A year after the nForce 600i was launched, NVIDIA comes up with an updated chipset series. This time it contains much fewer products (two chipsets versus 4+1 the last time), which can be easily explained: the nForce 700i does not have significant innovations, so the company updated only those products that needed the new features. Curiously, NVIDIA is the only company that previously offered a chipset consisting of a single chip, and now it offers another oddity: its desktop chipset includes three chips instead of traditional two (north and south bridges).

nForce 780i SLI

You'll understand why the third chip is added, if you have a look at the block diagram of the nForce 780i SLI:

So, the additional nForce 200 bridge is added for the PCI Express 2.0 interface. But why is this additional chip necessary at all? To answer this question, let's have a look at the list of the nForce 780i SLI characteristics and its differences from the nForce 680i SLI:

• Support for all Intel Celeron 400, Pentium Dual-Core, and Core 2 processors (it officially supports old Pentium 4/D CPUs, but this feature is not relevant anymore) FSB clock rate can officially reach 1333 MHz, there is also official support for future Intel processors on the Penryn core
• Dual channel DDR2-533/667/800 memory controller supporting up to four DIMM modules, up to 8 GB in total (without ECC) Support for SLI-Ready memory (with EPP) Integrated DASP 4.0 for hardware prefetch and data caching
• Bidirectional HyperTransport bus (1 GHz, 16 bit in each direction) to connect chipset bridges
• Three PCIEx16 graphics interfaces (two of them are based on the north bridge through nForce 200, the third interface is based on the south bridge) supporting 3-way SLI (x16+x16+x16), PCI-E ports on the north bridge support PCI Express 2.0
• PCIEx8 interface on the south bridge
• Another six PCI Express x1 interfaces (in total) from both bridges
• Up to five PCI devices
• Up to six Serial ATA ports for six SATA300 devices (SATA-II, the second generation of the standard) supporting NCQ, TCQ, and Hot-plug
• Up to two ATA133 devices (one channel)
• SATA RAID 0, 1, 0+1, and 5
• 2 x MAC controllers for 10/100/1000 Mbit/s network (Gigabit Ethernet) supporting such functions as unloading processor when processing TCP/IP traffic, traffic shaping, balancing the load, channel teaming and duplexing
• Up to ten USB 2.0 devices
• High Definition Audio support.

The new chipset uses the same south bridge, and the north bridge has been modified: instead of PCIEx16 (PCI-E 1.0), it has an interface to nForce 200 (which is used to add two PCIEx16 (PCI-E 2.0) or four PCIEx8 interfaces). We have the following question: what is the bandwidth of the interface between the nForce 780i SLI SPP and the nForce 200 (and are there any differences between the nForce 780i SLI SPP and the nForce 680i SLI SPP?) NVIDIA specifications give a vague answer to this question: the bandwidth is 4.5 GT/s per link, which is sufficient for maximum performance of PCI Express 2.0 graphics cards. We don't understand what "4.5 GT/s per link" means, because we are not sure what they mean by "links" (and how many links there are at all). The most logical explanation is that it's a raw data transfer rate in one direction through the PCIEx16 (1.0) port. In this case, the above question has only one answer.

Indeed, unidirectional bandwidth of a PCIEx16 (1.0) port is 4 GB/s (plus overheads), so the nForce 780i SLI SPP is just a renamed nForce 680i SLI SPP. There is also an indirect proof of our last statement (except for Internet rumors that asserted it from the very beginning). For example, [third-party] system monitoring tools detect the new nForce 780i SLI north bridge (which is not yet supported by them) as … nForce 680i SLI SPP. It means that the unique chipset ID in the new SPP hasn't been changed. Besides, only our assumption explains the necessity to add the nForce 200, which should support two "honest" PCI Express x16 (2.0) ports.

What are the consequences of this chipset architecture for common users? As a consequence, there is no practical use of such support for PCI Express 2.0 in the nForce 780i SLI: the necessary bus bandwidth (a graphics card reading data from system memory) will still be limited by the transfer rate of a single PCIEx16 (1.0) port to both PCIEx16 (2.0) ports, even though their total bandwidth is four times as high. On the other hand, the second version of PCI Express is hardly necessary for the existing graphics cards. Perhaps, NVIDIA is correct mentioning maximum performance of PCI Express 2.0 graphics cards.

Now let's analyze other interesting differences of the new chipset. There are two of them: full support for future Intel processors on the Penryn core (45-nm fabrication process) and support for 3-way SLI (three graphics cards operating in SLI mode). Both differences from older solutions are de-facto not differences at all. The 3-way SLI technology was announced prior to the nForce 700i. And it had been designed to be used with motherboards on the nForce 680i SLI at first. It was natural - this technology did not require special logic in a chipset, and motherboards on nForce 680i SLI were often equipped with three graphics slots (to say the truth, the third graphics slot in the nForce 600i was intended for a card to compute physics in games). The only advantage of the nForce 780i SLI is the x16 configuration for all three graphics slots (taking into account that PCI-E 2.0 is twice as fast, we get x32+x32+x16 versus x16+x16+x8 in the nForce 680i SLI). However, we've just discussed the reality of x16 in two slots. What concerns full support for Penryn and 1333 MHz FSB, they were promised in the times of the nForce 600i. Experiments of our colleagues prove that engineering samples of Penryn processors work fine at least on some prev-gen motherboards.

However, there is one more secret difference between the nForce 780i SLI and the nForce 680i SLI. It's caused by the new hot bridge, which will have to be placed under the same heat sink with the hot north bridge in new motherboards. As a results, a cooling system for this chipset must be either huge or noisy (at least, that was the case with the nForce 680i SLI). Don't forget about the south bridge, which was used as the north bridge in the 500-series chipsets from NVIDIA and was equipped with an imposing cooler. Attempts to design a passive cooling system for the nForce 700i will probably end just like in this Gigabyte motherboard, which represents the new top chipset in our today's review. By the way, you should take into account that the new chipset series from NVIDIA is intended for enthusiasts (and overclockers), and overclocking accompanied by raising chipset voltages increases requirements to a cooling system significantly.

nForce 750i SLI

The second chipset in the new series has the same new features (and architecture) as the nForce 780i SLI:

Indeed, it also uses the additional nForce 200 bridge, which provides PCI Express 2.0. And it also supports new processors from Intel. At the same time, the nForce 750i SLI, being the cheapest in the series, has some restrictions: SLI for two cards only in x8+x8 mode (considering that PCI-E 2.0 is faster, performance will be similar to full-speed x16+x16 in the old top chipsets), no support for EPP profiles, and the south bridge with fewer functions. Here are specifications of the nForce 750i SLI:

• Support for all Intel Celeron 400, Pentium Dual-Core, and Core 2 processors (it officially supports old Pentium 4/D CPUs, but this feature is not relevant anymore) FSB clock rate can officially reach 1333 MHz, there is also official support for future Intel processors on the Penryn core
• Dual channel DDR2-533/667/800 memory controller supporting up to four DIMM modules, up to 8 GB in total (without ECC) Integrated DASP 4.0 for hardware prefetch and data caching
• Bidirectional HyperTransport bus (1 GHz, 16 bit in each direction) to connect chipset bridges
• PCIEx16 and PCIEx8 graphics interfaces (from the north bridge through the nForce 200) supporting SLI (x8+x8), PCI-E slots support PCI Express 2.0
• Another two PCI Express x1 interfaces on the north bridge
• Up to 5 PCI devices
• Up to four Serial ATA ports for four SATA300 devices (SATA-II, the second generation of the standard) supporting NCQ, TCQ, and Hot-plug
• Up to four ATA133 devices (2 channels)
• SATA RAID 0, 1, 0+1, and 5
• MAC controller for a 10/100/1000 Mbit/s network (Gigabit Ethernet) supporting such functions as unloading processor when processing TCP/IP traffic, traffic shaping
• Up to eight USB 2.0 devices
• High Definition Audio support.

What concerns the south bridge, restrictions are typical of the latest NVIDIA products: fewer USB ports, PCI-E interfaces, and SATA ports; only one gigabit network controller. There are still two PATA channels for owners of hard drives with this interface. In conclusion, we traditionally send you to the nForce 500 description, because details and proprietary technologies of network and drive controllers have not changed since that time.

Performance tests

Testbed configuration:

• CPU: Intel Core 2 Duo E6600 (2.4 GHz, 1066 MHz FSB), Socket 775
• Motherboards:
  • Gigabyte 780SLI-DS5 (BIOS D2) on NVIDIA nForce 780i SLI
  • EVGA nForce 680i SLI (BIOS P17) on NVIDIA nForce 680i SLI (the link will open photos of the motherboard)
  • Gigabyte 965P-DQ6 (BIOS D25) on Intel P965

• Memory: 2 × 1 GB PC2-9136(DDR2-1142) DDR2 SDRAM DIMM Corsair Dominator (CM2X1024-9136C5D)
• Graphics card: ATI Radeon X1900 XTX 512 MB
• HDD: Seagate Barracuda 7200.7 (SATA), 7200 rpm

Software:

• OS and drivers:
  • Windows XP Professional SP2
  • DirectX 9.0c
  • Intel Chipset Drivers 8.0.1.1006
  • the latest version of NVIDIA drivers (9.46)
  • ATI Catalyst 5.2

• Test applications:
  • 7-Zip 4.10b
  • XviD 1.0.2 codec (29.08.2004)
  • Doom 3 (v1.0.1282)
  • FarCry (v1.1.3.1337)

Test results

Unfortunately, the motherboard from Gigabyte, representing the new chipset in this review, was not ready to demonstrate maximum performance in our rig. You can read the details in the review of this motherboard. We can only mention that the BIOS beta flashed in our sample (we are almost 100% certain that it's not a chipset problem) failed to support the memory mode we use in our tests. The DDR2-800 memory was naturally faster at 4-4-4-1T versus 5-6-6-2T. So even our minimal interest to the speed of the new chipset (considering its differences from the precursor) was not satisfied.

For form's sake, we'll publish several test results. However, you should keep in mind that the performance differences were caused by our testbed's configurations, not by chipset characteristics.

Conclusion

Before we draw conclusions about the new chipsets, we want to touch upon NVIDIA's plans for the nearest future. Here they are: the company will soon launch a new chipset in this series - nForce 790i SLI (approximately in Q1 2008). This chipset will not use additional chips (only two bridges), but it will still support PCI Express 2.0. But the memory type will change from DDR2 to DDR3, and we don't know how expedient this memory type will be in the beginning of the next year.

Now we can draw our conclusions. The new chipset series from NVIDIA offers a number of new technologies (for example, the ESA standard). But if we speak of the general situation on the market, the unique nature of the new product comes to nought. Support for Intel Penryn processors and 3-way SLI (relevant from the marketing point of view) can be provided not only by competing chipsets from Intel, but also by the previous solutions from NVIDIA (nForce 600i). And such support for PCI Express 2.0 (actually any support these days) will hardly persuade you to buy this product (on the other hand, Intel X38 offers sterling support for PCI-E 2.0). On the positive side, NVIDIA offers SLI. That is the nForce 700i indeed manages to offer a unique feature… However, we've heard rumors about official SLI support in future chipsets from Intel… So, we can recommend the new chipsets now only to a narrow group of users, who found all the words they wanted in our today's article. But we should repeat to these users that they will have to provide a good cooling system. The other users should wait till the next year. If you want to buy a motherboard right now, you should focus on Intel chipsets or the previous generation of chipsets from NVIDIA—nForce 600i.

Write a comment below. No registration needed!