Summary: With the arrival of Socket AM2 comes two new chipsets from ATI and NVIDIA, the CrossFire Xpress 3200 and nForce 500 family. In today's article we go over what's new with both chipsets as well as run them through our full suite of benchmarks to see how they perform. Does NVIDIA come out on top over ATI? Find out inside!
For past few years now, NVIDIA’s had a decisive advantage in the chipset market over not just ATI, but all chipset manufacturers in the AMD space. Take for example the first generation of PCI Express capable chipsets for the AMD platform. Both ATI and NVIDIA’s chipsets first shipped in the same general timeframe around the end of 2004, but it was NVIDIA’s nForce4 chipset that was perceived as the high-end part, with most of ATI’s XPRESS 200 chipset sales going to OEMs who opted for the ATI chipset due to the fact that they were the only ones to offer a PCI Express chipset with integrated graphics. Once NVIDIA provided their integrated solution, ATI’s XPRES 200 sales quickly dried up. nForce4’s dominance continued for all of 2005 and into 2006 as well.
Then about 2 months ago ATI dropped a bombshell on the chipset space with the launch of their CrossFire Xpress 3200 chipset. Not only did ATI finally have a viable answer to NVIDIA’s SLI, Xpress 3200 also delivered excellent performance and stunning overclocking capabilities, all in a package that ran cooler than NVIDIA’s latest nForce offerings. It was as if ATI built a chipset specifically for hardware enthusiasts who were into overclocking and tweaking their system’s performance. Sure, Xpress 3200 didn’t natively have all the frills and gadgets found in NVIDIA’s latest chipset offerings, but that didn’t stop motherboard manufacturers like ASUS from incorporating as many of those features as they could into their own Xpress 3200 motherboards.
With the arrival of CrossFire XPRESS 3200, ATI for the first time had a true high-end solution that was capable of taking on the latest and greatest from NVIDIA.
Now with the arrival of AMD’s brand new AM2 socket, both ATI and NVIDIA have a new platform to renew their battle on. ATI’s decided to stick largely with their tried and tested Xpress 3200 platform first launched in March, while NVIDIA’s come up with a new nForce chipset that’s a little more revised in comparison to the competition. We’ll start with ATI first since it’s roots are basically unchanged…
If you recall, CrossFire Xpress 3200 supports up to 40 lanes of PCI Express connectivity, enough to run two x16 PCI Express graphics slots. The other 8 lanes are split between ATI’s chip-to-chip interconnect and expansion, with 4 lanes going to ATI’s A-Link Xpress2 chipset interconnect, which connects the RD580 North Bridge to its SB600 South Bridge, while the remaining four lanes are devoted for expansion cards.
One trait of their RD580 North Bridge that ATI is particularly proud of is the fact that their RD580 North Bridge handles all 40 PCI Express lanes. ATI argues that NVIDIA’s architecture is less efficient due to the fact that the interconnect connecting NVIDIA’s North and South Bridge could become a bottleneck. We agree that ATI’s solution looks more elegant on paper, in reality though there have been no cases where NVIDIA’s architecture has presented a problem. In addition, ATI’s solution was due in part by necessity, by integrating all of the PCI Express lanes into the North Bridge, ATI’s motherboard partners can still provide full 16 lane operation to both graphics slots while using a third-party South Bridge from ULi.
Speaking of the South Bridge, ATI’s new SB600 chip is much improved in comparison to its predecessor. Whereas SB450 lacked support for even basic features like Gigabit Ethernet and 300MB/sec Serial ATA with NCQ, SB600 is stuffed with goodies.
First, ATI claims that SB600 has resolved the nagging USB performance issues that plagued SB450. In addition, ATI has finally added support for 300MB/sec Serial ATA transfer rates as well as Native Command Queuing. Up to four Serial ATA hard drives are supported. Finally, with SB600 ATI has integrated support for RAID Levels 0, 1, and 10.
Interestingly enough, ATI’s Xpress 3200 chipset omits an integrated networking engine. That’s right, ATI doesn’t even provide a basic 10/100 Ethernet controller with their Xpress 3200 chipset. Instead ATI relies on the motherboard manufacturers to implement whatever networking solution they wish. This makes things more flexible for the motherboard manufacturers, as they can choose whichever solution they’d like, but one downside is that this decision could lead to a wide variance when it comes to networking performance on various Xpress 3200 motherboards. For instance we wouldn’t be surprised to see a motherboard manufacturer or two drop a PCI-bound (rather than PCI Exress) GigE network controller on the unsuspecting public.
To see the power of ATI’s CrossFire Xpress 3200 AM2 chipset, we’re using the ATI reference RD580 Xpress3200 motherboard. We felt it was important to take a more in depth look at the ATI reference board, as it’s quite possible some of ATI’s motherboard partners may go into production with this design. Sapphire for instance has used ATI’s reference boards in the past for their final retail boards. Let’s start with the back of the reference board first.
The back panel of the reference board doesn’t have anything special, in fact it actually lacks some of the features found on many premium nForce4 and nForce 500 boards. There are only four rear USB ports, only one Ethernet jack and no optical out on the back. Again, keep in mind that this ATI board is reference only, we’re pretty sure ATI’s board partners have something up their sleeves to make their CrossFire Xpress 3200 boards stand out.
Here are the expansion slots for the reference board. There’s nothing much to see here, everything is quite standard. There’s enough space for video cards with dual slot cooling, you can also see a VIA Firewire controller. Towards the bottom corner is a useful diagnostic display for informing you of critical system messages, and a Molex connector. This connector is used to feed juice to the motherboard when Crossfire is enabled. The Molex connector is at the same spot here as on the NVIDIA nForce 590 reference board. This is a very strange position because the cables have to go over the video card to reach. Some power supplies have shorter cables, which may not reach the connector at all.
Here is the Realtek ALC880 codec. It provides the HD Azalia audio. To the right is the ITE 8712F chip that monitors system health, such as voltages and temperatures. You’ll have to excuse us for these two fairly blurry photographs.
As with NVIDIA, ATI provides six SATA ports on their reference motherboard. You can see the 24 pin ATX2.0 connector and floppy port. The four DDR2 240pin DIMM slots are present, but do not define the dual channel configurations. We’ll tell them to you instead. From top to bottom are DIMM slots 1-4. Slots 1 and 3 are Channel A, and slots 2 and 4 are Channel B.
ATI incorporates silent passive cooling on both bridges. The MOSFET/PWM area is also cooled. At the top left of the first picture you can spot a second Molex connector. This is for even more juice to the motherboard and is located quite conveniently.
NVIDIA has integrated quite a few new features into their nForce5 family of chipsets. Before we get into the specifics though, let’s look at the features list:
In the table above, you no doubt noticed lots of new buzzwords that have never been used by NVIDIA before. Words like LinkBoost, FirstPacket, and Teaming probably all sound pretty foreign to you, so we’ll go over all of the new technology NVIDIA is introducing with their nForce500 family first, before going over the specific features supported by each chipset in the nForce500 family.
One new feature NVIDIA is touting in nForce 590 SLI is known as LinkBoost. Once LinkBoost is activated, the nForce 590 automatically increases the bandwidth of the PCI Express Graphics slots (PEG, the x16 slots) by 25% to 125MHz. In addition, the speed of NVIDIA’s HyperTransport connection between the SPP and the MCP is increased from 200MHz to 250MHz, providing up to GB/sec between the two chips.
LinkBoost is a feature available only on the nForce 590. This is for two very good reasons. First of all, the nForce 590 is a dual chip configuration where each chip controls one x16 PCI Express slot, so this could theoretically create a bottleneck. LinkBoost is supposed to help alleviate this potential problem. (Of course, we should note that we haven’t run into any cases where NVIDIA’s decoupled PCI Express graphics lanes have created a bottleneck.) Secondly, LinkBoost is a premium feature, so it is only available on the premium class motherboards – the nForce 590.
When it comes to networking connectivity, NVIDIA’s new nForce chipsets certainly don’t disappoint. For starters you’ve got dual Gigabit Ethernet networking, a first for an NVIDIA chipset, previous nForce offerings only provided one GigE network connection, with the second being 10/100.
Each of the GigE controllers has its own TCP/IP offload engine. With TCP/IP Acceleration the chipset handles packet inspection and movement, offloading these duties from the CPU. Moving packets is a CPU and memory intensive process. TCP/IP Acceleration moves those processes to hardware, thus reducing CPU overhead. There is a drawback to this feature though. Software firewalls will be bypassed by this technology, rendering them useless. So unless you have a hardware firewall in your router, the CPU time saved is probably too little to risk the safety of your PC.
Another useful feature included in all versions of the nForce 500 except the nForce 550 is FirstPacket™ technology. FirstPacket prioritizes packets sent and received. Pictured above is a typical network. Normally, the gaming and other latency sensitive packets are treated the same as others, so these packets wait like any other packet from any other application. FirstPacket makes the network “intelligent” by prioritizing latency sensitive applications such as games and VoIP applications ahead of other programs. Here is FirstPacket in action:
DualNet® is another first by NVIDIA. All versions of the nForce 500 except the nForce 550 have DualNet available. The nForce 500 family is the first chipset to have dual gigabit connections on one physical chip. The two ports can be used separately, together, and in many other ways. One of these ways is called Teaming. Teaming makes two network connections appear as one, thus providing twice the bandwidth of a single connection. This is ideal for game servers that need to handle the demands of many connections, particularly during LAN parties.
Another method of utilizing DualNet is Failover. This method let’s you have a parallel connection. This way, in case one connection goes down, there will be another one to continue the work. This feature is critical for servers that upload important information. If a file is being uploaded/downloaded and someone trips over a wire, the other connection will continue the process seamlessly.
The name of this feature is a little confusing, as fundamentally it implies that it requires SLI in order to work, but it doesn’t. What SLI Memory refers to is a new feature that NVIDIA is working with memory manufacturers to aid end users in getting the maximum amount of performance out of their memory modules.
The last technology introduced by nVidia is MediaShield™, a RAID optimization. RAID, short for Redundant Array of Independent Disks is a method to boost disk capacity and transfer speeds using multiple hard drives. RAID is easy to set up, right in Windows with MediaShield. MediaShield tells you which disk to replace if one fails and reports the health status of the disks. As with nForce 4, the nForce 500 series supports Native Command Queuing, which reorganizes disk requests to minimize passes and improving latencies.
MediaShield also eliminates the bottleneck of having an additional RAID controller. An additional RAID controller means hard drives communicate with the system through one lane of the PCI Express bus. Having all 6 SATA ports native eliminates the bottleneck, increasing performance. The PCI Express lane only has a max bandwidth of 1.5GB/sec, which may be fine for a single disk, but with multiple disks, there will be a bottleneck. MediaShield and nVidia’s MCP redesign eliminate any possible problems.
The nForce 590 is the flagship of NVIDIA’s new AM2 chipset platform. It is the only dual chip version of the new platform like the nForce 4 SLI X16, NVIDIA’s previous high-end offering. It incorporates all of the new technologies NVIDIA now supports including native dual gigabit LAN, nVidia LinkBoost™, Teaming, TCP/IP Acceleration, nVidia MegaShield™, and finally Max overclocking. To see how nForce 590 SLI compares to its predecessor, nForce4 SLI X16, we’ve provided the following block diagram of the nForce4 SLI chipset:
As you can see, a few crucial aspects have changed besides the addition of new technologies. First of all, one IDE (PATA) connector has been removed on nForce 590, allowing only 2 parallel ATA drives. This is not an issue unless you still use IDE hard drives (which are not slower than SATA drives) with IDE optical drives.
Substituting the missing IDE port on nForce 590 are two extra native SATA 3Gb/sec connectors, giving NVIDIA support for up to six SATA drives versus ATI’s four. Again, NVIDIA is pushing the move to SATA hard drives only. The audio has been upgraded to HD Azalia. One final change in nForce 590 is that the PCI Express x1 slots are no longer controlled by the SPP (North Bridge) instead the x1 PCI Express connections are hanging off the South Bridge. NVIDIA expects nForce 590 SLI motherboards to sell for $150 and up.
Fortunately, the nForce 570 still has all the premium options of the nForce 590, such as 6 SATA 3Gb/sec ports, and dual native gigabit Ethernet.
In terms of pricing, NVIDIA has the nForce 570 Ultra pegged in the $110 price range, while nForce 570 SLI motherboards should start around $130, about $20 less than nForce 590 SLI. Considering the extra features found in nForce 590 SLI, we expect that most enthusiasts will probably opt for that chipset instead of nForce 570 SLI if the prices NVIDIA is projecting end up holding true.
Finally comes the nForce 550, the least expensive and thus least enriched version of the nForce 500 series. Like the nForce 570, the nForce 550 is based on a single-chip configuration. It is very stripped down though. Gone is NVIDIA’s DualNet™ Technology, which includes dual native gigabit Ethernet, Teaming, and TCP/IP acceleration. Also missing is the FirstPacket™ Technology as well as RAID 5 capabilities. Two of the 6 SATA 3Gb/sec have also been removed. The nForce 550 does have the essential core components of the nForce 500 series, such as HiDefinition audio, 10 USB 2.0 ports, SATA 3Gb/sec and gigabit Ethernet though.
NVIDIA expects nForce 550 motherboards to sell in the $70 range.
While the board above is technically manufactured by Foxconn (the C51XEM2AA), it’s actually based entirely on NVIDIA’s reference board design for the nF590 SLI chipset. Foxconn makes no changes to the NVIDIA reference design, so we’re essentially comparing ATI’s reference motherboard to NVIDIA’s in this article.
Although there is no passive cooling like on other premium motherboards, this MCP Southbridge cooler was extremely quiet even under load. The SPP Northbridge cooling was passive, but was extremely hot to the touch under load. I could not hold my finger on it for more than a couple of seconds at a time, or else I would have been burned. The PWM and MOSFETs aren’t cooled at all, something we don’t see on premium motherboards. Our guess is that NVIDIA built these components to withstand high temperatures.
As we noted before, one of the IDE ports has been removed. In its place are two more native SATA ports for a total of six. You’ll also notice four 240 pin DDR2 slots clearly showing the dual channel configuration placement. The blue slots are one channel, and the black slots make up the other channel.
Nothing special here, except a typical SLI layout with two slots in between for video cards with thick, dual slot coolers such as the GeForce 7900 GTX. You can also see a Molex connector at the bottom edge of the board. It is meant to feed juice to the motherboard when you have an SLI configuration. This is an awkward placement, and some power supplies may not even have long enough cables to reach that location.
Above you can see the Texas Instruments Firewire controllers and the Realtek ALC880 audio codec supporting the HiDef Azalia audio.
Here are some great goodies for the tweaking community. The power and reset buttons on the motherboard means you don’t have to have your system in a case for initial tweaking and testing. The cell battery can also be easily replaced, with a vertical placed battery unlike some of the older flat oriented niches. From the back panel we see the standard layout of a premium motherboard. There are two LAN jacks, 6 USB 2.0 ports, two Firewire ports, one standard and one small, and the standard PS/2 keyboard and mouse ports. The HiDef audio with optical out is seen as well. The serial and parallel legacy ports have disappeared, which is very normal for current premium motherboards. One thing that we missed is the external SATA ports, featured on several of the high end nForce 4 and nForce 5 motherboards. We can’t really blame anyone though, since this is a reference board.
Here is the test setup for this review:
AMDX2 5000+ (2x 2.6GHz 512Kb cache)
2GB (2x1GB) Corsair PC2-6400 DDR2 (4-5-5-5 2.3V)
Foxconn C51XEM2AA motherboard – nF590 first release drivers
ATI Xpress 3200 AM2 reference motherboard – first release drivers from ATI
BFG 7900GTX (670/800) – 91.27 drivers
Seagate 160GB 7200.7 IDE
Antec SP2.0 450W
Benchmarks that are being conducted are following:
- Memory Read
- Memory Write
- Memory Latency
- Float Memory Bandwidth
- Integer Memory Bandwidth
Windows Media Encoder 9
- 1600x1200 Ultra Quality 4xAA 16xAF
- 1280x1024 Ultra Quality 4xAA 16xAF
Call of Duty 2
- 1600x1200 16xAF
- 1600x1200 4xAA 16xAF
- 1600x1200 16xAF Maximum Settings
- 1600x1200 4xAA 16xAF Maximum Settings
Elder Scrolls IV: Oblivion
- Outdoors 1600x1200 4xAA 8xAF
- Indoors 1600x1200 4xAA 8xAF
- Foliage 1600x1200 4xAA 8xAF
From these results, you can see that memory performance is about equal and will be unnoticeable in real world applications.
We can once again infer that memory performance is even on these two platforms. This shouldn’t be too surprising, considering that the memory controller is located on the CPU itself on the Athlon 64 platform.
3DMark as its name states heavily relies on the video card and the CPU. As you can tell above, the motherboards made no difference, and had equal performance.
Aquamark is another graphically intense program. It is also affected by the CPU greatly. The motherboards made no impact here, and once again, the performance was even.
Encoding is mainly a CPU task that involves the memory and hard drive. The motherboards played no role in affecting the results, and performed evenly.
RightMark 3D Audio CPU Utilization Test
ATI Xpress 3200
nVidia nForce 590
Look at the highlighted area in the table for results, not the graphs. As you can see from the results, the two audio codecs perform almost exactly alike. This is because both motherboards use the same exact audio chip – the Realtek ALC 880.
nVidia nForce 590
NVIDIA edges out ATI here by 3.1%. Although these results won’t be noticeable in real world applications, the difference between the storage technologies is shown. NVIDIA has edged ATI in most of the tests so far, but except for the HD Tach test, the boards are even. Let’s move on the gaming tests.
This is quite a margin. Although the FPS difference may be small at first glance, the percentage difference is much larger.
Once again, nVidia takes the lead, but a moderate margin.
The difference here was much smaller than the two pervious games. Quake 4 is not as taxing of a game either, so that may be a reason for the differences.
This is one of the most taxing games to date, yet the differences are moderate. So the advantage of NVIDIA’s new platform varies from game to game. However one thing is clear from these benchmarks if you crave the most performance for gaming: if you plan on using a single ATI or single/dual NVIDIA cards, go for the NVIDIA nForce 500 platform. If you want Crossfire, you’re going to have to go the ATI route.
NVIDIA nForce 590
ATI CrossFire Xpress 3200
Fortunately, NVIDIA did improve greatly on their nForce 4 platform with the new nForce 500 series. Technologies such as DualNet and Teaming that let you configure the two native gigabit Ethernet ports so they can be used separately, or together, should appeal to home networking buffs or anyone who’s running a small gaming LAN in their home. FirstPacket is another intelligent improvement that prioritizes latency sensitive applications such as games and VoIP applications ahead of other programs.
On the other hand, ATI didn’t improve much on their existing platform. They did what they were supposed to do. ATI made a successful and stable chipset for AM2 by converting the existing Xpress 3200 platform and adding the new SB600 Southbridge. The SB600 did improve USB and storage performance, but that was just about the only thing ATI did different during the transition.
So in conclusion we’d like to say that both platforms turned out to be a success. One the one hand you’ve got NVIDIA’s nForce 500 which is jam-packed with features natively right out of the gate. Any gamer who plans on purchasing a GeForce graphics card (particularly a 7900 GTX) is probably naturally going to want to opt for an nForce motherboard to go along with it. But NVIDIA’s integrated just enough features to tempt ATI users who’d like to stick to single card setups to go the nForce route.
On the other hand, ATI’s Xpress 3200 CrossFire is no slouch either. Performance was a hair behind the nForce 590 in our gaming tests, but everywhere else ATI’s performance was on par with the NVIDIA chipset. On top of that, ATI’s solution runs cooler.
ATI’s chipset may not match NVIDIA feature-for-feature, but motherboard manufacturers will no doubt try and integrate as many features as they can to make it competitive with nForce 590. Adding dual GigE controllers and an extra Serial ATA controller shouldn’t be a problem for most manufacturers.
In a lot of ways, this chipset battle between ATI and NVIDIA is probably going to ultimately boil down largely to price and availability. It’s here where NVIDIA has the early advantage. Whereas we’ve already received a bevy of nForce 590 SLI motherboards, we haven’t received a single ATI-based AM2 motherboard from anyone outside of ATI. ATI’s certainly got all the major players onboard for the AM2 adaptation of CrossFire Xpress 3200, but it appears that most manufacturers have focused their early efforts on NVIDIA’s chipsets rather than ATI, this may be due in some part because of SB600’s availability. We’ve heard reports that production is lagging and that as a result supply isn’t meeting expectations.
If these rumors are true, ATI needs to get these issues resolved immediately or else they could end up letting a perfectly good opportunity to gain some ground on NVIDIA in the high-end space go to waste.
|© Copyright 2003 FS Media, Inc.|