BIOS settings.

BIOS

The BIOS of Gigabyte 8N-SLI is built on the Award BIOS Phoenix version.

On pressing the "Ctrl+F1" key combinations, you get access to the memory latency timings gathered in the "MB Intelligent Tweaker" section, or "M.I.T" for brief:

Apart from a number of memory latency settings, the user can choose the memory frequency:

The frequency values are listed in the x4 form, which makes the choice a bit difficult. You can also change the multiplier of the HyperTransport bus:

We found no special issues of compatibility to memory modules. In particular, Kingmax DDR2-533 modules were running trouble-free.

Now move on to the section to do with the system monitoring.

The board keeps track of the processor's current temperature and the rotational speed of the three fans. In my view, that is an extremely scanty selection of sensors and monitoring features for high-end boards. Besides, the speed of the processor cooler can be set dependent on the CPU temperature with the Smart FAN feature.

Also, note the support for power-saving and security features embedded in the most recent steppings of Prescott and Prescott 2M cores.

In particular, that is support for the C1E mode

and the Execute Disable Bit feature (or NX-bit)

Also, notable is that the BIOS settings can be displayed in several languages:

Finally, the board offers support for the DualBIOS technology:

This technology allows experimenting safely with the BIOS, e.g. changing the POST screen image with the FaceWizard utility.

Overclocking and stability

Let's take a look at the power converter. It uses a 4-phase power scheme in which there are four 1000 mkF, two 1500 mkF, and five 560 mkF capacitors.

But if we enable the VRM_CONN slot and install the DPS module, then the power supply modules operates following the 8-phase scheme.

By the way, in fact the DPS module uses four 1000 mkF and two 560 mkF capacitors. Besides, it uses its own cooling system with a heat pipe and a copper radiator.

Despite the theoretical advantage of using the DPS, we found no benefits in practice. In particular, the board runs absolutely stably without the DPS module, and no improvements in overclocking results were found after installing it. Besides, installation of the DPS module makes the assembly operator's job a bit complicated.

Now on to the overclocking features.

First, the Gigabyte 8N-SLI allows adjusting the system bus speed within 100 to 325 MHz in 1 MHz increments. And programmers at Gigabyte use a quadruple value in selecting the FSB speed, which allows raising the speed in 0.25 MHz increments (which in fact has no practical sense).

The memory frequency is adjustable in the same way (see the previous page). Advanced users can set the memory frequency manually, but the beginner users can only make use of the Linked mode at which the frequency is raised automatically.

Another item is the feature for changing the processor's multiplier.

The adjustment range is within 14 to 18 (the maximum for the test processor P4 660) in 1 increments. The next item deals with raising voltage on the processor (Vcore) within 0.8375V to 1.6V in 0.0125V increments.

The next item deals with raising the memory voltage.

The Vmem voltage can be raised by +0.3V in 0.1V increments. Besides, the overclocker can increase voltage on the FSB by 0.3V in 0.1V increments,

on the PCI Express bus (within the same range),

and on the SerialATA bus (within the same range),

Finally, the user can adjust the PCI Express bus within 100 MHz to 148 MHz.

Now on to the practical overclocking. At that, the board was not fun at all - with Extreme Edition, stable operation was possible at frequencies not higher than 306MHz.

Once a 200MHz bus processor (Pentium4 660) is installed, the maximum stable FSB speed was 240 MHz. In fact, in view of the preliminary character of the board's revision (that applies to the chipset, to all appearances), the overclocking results proved unsatisfactory. On the same grounds, we decided not to criticize Gigabyte for that the EasyTune 5 utility failed to operate properly with the 8N-Royal board.

Going on with the topic of software-driven overclocking, we note that the proprietary utility nVidia nTune (version 2.05.09) was able to recognize the chipset and was running quite stably on our board in question. In particular, the feature for adjusting the FSB speeds, memory frequency and memory latency timings was running properly. However, some other features like nTune Monitor did not work.

Our board offers quite interesting feature "Robust Graphics Booster" which should have somehow boosted the performance of the graphic subsystem.

But in practice we found no effect of enabling the feature (perhaps the situation will change in future versions of the BIOS).

Among Gigabyte's advanced technologies, there is "M.I.B. 2" aimed at automated optimization for memory settings. However, in the current BIOS version it hasn't been implemented yet.

Performance

Interestingly, Gigabyte 8N-SLI is quite precise at setting the nominal FSB speed:

As a contender to our board in question we used Asus P5AD2-E Premium (Intel 925XE).

In our test setup, we used the following hardware:

Let's first take a look at the results of synthetic benchmarks.

Now on to the gaming benchmarks (i.e. real-world applications).

Remarkably, in FarCry the nForce4 IE board demonstrated a substantial superiority (which is still hard to explain).

Kb/s. the more the better

Note separately that the Asus board built on the i925XE chipset overstated the FSB speed by 2 MHz, which gives additional superiority of nForce4 IE over i925XE at the chipset speed.

Final Words

Let's first draw conclusions regarding the nForce4 SLI Intel Edition chipset. No doubt, this product will add much of "fresh new blood" onto the market of Intel Pentium4 processors and will make this platform more popular. The main "pro" arguments in support of this statement are: very high performance (as high as that for i925XE-based motherboards), excellent expansion capabilities which surpass the features of ICH6, as well as support for the SLI technology. Of special mention is that the operating performance in running two video cards in the SLI mode makes almost no difference from that in nF4 SLI for the AMD platform (in percentage).

The new chipset is not free of shortcomings either. In particular, there are problems with overheating the north bridge, which is a negative point in terms of stability of the current stepping in nF4 IE.

Finally, questions on the prices for this chipset are rising, as well as regarding nVidia's plans for producing the "lite" versions of the chipset and versions with the video core integrated. But the most important conclusion is this - by its appeal, boards on the base of the nForce4 IE look much better than all the remaining boards for Intel processors (and this situation will not change until the release of i955X and i945P chipsets).

Secondly, here are the findings on the Gigabyte 8N-SLI. Of note is that the company engineers produced a really good product of great capabilities. In particular, using additional controllers they were able to expand the already rich capabilities of the nForce4 IE chipset. The list of overclocking features is quite impressive, and even the preliminary version of the board (its number is 0.2) shows pretty decent results (FSB>300 MHz). Anyway, there are some rough edges both in the implementation of the BIOS features and in the system functioning in general (which will most likely be fixed in new BIOS versions).

So far, the retail price is not defined yet. But the balance of these two parameters will be on par with other motherboard manufacturers.

Conclusion

• Excellent stability and performance;
• Support for the SLI (two PCI Express x8x16 slots);
• Support for SerialATA (4 channels of SATA II; 2 SATA II channels - additional Promise controller; RAID);
• Support for ParallelATA (2 nForce4 channels, 1 channel - additional Promise controller);
• Integrated 8-channel audio and two Gigabit LAN controllers;
• Support for the USB2.0 (10 ports) and IEEE-1394b (3 ports);
• A pack of Gigabyte proprietary technologies: (DualBIOS, Smart FAN, @BIOS, FaceWizard);
• Rich package bundle (including a WiFi module).

• Compulsory cooling of the board's north bridge

• Powerful overclocking features and quite decent results.
• DPS (Dual Power System) technology

Join the conference with all your questions, remarks and suggestions.