NVIDIA nForce 500 Chipset Family (Socket AM2)

The NVIDIA nForce 500 MCP

In the past few weeks, we've previewed a few choice motherboards from ASUS, MSI and Gigabyte showcasing not only the new AMD Socket AM2 but also a taste of NVIDIA's next generation core logic chipsets the nForce 570 and nForce 590 SLI. Since we were then bound under NDA, we could not divulge any details regarding the new features and functionality of the chipset besides what could be physically seen on the boards itself. If you were excited to have seen glimpses of the motherboards, you've seen nothing yet. NVIDIA has removed the veil surrounding the nForce 500 series of chipsets and we'll be taking an in depth look at the technologies that make up the nForce 500 Media Communication Processors (MCP).

NVIDIA has refreshed its entire nForce core logic line-up and the nForce 500 series is looking to extend NVIDIA's lead in the chipset market. The nForce 500 chipsets will initially be launched for the AMD Socket AM2 platform with four different models: The nForce 590 SLI at the highest end with dual x16 PCIe SLI and additional performance features, nForce 570 SLI with toned down features and standard dual x8 PCIe SLI support, nForce 570 Ultra with single GPU support and the stripped-down mainstream oriented nForce 550. The following feature comparison table shows the major differences between the MCPs.

Looking at the table, you'd notice that the feature scaling and market positioning of the nForce 500 series is very similar to the current nForce 4. The nForce 500 MCPs still retain a 16-bit bi-directional HyperTransport link operating at 200MHz with a default link speed of 1000MHz at a 5x HTT multiplier. NVIDIA maintains their single chip design for most of the nForce 500 MCPs though the nForce 590 SLI will continue to use an SPP PCI Express bridge chip to deliver the extra PCIe lanes. What is interesting though is that the nForce 590 SLI has a total of 46 PCI Express lanes, six more than the nForce4 SLI X16, which incidentally explains how the Gigabyte GA-M59SLI-S5 (you can see the preview ) is fitted with three PCIe x16 slots.

The rest of the nForce 500 MCP's updated features that we have already seen such as dual Ethernet MACs, six SATA 3.0Gbps channels are available. It is also nice to see NVIDIA embracing better onboard audio with Azalia HD Audio standard for the nForce 500, which means we can say good bye to crappy AC'97 for good. However, we'd like to draw attention to the mainstream nForce 550, which does not carry any of the enhancements found in the rest of the series. As far as specifications are concerned, the nForce 550 just seems to be a re-badged nForce4 Ultra for Socket AM2 with HD Audio support.

Besides sporting updated hardware components and CPU support, the main highlight of the nForce 500 series is actually a number of proprietary enhancements made into the core logic itself. The main features we'll be covering are NVIDIA's new LinkBoost, FirstPacket and SLI-Ready Memory technologies. We'll also look into the nForce 500 MCPs updated DualNet network functionality.

nForce 500 MCP Block Diagrams

LinkBoost

LinkBoost is an nForce 590 SLI exclusive feature that simply put, will automatically boost the SPP/MCP HyperTransport and PCIe x16 bus bandwidth to 125% if compatible hardware has been detected. LinkBoost works by applying a base frequency overclock, which means that the 200MHz HT link is boosted to 250MHz. With a 5x HT multiplier, it is effectively running the SPP/MCP link at 1250MHz, providing more bandwidth for demanding SLI operations at full PCIe x16 speeds.

LinkBoost will also overclock the PCIe bus by 25%, but it is isolated towards the two PCIe x16 lanes for increased bandwidth to the GPU. These clocks are controlled independently. Thus if you had only one GPU installed, only the particular PCIe x16 slot gets the 125MHz overclock and both PCIe x16 slots are overclocked if you have a dual GPU or SLI setup. The rest of the PCI Express bus outside the graphics lanes will still operate at a base 100MHz specification.

Yes, NVIDIA is basically guaranteeing a 25% bandwidth increase to the graphics bus subsystem, a feature clearly designed to appeal to gamers. However, there is currently a very strict restriction to this feature. Besides being only available on the nForce 590 SLI, LinkBoost will only work when a GeForce 7900 GTX card is detected by the BIOS. Thus, this feature is basically limited to the repulsively rich, die hard gamers or the repulsively rich die hard gamers.

Now before you write off LinkBoost as a marketing ploy, it is basically just an automated overclocking functionality built into the chipset. In future, there may be more graphics cards validated to support LinkBoost, but somehow we have a feeling that it would only be tied with faster graphics cards and not the mainstream variants. In any case, it makes sense because only high quality settings with high antialiasing sampling levels would tax the graphics card and the system bus consecutively to be worthy of the extra bandwidth available. Enthusiasts will be quite happy to know NVIDIA provides very detailed BIOS options with incredible granularity so that LinkBoost can be disabled and there is no one stopping you from tweaking the HT link or PCIe bus speeds to your own hearts content the way it's meant to be tweaked - manually.

GPU Ex

There is a very little documented feature in the BIOS called GPU Ex that can be toggled enabled or disabled. When contacted, NVIDIA has revealed that it is actually another SLI performance optimization that works with the forthcoming ForceWare Release 90 drivers. What this particular feature actually does is still a mystery.

For this article, we've done some preliminary testing with both GPU Ex and LinkBoost enabled independantly and working in tandem. Running a battery of tests with Quake 4, 3DMark05 and 3DMark06 at 8xAA/16xAF with a pair of GeForce 7900 GTX, we managed to achieve an incredible performance boost of - nothing.

The average framerate differences with LinkBoost and GPU Ex enabled can be counted within a single digit, but bear in mind that we've only done simple tests that might not have fully taxed the GPU bandwith yet. We've also ran into some slight problems enabling SLI AA with the supplied ForceWare 91.27 drivers so we will cover this topic in further detail when we manage to perform more comprehensive tests.

SLI-Ready Memory

Before you start getting a headache just by looking at the title above and trying to visualize how in the world can memory be SLI ready, let us break this down for you. SLI-Ready memory modules is a term that the clever marketing people at NVIDIA have coined for their excellent implementation of a new memory profile type called Enhanced Performance Profile, or EPP for short. EPP is an extension to the existing JEDEC SPD specifications and is actually programmed into the DIMM EEPROMs. Memory modules that carry these EPP codes will have more advanced timing and performance information stored than the basic SPD timings.

So how does EPP actually work? Every DIMM module today has its own 256Byte EEPROM (Electrically-Erasable Programmable Read-Only Memory) where the SPD information is stored. However, the SPD only takes up the first 98Bytes leaving Bytes 99-256 literally free to be used. What NVIDIA has introduced is a new EPP profile that stores advanced timings including voltage, frequency and signal strength information that is more essential to overclocking and pushing the particular DIMM to its limits. EPP utilizes Bytes 99 to 127 of the EEPROM and comes in either a 'Full Performance' profile or 'Abbreviated Performance' profile. We will not go into too much detail on the actual profiles though the technically inclined can view the two different profile maps below for more information on what EPP stores in the EEPROM.

• EPP EEPROM Abbreviated Performance Profile Map
• EPP EEPROM Full Performance Profile Map

Just like how current motherboards detect and apply optimal SPD timings today, an nForce 590 SLI motherboard can detect and apply these advanced EPP timings to provide higher memory performance. Note that EPP does not only tweak memory timing parameters, but is an actual overclocking profile. When the nForce 590 SLI BIOS detects EPP settings, it will try to tweak the system according to options that have been selected by the user. As such, the SLI-Ready option can be disabled in the BIOS if users do not want to utilize the extended EPP functions. The following settings are available for SLI-Ready memory functionality.

CPU MAX – This option will tell the BIOS to try and match the EPP specifications through overclocking of the memory controller, HT link and CPU as well. This is most useful if you are using DIMM modules that are of higher frequency than the base FSB and you need to overclock the CPU to match with the memory speeds. Reading the EPP parameters off the DIMM, the nForce 590 SLI will be able to detect the maximum overclock capable for the particular DIMM. Depending on the EPP profile selected, the BIOS will then try its best to match the EPP profile in terms of timing and frequency. Of course, any overclocker will know that overclocking is a fickle art. No two hardware components are alike and your overclocking mileage may vary.

CPUOC 0% - As the options suggests, setting it to this option will tell the BIOS to try and increase memory performance while keeping the CPU at stock stock speeds. This is the safest option to use and will only prompt the BIOS to perform memory controller overclocking. The images below show a set of Corsair EPP DDR2-800 DIMMs used. With SLI-Ready Memory enabled and set to CPUOC 0%, the BIOS was able to overclock the memory to from 800MHz @ 12-4-4-4 to 932MHz @ 15-5-5-5. While the CPU is kept at about stock speeds, this parameter still fiddles with the CPU multiplier and HTT link speeds to achieve the higher memory overclock, but tries to keep the CPU clock within limits.

Other settings include CPUOC from 1%-6%, though the most impressive modes are the two described simply because of what the BIOS tries to do. This isn't your typical automatic overclocking which you see in many of today's motherboards. With EPP profiles and and SLI-Ready Memory BIOS, the nForce 590 SLI performs overclocking as close to what is possible through manual means, it is almost human and the level of intelligence is a little scary. This function just opens up care-free overclocking for many who want to achieve the best possible performance, but not necessarily the time to tweak.

Being NVIDIA's partner for SLI-Ready Memory, Corsair will be launching two 2GB DDR2 kits carrying the EPP profiles: TWIN2X2048-6400C4 800MHz low-latency DIMMs and the TWIN2X2048-8500C5 high-frequency 1066MHz DIMMs. As this article is being written, OCZ Technology has also announced that they will be jumping onto the SLI-Ready Memory bandwagon with a line of EPP memory modules as well.

We can really see EPP type DIMMs taking off in the future and while the nForce 590 SLI again benefits from this SLI-Ready Memory function, there is no stopping of any other chipset or manufacturer to program a BIOS to take advantage of the EPP profiles. Also if you happen to purchase EPP DIMMs in the future, you'd be rest assured that the piece of memory has a high overclocking potential, even if you do not have an SLI-Ready Memory capable motherboard or BIOS. Nifty features, but they all come at a cost that's more than the usual as well.

FirstPacket Technology

FisrtPacket technology is the first of the new networking enhancements that the nForce 500 MCPs carry. In NVIDIA's own words, FirstPacket is essentially packet prioritization. Think along the lines of D-Link's DGL-4300 GamerLounge Wireless 108G Gaming Router and you'd have a pretty good idea on what FirstPacket technology intends to do.

It is true that broadband is becoming more accessible and bandwidth limits are increasing progressively. However, pumping in more bandwidth isn't the only solution to a smoother networking experience. Basically, there are two types of applications in a network, the latency tolerant ones that benefit from increased bandwidth, like FTP or web-browser transfers where continuous full frames are sent and received. Then there are the applications that usually send small data packets and do not require much network bandwidth, but very sensitive to latency and reliant on synchronization between client and server, much like network games, VoIP applications or audio/video streaming. Most avid on-line gamers will usually tell you that they turn off almost all other network software (Instant Messaging, file sharing, e-mail) except those that are required by the game to play and communicate in order to get the best 'ping'.

This is essentially the same concept. In an idle system with little to no network traffic, latency is a non-issue since packets are sent as and when they are made available. However, if you were to play the same game with an FTP running, you will most probably run into increased latency and lag spikes, which can equate to a jittering effect in-game. This is because regular network hardware and software are unable to differentiate between latency tolerant and latency sensitive applications. They will all be treated in the same way and applications that send continuous large frames will choke up the network for latency sensitive applications.

How FirstPacket really works is to optimize the way applications utilize upstream bandwidth so as to give latency sensitive applications like games, VoIP and video streaming a smoother experience. NVIDIA's FirstPacket technology acts as the middle-man in the equation and virtually creates two data queues internally, the 'fast queue' and the 'slow queue'. This is very similar to the priority queues given to preferred customers at banks. The FirstPacket filter re-arrange the data queue giving priority to the data packets in the 'fast' queue so they go out first. Gamers will benefit most out of FirstPacket technology since a lower latency will effectively reduce 'lag' and allow them to react better. No longer will you bang your head with frustration as your '1337 skillz' are defeated by 'lag'.

Of course FirstPacket technology isn't only limited to the gaming crowd, the effects of reduced latency and increased application network performance will allow most users to multi-task in a network or on-line environment more effectively than before. File-sharing, video streaming and VoIP can concurrently be performed with lowered latency effects.

However unlike the aforementioned D-Link gaming router which analyzes gaming packets and prioritizes UDP/TCP and port traffic, NVIDIA's FirstPacket technology is more direct and simpler in implementation. Users are required to choose which applications are given access to the 'fast' queue and all network traffic generated by the said application benefits from FirstPacket. It doesn't actually worry about what data is sent, the type of packet involved or what ports are used.

FirstPacket may not be too advanced or intelligent in this sense, but it allows the user to manage their networks more efficiently. However, It's great when you're just worried about the one or two games you're really having problems with, but if you live and breathe a networked life, setting FirstPacket profiles for all the applications you have will initially be a chore. Good thing, it is a one time setup.

DualNet

NVIDIA has gradually improved the networking features of their core logic chipsets over the years, introducing the industry's only built-in Gigabit Ethernet and hardware firewall with the nForce 2, TCP/IP task offloading from the nForce3 onwards to full TCP/IP acceleration and hardware ActiveArmor Secure Networking Engine in the nForce4 and nForce4 professional.

The nForce 500 series will be the first consumer core logic chipset to feature two Gigabit Ethernet MACs on one chip. However, what NVIDIA calls their DualNet technology isn't just an excuse to one-up the competition in terms of numbers. The TCP/IP acceleration engine is available to both Gigabit links and it will be the first onboard consumer networking controller to feature high-end server-like capabilities such as teaming, load balancing and network fail-over. These features are only available on the nForce 590 SLI, 570 SLI and 570 Ultra. The nForce 550 only comes with a single Gigabit Ethernet MAC, much like the older nForce4s so it doesn't carry the DualNet features.

Teaming, Load Balancing and Fail-over

Both Gigabit Ethernet controllers within the nForce 500 series MCP can operate independently as separate ports or if you do need the bandwidth, they can be combined to form one fat two Gigabit pipe. This unique feature is called 'Teaming' (not to be confused with 'Bridging' where a PC with multiple network adapters can be used to connect different LAN segments). However, that is not all. An nForce 500 series MCP with teaming enabled is also capable of intelligent load balancing and fail-over features.

With teaming enabled, both the Gigabit Ethernet links still maintain their individual MAC addresses internally, though the teamed line share one IP address that is visible to the network. Due to this, the network driver is able to perform load balancing for both inbound and outbound traffic to optimize performance and utilization.

For outbound traffic, the network driver can assign the destination IP address to either one of the Gigabit links for transmission purposes and in the case that one link is idle, the driver dynamically adjusts the connections so that transfers are always balanced between the two links. Load balancing for inbound traffic is more straightforward and is performed through a steering distribution method. In the case of multiple server connections, the traffic from different servers are also load balanced and distributed within the two Gigabit links.

Lastly, the nForce 500 series MCP also supports fail-over functionality when teaming is enabled. If one of the Gigabit links malfunction, drop out or is disabled, the driver automatically switches over to the remaining link to maintain the network. If and when the disabled connection is reestablished, the driver will also be able to rebuild the teamed link and continue working.

Deactivated Armor

If you've looked at the specifications on the first page, you'd have noticed that NVIDIA has removed the ActiveArmor Firewall and Secure Networking Engine component of the MCP. After three generations of promoting a built-in firewall solution, they're suddenly none to keen on continuing down that path with the nForce 500 series. Apparently, the decision to drop ActiveArmor had something to do with Microsoft Windows Vista's updated networking and security capabilities. However, looking at the new networking features of the nForce 500 MCP, we'd take FirstPacket and DualNet over ActiveArmor any day so there is no love lost there.

NVIDIA MediaShield Storage

From our previews, we've seen that the new nForce 500 series will feature expanded storage support with up to six SATA 3.0Gbps channels. NVIDIA has also finally come around to increase the RAID capabilities of the controller (and about time too!). They have so far stuck with the basics of RAID 0 and 1 combinations, but the nForce 500 series will change that with the introduction of RAID 5 support. Note that this isn't an industry first (Intel's Matrix Storage can do RAID 5 and 10), but it is the first for the nForce chipset family. NVIDIA also seems to have followed Intel's lead in regards to older connections, as the nForce 500 loses one of its IDE channels so there is support for only two Ultra ATA devices max. Note that there's no IDE RAID capability built in anymore and cross-controller RAID involving both IDE and SATA drives are no longer supported as well. You lose some and you gain some, but since nForce 500 is a forward looking chipset, it was designed to go easy on the IDE side and bulk up what the SATA RAID controller can do.

The nForce 500 series SATA controller also has full NCQ support. Other MediaShield features that have made it into the nForce 500 series include RAID Morphing (on-the-fly RAID configuration changing), hot-plug support and the disk alert system where device problems or disk failures are graphically highlighted to ease troubleshooting. Most of these features were already present on the nForce4 series and were successfully replicated across the nForce 500 series as well.

HD Audio (Azalia)

Rather than focus on creating their own audio solution (which they did for the highly successful SoundStorm introduced in the nForce2), the entire line-up of nForce 500 MCPs will have updated audio support for Intel's Azalia HD Audio specifications.

nTune 5.0

NVIDIA has really beefed up their nTune utility for the nForce 500 series launch as the updated application has more or less undergone an overhaul. nTune now has access to a host of tweaking capabilities that it has become more or less redundant to mess around with the BIOS. The nTune utility is now a powerful performance analyzer and tweaking tool that you can access even more detailed functions than most BIOS' ever had.

The user interface is very accessible and while the sheer amount of options may be daunting to the initiate dabbling into overclocking, we're very sure enthusiasts will be very happy with this new version.

Final Thoughts

With AMD launching their Socket AM2 range of processors today, some might make the mistake of thinking that NVIDIA's concurrent release of the nForce 500 series is just a minor update to support the new CPU and its features. All in all, the nForce 500 series MCP chipsets offer some very significant enhancements to the current core logic chipset market with features that are ready to take advantage of Socket AM2, faster memory and advanced operating systems like the upcoming Windows Vista.

In terms of actual technology innovation, the nForce 500 doesn't really bring anything revolutionary to the table. There isn't anything that has not really been done before in some form or other. NVIDIA just seems to have taken a long hard look at the industry and implemented all the best features right onto the core logic chipset itself and built into it intelligence to boot. Motherboards based on the nForce 590 SLI will be so feature rich right out of the chip that it will become hard for motherboard manufacturers to really come up with more differentiating factors. Dual Gigabit LAN with server features, six SATA 3.0Gbps with multi RAID configuration, full SLI x16 graphics with a total of 48 PCI Express lanes and ultra comprehensive memory and system overclocking, what more can any enthusiast really want out of a motherboard, let alone the chipset.

However, if you consider yourself a power user, the only choice that you can go for 'IS' the nForce 590 SLI, which is beyond the reach of most users if one is to fully embrace everything it has to offer. Our only gripe is the way NVIDIA is promoting a set of performance enhancing options for the nForce 500 MCPs, but only making it available in the highest-end hardware. The two main performance features (LinkBoost and SLI-Ready Memory) are only available to the nForce 590 SLI, which is really a shame since the new DIMMs with EPP profiles can have a great potential and better take-up if a variety of motherboards support it. If EPP successfully goes mainstream (and we think it would given a span of time), we could possibly see better memory performance handling across the industry.

The nForce 500 series of chipsets will not only be locked to the AMD platform as today, NVIDIA also announced the availability of the nForce 590 SLI Intel Edition. Since the Intel platform relies upon a memory controller being built into the chipset, and seeing how much NVIDIA has done for the AMD64 platform that doesn't even have a memory controller on the board, NVIDIA has plenty of room to play around to deliver what might possibly be the 'ultimate' memory controller. After all, we've seen the nForce4 SLI Intel Edition's strong performance lead over Intel's own chipsets. Following which in the long run, though NVIDIA wouldn't comment on future plans, we expect the rest of the chipset versions to be 'Intelified' if the progression of the nForce4 is a good track record. This article will be followed up shortly with a performance review of the nForce 590 SLI chipset, so stay tuned.

Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.