In many ways, the latest core logic offerings from ATI and NVIDIA are evolutionary designs that address problems with previous chipsets. ATI claims its SB600 resolves the I/O performance problems that plagued the SB450, and NVIDIA promises the nForce 500 series’ Gigabit Ethernet acceleration sheds the hardware bug that afflicted the nForce4’s ActiveArmor. New features are also on the menu. The SB600 is ATI’s first stab at Serial ATA with 300 MB/s and Native Command Queuing, and the nForce 500 series is virtually bursting at the seams with fancy feature names, including FirstPacket, LinkBoost, and DualNet.
Is the combination of ATI’s CrossFire Xpress 3200 and SB600 potent enough to prevent NVIDIA’s nForce 590 SLI from inheriting the Athlon 64 core logic crown? We’ve subjected both chipsets to an exhaustive array of application, peripheral, and power consumption tests to find out, and the answer might surprise you.
ATI’s CrossFire Xpress 3200 for AM2
Oddly enough, the north bridge component of ATI’s CrossFire Xpress 3200 for AM2 chipset isn’t new at all. It’s the same chip ATI introduced several months ago for Socket 939, but this time around, it’s connected to Socket AM2. The chip offers 40 lanes of PCI Express connectivity, allowing it to feed two full-bandwidth PCI-E x16 slots. ATI has made much of this capability, claiming that multi-GPU configurations that hang one graphics card off the south bridge suffer from poor performance due to limited chipset interconnect bandwidth. Consolidating all of a chipset’s PCI Express lanes in the north bridge is certainly a much cleaner approach, but the performance of NVIDIA’s nForce4 SLI X16 chipset certainly doesn’t seem to suffer due to its dual-chip layout.
The multi-GPU configuration most likely to benefit from having both graphics cards hanging off the north bridge would be one that relies solely on PCI Express to pass data between cards. That type of setup is only common with low-end graphics cards, though; high-end configs use a CrossFire dongle or SLI bridge connector to pass data between graphics cards, largely bypassing PCI Express. Since the CrossFire Xpress 3200 is a high-end chipset, it’s unlikely to be paired with the kind of low-end or even mid-range graphics cards that stand the best chance of benefiting from its PCI Express configuration.
With 32 of the CrossFire Xpress 3200’s PCI Express lanes occupied by a pair of 16-lane graphics slots, eight lanes remain. Half are available to PCI Express peripherals and x1 slots, with the last four reserved for ATI’s Alink2 chipset interconnect.
Using standard PCI Express lanes for a chipset interconnect allows the CrossFire Xpress 3200 north bridge to easily interface with a range of south bridge chips, including ULi’s M1575. The M1575 was a popular way for motherboard manufacturers to avoid the liabilities associated with ATI’s older SB450 chipset, but with NVIDIA’s recent acquisition of ULi, the arrival of ATI’s new SB600 south bridge is particularly timely.
With the SB600, ATI claims it has addressed the I/O performance issues that plagued the SB450. There’s more to the SB600 than just performance improvements, though. ATI has also added features, including support for 300 MB/s Serial ATA transfer rates and Native Command Queuing to the chip’s SATA controller. That controller’s RAID capabilities have also been bolstered, with ATI adding support for four-drive RAID 10 arrays to complement RAID 0 and RAID 1. It’s a little disappointing that RAID 5 didn’t make the cut for the SB600, but given the poor write performance we’ve seen from several chipset-level RAID 5 implementations, it’s not a huge blow to the SB600’s appeal.
With the exception of RAID 5, the SB600 has everything else you’d expect from a modern south bridge chip, including support for AC’97, High Definition Audio, and 10 USB ports. A single ATA channel has also become the norm, for better or worse, limiting users to just a pair of parallel ATA devices. At the very least, the limited number of available ATA channels in new core logic chipsets should encourage optical drive manufacturers to offer a greater number of Serial ATA options.
One thing you won’t find in the SB600 is a networking component. ATI doesn’t even integrate a basic 10/100 Fast Ethernet controller, instead relying on motherboard manufacturers to offer Ethernet options via PCI or PCI Express. There are some pretty spiffy PCI-E Gigabit Ethernet controllers out there, so a lack of integrated networking isn’t necessarily a strike against the SB600. However, some GigE chips have less attractive performance characteristics, so prospective buyers have to be careful about which chip a mobo maker chooses to integrate.
ATI’s CrossFire Xpress 3200 for AM2 came to us riding a fancy-pants reference board with a white base and red trim. Look beyond the fancy colors, though, and it’s worth paying special attention to the reference design’s tiny passive coolers. ATI’s core logic chips have a reputation for running cool, and the Xpress 3200 and SB600 are no exception. The former is built using 90-nano fabrication technology and requires little more than a tallish passive cooler. The SB600’s heat outputor lack thereofis even more impressive. That chip is built using an older 130-nano process, but it makes do with just a tiny low-profile cooler.
NVIDIA’s nForce4 chipset family was introduced more than a year and a half ago, so it’s long overdue for a replacement. The nForce4 SLI X16 refresh did little more than add PCI Express lanes via a discrete north bridge chip, leaving plenty of room for NVIDIA to break new ground with its nForce 500 familyand it has, expanding and refining features found in the original nForce4 while adding new bells and whistles along the way.
The nForce 590 SLI occupies the high end of the nForce 500 series, and as one might expect, it’s capable of providing a pair of PCI Express graphics cards with 16 lanes of bandwidth each. Rather than consolidating those lanes in a single chipset component, NVIDIA splits them between the nForce 590 SLI’s north and south bridge chips. This arrangement works well enough with the nForce4 SLI X16 chipset, perhaps in part because NVIDIA uses a 16-bit, 1 GHz HyperTransport chip-to-chip interconnect that offers a whopping 8 GB/s of bandwidth. That’s as much bandwidth as the Athlon 64’s HyperTransport link to the rest of the system, making it unlikely the nForce 590 SLI’s chipset interconnect will become a bottleneck.
Just in case that’s not enough, the nForce 590 SLI also incorporates what NVIDIA calls LinkBoost technology. LinkBoost increases the clock speed of the chipset’s interconnect and PCI Express graphics links by 25%, boosting the bandwidth available to each to 10 GB/s. NVIDIA is adamant that LinkBoost isn’t overclockingthe chipset’s PCI Express and interconnect links have all been tested and qualified up to 125% of their default speeds. There is one catch, though. Because LinkBoost increases the clock speed of the PCI Express bus, it requires a graphics card that can handle the extra speed. To date, the only graphics card that NVIDIA has deemed LinkBoost-compatible is the GeForce 7900 GTX.
LinkBoost is a neat trick, but with such restrictions, its potential benefits won’t be universal. Those benefits may be few and far between, anyway, since the chipset’s link to the CPU (and thus to main memory) tops out at 8 GB/s.
At least the nForce 590 SLI won’t be bottlenecked when it comes to networking performance. The chipset’s south bridge component has two Gigabit Ethernet MACsone more than the nForce4 series. Each of those MACs has its own TCP/IP offload engine, promising lower CPU utilization during GigE transfersfor real this time, NVIDIA says. The nForce4 series apparently had a hardware bug that limited the TCP/IP offload engine’s effectiveness, forcing NVIDIA to back off the chipset’s hardware acceleration to avoid data corruption. According to NVIDIA, that bug’s been fixed in the nForce 590 SLI, a claim we’ll put to the test in our Ethernet performance tests.
NVIDIA has actually ditched the ActiveArmor moniker for the nForce 590 SLI’s Gigabit Ethernet controllers, perhaps to avoid association with the CPU utilization and data corruption problems that dogged the nForce4 family. The nForce4’s personal firewall software hasn’t made the transition to the nForce 500 series, either. That creates interesting implications for hardware acceleration, because the nForce 590 SLI’s TCP/IP offload engine isn’t compatible with third-party firewall software. You can use the TCP/IP offload engine or you can run third-party firewall software, but not both together.
NVIDIA has added a few new wrinkles to nForce 500 series networking. The most interesting new addition is a quality-of-service feature that NVIDIA calls FirstPacket. With FirstPacket, users can set higher priority levels for certain applications, allowing packets from those apps to cut in line and be transmitted ahead of packets from lower priority programs. Application priority is defined through an easy-to-navigate Windows driver control panel, making FirstPacket configuration a snap.
FirstPacket looks particularly promising for gamers looking for more consistent ping times while transferring files or running programs like BitTorrent in the background, but there are limitations. The nForce 590 SLI only has control over the packets it sends out, since it must accept every packet it receives. Therefore, FirstPacket is only capable of prioritizing outbound packets. That may be a significant limitation for folks who spend more time downloading than uploading, but NVIDIA only has control over the chipset, so they can’t impose a quality-of-service scheme on other network clients.
In addition to FirstPacket, the nForce 590 SLI also sports an EtherChannel-like Gigabit Ethernet teaming feature dubbed DualNet. DualNet takes advantage of the nForce 590 SLI’s pair of GigE MACs by combining them to act as a single networking controller. This double-wide Gigabit Ethernet connection can be used to push additional data. If one connection fails, the second will maintain service, adding a measure of fault tolerance. NVIDIA declined to reveal exactly how it presents the nForce 590 SLI’s dual GigE controllers as a single unit, but admitted that the scheme involves “playing some games” with ARP spoofing, among other tricks.
DualNet’s ability to boost networking throughput to a theoretical peak of 2 Gbps is an intriguing prospect, but one that few users will be capable of exploiting. It would take multiple GigE-capable clients to saturate a 2 Gbps connection, and even then, NVIDIA is only guaranteeing a 40% performance boost. That said, in a multi-client demo it showed the press, NVIDIA was able to leverage DualNet to boost networking throughput by 70%. Ultimately, DualNet’s performance potential looks more appropriate for server and multi-user environments where the nForce 590 SLI will be rare at best. DualNet’s fault tolerance also seems better suited to server environments. Perhaps we’ll see it appear in NVIDIA’s next-gen workstation and server chipsets.
However, for those who lack a Gigabit Network, DualNet is capable of teaming a pair of 10/100 Fast Ethernet connections. Home users could find that capability useful, especially at LAN parties where systems are often share a large number of files with multiple clients.
With the nForce 590 SLI brimming with new networking buzzwords, the chipset’s additional Serial ATA ports are almost an afterthought. The nForce 590 SLI supports up to six Serial ATA devicestwo more than other core logic chipsets. Those drives can be configured in RAID 0, 1, 0+1, and 5 arrays, and with support for six drives, it’s even possible to run a pair of three-drive RAID 5 arrays side-by-side. Unfortunately, NVIDIA admits it hasn’t done much to improve the performance of its chipset RAID 5 implementation. We experienced dismal write performance when we tested RAID 5 on the nForce4 last year, and although NVIDIA is aware of the issue, it has been hesitant to employ performance enhancements because it’s worried about inadvertently compromising data integrity.
Speaking of performance enhancements, the nForce 590 SLI’s Serial ATA controller can be fine-tuned to maximize performance for specific hard drive models. Each hard drive model has unique performance characteristics, and tweaking things like the Native Command Queuing queue depth can make a difference with some drives. Thus far, NVIDIA has only created a performance profile for Western Digital’s latest 150 GB Raptor, but there are plans to profile additional drives that are popular among enthusiasts. Support for performance profiles will also make its way to the nForce4 family through a driver update, although NVIDIA isn’t keen to let users fiddle with profiles on their own.
While bulking up elsewhere, the nForce 590 SLI has dropped an ATA channel, limiting support to just two ATA devices. That limitation is shared by ATI’s SB600 south bridge and even Intel’s ICH7, so it’s hardly a new phenomenon.
We may not be surprised to see NVIDIA dropping an ATA channel from the nForce 590 SLI, but we’re a little shocked to see support for AC’97 audio missing from the nForce 590 SLI spec. NVIDIA has taken seemingly forever to offer an alternative to AC’97, and rather than straddling the fence, it’s moving the entire nForce 500 series over to the “Azalia” High Definition Audio spec. NVIDIA sees no need to retain compatibility with the older audio standard, and we’re inclined to agree. AC’97 support would give motherboard manufacturers additional flexibility to use cheaper codec chips, but we’d rather they not have that option.
Rather than simply giving its board partners a basic reference platform for the nForce 590 SLI, NVIDIA says it put a lot of effort into designing a quality reference board with an intelligent layout, a feature-laden BIOS, and loads of overclocking headroom. Foxconn has fully implemented that reference design in a retail board, but unfortunately, our sample didn’t arrive in time for testing. Instead, we’ll be using Asus’s new M2N32-SLI Deluxe Wi-Fi Edition to evaluate the nForce 590 SLI. Expect more in-depth coverage of the Foxconn board and other nForce 590 SLI platforms soon.
Although the M2N32-SLI Deluxe doesn’t follow NVIDIA’s nForce 590 SLI reference design exactly, it’s worth noting that its chipset cooling is considerably more aggressive than that of ATI’s CrossFire Xpress 3200 reference board. The nForce4 SLI chipset has a reputation for running hot, and given the extensive array of heat pipes and cooling fins on the Asus board, it would appear that the nForce 590 SLI isn’t much cooler.
Oddly enough, the nForce 590 SLI’s north bridge component is manufactured using 90-nano fabrication technology, just like the CrossFire Xpress 3200 north bridge. Like the SB450, the nForce 590 SLI MCP is a 130-nano chip, although its more extensive feature set likely requires a significant number of additional transistors.
NVIDIA is taking advantage of the nForce 500 series’ launch to unveil the latest version of its nTune system utility. This latest rev offers an impressive number of new features, including more extensive BIOS tweaking, better hardware monitoring options, and support for multiple profiles that can be invoked based on system variables or application launches. NVIDIA is also eager to promote the nForce 590 SLI’s support for what it calls SLI memory. SLI has nothing to do with memory, of course, but NVIDIA appears determined to apply its SLI brand to as many platform components as possible. SLI memory merely refers to memory that adheres to the open Enhanced Performance Profile spec NVIDIA developed with Corsair. Enhanced Performance Profiles are a good thing, but support for them appears to be more of a BIOS feature than a chipset attribute.
Thus far, we’ve only discussed the flagship member of the nForce 500 family, the nForce 590 SLI. NVIDIA is also introducing several other members of the nForce 500 series today, including the nForce 570 SLI and nForce 570 Ultra. Both 570-series chipsets have dual hardware-accelerated Gigabit Ethernet controllers with support for DualNet and FirstPacket, and both feature six Serial ATA ports with support for RAID 0, 1, 0+1, and 5. However, neither member of the nForce 570 series features LinkBoost. The nForce 570 SLI also only offers eight lanes of PCI Express to each of a pair of graphics cards in SLI, while the Ultra derivative doesn’t support GPU teaming at all. Both are single-chip implementations, though. That makes us suspect that the nForce 590 SLI’s south bridge component is physically the same chip as the nForce 570 SLI.
Before diving into our performance tests, we’ve whipped up a couple of handy spec comparison tables that contrast the CrossFire Xpress 3200 and nForce 590 SLI with leading Socket 939 chipsets. The north bridge is up first, and since the same CrossFire Xpress 3200 chip is used for Socket AM2 and 939 platforms, we’ve only listed it once.
|CrossFire Xpress 3200||nForce4 SLI X16 SPP||nForce 590 SLI SPP|
|Processor interface||16-bit/1GHz HyperTransport||16-bit/1GHz HyperTransport||16-bit/1GHz HyperTransport|
|PCI Express lanes||40*||18||18|
|Chipset interconnect||PCI Express x4||16-bit/1GHz HyperTransport||16-bit/1GHz HyperTransport|
|Peak interconnect bandwidth||2GB/s||8GB/s||8GB/s|
Note the huge disparity in the number of north bridge PCI Express lanes between the CrossFire and nForce chipsets. NVIDIA makes it up at the south bridge, of course, and perhaps that’s why the nForce4 SLI X16 and 590 SLI have a much higher bandwidth speed chipset interconnect than the CrossFire Xpress 3200.
Next, we turn our attention south. We’ve added the ATI SB450, ULi M1575, and NVIDIA nForce4 SLI X16 MCP to accompany the south bridge components of our Socket AM2 chipsets.
|M1575||SB450||SB600||nForce4 SLI X16 MCP||nForce 590 SLI MCP|
|PCI Express lanes||4*||4*||4*||20||28|
|Serial ATA ports||4||4||4||4||6|
|Peak SATA data rate||300 MB/s||150 MB/s||300 MB/s||300 MB/s||300 MB/s|
|Native Command Queuing||Y||N||Y||Y||Y|
|Max audio channels||8||8||8||8||8|
|Ethernet||10/100||N||N||10/100/1000||2 x 10/100/1000|
The nForce 590 SLI’s cornucopia of PCI Express lanes, Serial ATA and Gigabit Ethernet options, and supported RAID levels looks mighty impressive when compared with what ATI offers with the SB600. Still, the SB600’s array of south bridge features should be enough for most folks, especially when paired with a decent PCI Express Gigabit Ethernet chip or two.
Note that while both the SB600 and nForce 590 SLI support Native Command Queuing (NCQ), only the SB600 does so through Intel’s Advanced Host Controller Interface for Serial ATA. NVIDIA has supported NCQ since the nForce4, but eschews AHCI in favor of a proprietary spec it developed before AHCI existed. According to NVIDIA, its approach offers better performance with no real penalties because AHCI is just a register-level interface spec. NVIDIA’s approach offers another perk in that additional drivers aren’t necessary to install Windows XP to a single Serial ATA drive; auxiliary drivers are required to install Windows to drives running in AHCI mode on the SB600. That will become a non-issue when Vista arrives because both ATI and NVIDIA are promising to have in-the-box drivers for Microsoft’s next operating system.
Microsoft drivers are actually responsible for the asterisk beside the nForce4 SLI X16 MCP’s support for ATA RAID. The chipset initially supported RAID for ATA devices, but Microsoft apparently didn’t take too kindly to its mass storage controller driver being used for a RAID device. Rather than taking on Redmond, NVIDIA has dropped support for ATA RAID on newer nForce4 chipsets, and left it out of the nForce 590 SLI. RAID isn’t really worth it on a single ATA channel, anyway.
Since we have a Socket AM2 processor article in the works, we’ve limited the bulk of our testing to chipset features and peripherals. Expect more extensive application performance results for a wide range of Socket AM2 processors soon.
Our performance testing focuses on the CrossFire Xpress 3200 for AM2 and the nForce 590 SLI, but we’ve also thrown in a couple of Socket 939 platforms for reference. Those platforms are based on the CrossFire Xpress 3200/ULi M1575 and nForce4 SLI X16 chipsets, respectively. Note that in the graphs, our Socket 939 CrossFire Xpress 3200 appears as the “CrossFire Xpress 3200,” while the Xpress 3200 for AM2 is listed as the “CrossFire Xpress 3200 AM2.”
In order to ensure comparable results between platforms, we tested the Socket 939 boards with an Opteron 180, which runs at 2.4 GHz, and we tested the AM2 platforms with an Athlon 64 FX-62 underclocked to 2.4 GHz. Both chips are dual-core K8 processors with 1MB of L2 cache, and for all intents and purposes, the Opteron 180 is equivalent to the Athlon 64 X2 4800+. The underclocked Athlon 64 FX-62 is also equivalent to the AM2 version of the X2 4800+. In order words, this should be a like-to-like match of processors between Socket 939 and Socket AM2, leaving the chipsets on equal footing. We also used relatively relaxed timings for our Socket 939 platform’s DDR memory because the DDR2 we used for testing the AM2 boards would only run at less aggressive 5-5-5-12 timings.
To prevent BIOS- or motherboard-specific memory timing defaults from skewing our results, we made sure that as many memory timings as possible were equalized across our respective Socket 939 and AM2 platforms. We also disabled LinkBoost on the nForce 590 SLI because it’s only compatible with the GeForce 7900 GTX. A GeForce 7900 GTX was used in our test systems, but as much as manufacturers would like us to evaluate entire platforms, today we’re targeting the chipset. To indulge those wondering whether LinkBoost really does make a difference, we’ve snuck a few extra results into our multi-GPU performance tests.
Our testing methods
All tests were run at least twice, and their results were averaged, using the following test systems.
|Processor||Opteron 180 2.4GHz||Athlon 64 FX-62 @ 2.4GHz|
|System bus||HyperTransport 16-bit/1GHz|
|Motherboard||Abit AT8 32X||Asus A8N32-SLI Deluxe||ATI reference||Asus M2N32-SLI Deluxe|
|North bridge||ATI CrossFire Xpress 3200||nForce4 SLI X16 SPP||ATI CrossFire Xpress 3200||nForce 590 SLI SPP|
|South bridge||ULi M1575||nForce4 SLI X16 MCP||ATI SB600||nForce 590 SLI MCP|
|Chipset drivers||ULi integrated 2.13||ForceWare 6.85||ATI 2.5.1540.25a||ForceWare 9.34|
|Memory size||2GB (2 DIMMs)||2GB (2 DIMMs)||2GB (2 DIMMs)||2GB (2 DIMMs)|
|Memory type||Corsair CMX1024-3500LLPRO DDR SDRAM at 400MHz||CorsairTWIN2X2048-6400PRO DDR2 SDRAM at 800MHz|
|CAS latency (CL)||2.5||2.5||5||5|
|RAS to CAS delay (tRCD)||4||4||5||5|
|RAS precharge (tRP)||4||4||5||5|
|Cycle time (tRAS)||8||8||12||12|
|Audio codec||Integrated M1575/ALC882D with Realtek HD 1.37 drivers||Integrated nForce4 SLI X16 MCP/ALC850 with Realtek 3.87 drivers||Integrated SB600/ALC880 with Realtek HD 1.37 drivers||Integrated nForce 590 SLI MCP/AD1988B with 220.127.116.11 drivers|
|Graphics||GeForce 7900 GTX 512MB PCI-E with ForceWare 84.21 drivers|
|Hard drive||Seagate Barracuda 7200.7 NCQ 160GB|
|OS||Windows XP Professional|
|OS updates||Service Pack 2|
Thanks to Corsair for providing us with memory for our testing. 2GB of RAM seems to be the new standard for most folks, and Corsair hooked us up with some of its 1GB DIMMs for testing.
We used the following versions of our test applications:
- WorldBench 5.0
- trq4demo1 demo
- Half-Life 2: Lost Coast with trcoast1 demo
- FutureMark 3DMark06 Build 1.0.2
- F.E.A.R. 1.04
- TCD Labs HD Tach v3.01
- Futuremark 3DMark06 Build 1.02
- RightMark 3D Sound 2.2
- Intel iPEAK Storage Performance Toolkit 3.0
- Intel IOMeter v2004.07.30
The test systems’ Windows desktop was set at 1280×1024 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests. Most of the 3D gaming tests used the Medium detail image quality settings, with the exception that the resolution was set to 640×480 in 32-bit color.
All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.
WorldBench overall performance
WorldBench uses scripting to step through a series of tasks in common Windows applications. It then produces an overall score. WorldBench also spits out individual results for its component application tests, allowing us to compare performance in each. We’ll look at the overall score, and then we’ll show individual application results alongside the results from some of our own application tests.
Although only one point separates the CrossFire Xpress 3200 AM2 chipset from our Socket 939 platforms, the nForce 590 SLI trails a few points off the pace. Perhaps WorldBench’s individual test results can shed some light on why.
Multimedia editing and encoding
Windows Media Encoder
VideoWave Movie Creator
The chipsets remain close through most of WorldBench’s media editing and encoding tests, but the nForce 590 SLI trails in the Movie Creator and Windows Media Encoder tests. With the Athlon 64’s memory controller integrated into the CPU die, it’s rare to see a chipset impact performance in these kinds of application tests. The Asus M2N32-SLI motherboard we used for testing could be the culprit here.
3D Studio Max
The nForce 590 SLI continues to trail in both of WorldBench’s 3D Studio Max tests, while the CrossFire Xpress 3200 for AM2 is about even with our Socket 939 platforms.
Photoshop proves problematic for the nForce 590 SLI, but NVIDIA’s latest just edges out the Xpress 3200 in ACDSee.
Multitasking and office applications
Mozilla and Windows Media Encoder
The nForce 590 SLI pulls off a rare win in WorldBench’s Office XP test, but its glory is short-lived. The Xpress 3200 for AM2 chipset reigns supreme in the remainder of WorldBench’s multitasking and office tests, beating the nForce 590 SLI by a healthy margin in the Mozilla test.
Winzip and Nero are close, with the Xpress 3200 finally showing some weakness in the latter.
Our first round of gaming tests focus on detail levels and resolutions low enough to eliminate our GeForce 7900 GTX graphics card as the bottleneck. Don’t you just hate it when a GeForce 7900 GTX holds you back?
The nForce 590 SLI redeems itself in this first round of gaming with strong showings in F.E.A.R., Quake 4, and 3DMark06’s CPU test. ATI’s CrossFire Xpress 3200 AM2 chipset isn’t far off the pace, but the Socket 939 platforms are consistently slower in these tests.
Our first round of gaming tests were conducted with modest in-game detail levels and display resolutions, but we’ve cranked things up for a second round. These tests use high resolutions, maximum detail levels, and anisotropic filtering and antialiasing. We’ve tested the SLI chipsets with single and dual GeForce 7900 GTX graphics cards and NVIDIA’s latest ForceWare 91.27 graphics driver. The CrossFire Xpress 3200 platforms were tested with single and CrossFire Radeon X1800 XTs and Catalyst 6.4 drivers. We also ran some extra tests on the nForce 590 SLI system with LinkBoost enabled.
When looking at our multi-GPU performance results, pay special attention to the jump in performance from single- to multi-card configurations. We’re not out to compare the Radeon X1800 XT’s performance with that of the GeForce 7900 GTX; we just want to see how the chipsets handle the addition of a second graphics card.
Our CrossFire Xpress 3200 AM2 CrossFire configuration crashed 3DMark06’s FireFly Forest test at 1600×1200 with 4x antialiasing and 16x anisotropic filtering. The setup was fine at other resolutions and in other tests, but without the FireFly Forest test, 3DMark06 can’t generate a Shader Model 2.0 score.
Adding a second graphics card boosts performance across the board, as one might expect, and there’s little difference between the Socket 939 and AM2 platforms in most tests. It’s interesting to note that LinkBoost has virtually no impact on graphics performance at these resolutions and detail levels.
The Serial ATA disk controller is one of the most important components of a modern core logic chipset, so we threw each platform a selection of I/O-intensive storage tests.
We’ll begin our storage tests off with IOMeter, which subjects our systems to increasing multi-user loads. Testing was restricted to IOMeter’s workstation and database test patterns, since those are more appropriate for desktop systems than the file or web server test patterns. Interestingly, only the ULi M1575 south bridge on our CrossFire Xpress 3200 Socket 939 platform was capable of running IOMeter with 128 and 256 simultaneous I/O requests. Both nForce boards and the Xpress 3200 for AM2 would only go up to 64 concurrent I/Os, but that’s a common limitation for desktop chipsets.
The M1575 may support a greater number of simultaneous I/O requests, but its performance doesn’t ramp until we move past 16 I/Os; the Xpress 3200 AM2 and nForce platforms scale at much lower I/O levels. Overall, the nForce 590 SLI takes the cake, scaling to match the Xpress 3200 AM2 at lower I/O levels, and achieving higher transaction rates up to 64 outstanding I/Os.
With the exception of the M1575, which is a little slow out of the gate, response times are consistent across the board. The nForce 590 SLI does have a slight advantage over the Xpress 3200 AM2 under heavier I/O loads, though.
The nForce platforms consume marginally more CPU cycles than their Xpress 3200 counterparts, but we’re talking about fractions of a percentage point here.
We recently developed a series of disk-intensive multitasking tests to highlight the impact of command queuing on hard drive performance. You can get the low-down on these iPEAK-based tests here. The mean service time of each drive is reported in milliseconds, with lower values representing better performance.
Among our AM2 platforms, the nForce 590 SLI is ahead of the Xpress 3200 in all but one of this first wave of iPEAK tests. The Socket 939 platforms steal the show in a couple of test patterns, though.
The results of our second round of iPEAK tests are more mixed, with the nForce 590 SLI and CrossFire Xpress 3200 passing the Socket AM2 performance crown back and forth with each test.
We used HD Tach 3.01’s 8MB zone test.
There’s very little variance in our HD Tach results, with the nForce platforms only showing a slight edge in the read burst speed test. That test is hampered by our Western Digital Caviar RE2 hard drive’s lack of support for 300 MB/s transfer rates, anyway.
Curiously, both our AM2 platforms have slightly slower random access times, although CPU utilization results are within HD Tach’s +/- 2% margin for error in that test.
ATA performance was tested with a Seagate Barracuda 7200.7 ATA/133 hard drive using HD Tach 3.01’s 8MB zone setting.
Scores are pretty close, but NVIDIA’s chipsets deliver slightly faster burst and write speeds.
Our USB transfer speed tests were conducted with a USB 2.0/Firewire external hard drive enclosure connected to a 7200RPM Seagate Barracuda 7200.7 hard drive. We tested with HD Tach 3.01’s 8MB zone setting.
Our CrossFire Xpress 3200 AM2 platform manages a respectable performance in our USB tests, proving that ATI has addressed the SB450’s dismal USB performance with the SB600. NVIDIA still offers higher throughput with lower CPU utilization, but the difference in performance isn’t as stark as it’s been in the past.
We used RightMark 3d Sound’s Idle Threads method to measure CPU utilization with 2D and 3D audio playback. We found that this method delivers more consistent results when using dual core processors.
The nForce 590 SLI looks like a champ here, but we should note that codec drivers often define performance in this test, and with 3D audio in general. The nForce4 SLI X16 and Xpress 3200 platforms use Realtek codecs and audio drivers, while our nForce 590 SLI platform uses an Analog Devices codec and drivers.
We evaluated Ethernet performance using the NTttcp tool from Microsoft’s Windows DDK. The docs say this program “provides the customer with a multi-threaded, asynchronous performance benchmark for measuring achievable data transfer rate.”
We used the following command line options on the server machine:
ntttcps -m 4,0,192.168.1.25 -a
..and the same basic thing on each of our test systems acting as clients:
ntttcpr -m 4,0,192.168.1.25 -a
Our server was a Windows XP Pro system based on Asus’ P5WD2 Premium motherboard with a Pentium 4 3.4GHz Extreme Edition (800MHz front-side bus, Hyper-Threading enabled) and PCI Express-attached Gigabit Ethernet. A crossover CAT6 cable was used to connect the server to each system. The nForce4 board was tested with the NVIDIA Firewall and Jumbo Frames disabled.
Only the nForce platforms actually integrated Gigabit Ethernet controllers in the chipset, and they do so with exceptional throughput and reasonably low CPU utilization. The nForce 590 SLI’s low CPU utilization is especially impressive, although the latest drivers for Marvell’s 88E805x Gigabit Ethernet controllers do a pretty good job, too. That makes the Xpress 3200 AM2 platform look comparatively good in this test, although the Socket 939 Xpress 3200’s high CPU utilization with a Realtek GigE chip nicely illustrates the fact that motherboard manufacturers don’t always spec boards with the best chips.
We used the same ntttcp test methods from our Ethernet tests to examine PCI Express throughput using a Marvell 88E8052-based PCI Express x1 Gigabit Ethernet card.
Throughput is consistent across each platform, but the CrossFire Xpress 3200 AM2 manages slightly lower CPU utilization than the rest of the field, followed by the nForce 590 SLI. PCI performance
To test PCI performance, we used the same ntttcp test methods and a PCI VIA Velocity GigE NIC.
NVIDIA’s nForce chipsets manage higher throughput here, with the Xpress 3200 trailing the nForce 590 SLI by roughly 70 Mbps. That’s not much in the grand scheme of things, especially considering the nForce 590 SLI’s higher CPU utilization. We were a little perplexed to see such high CPU utilization from the nForce4 SLI X16 chipset in this test, but performance was consistent across numerous test runs and even after re-imaging the test system.
System power consumption was measured at the wall outlet using a Watts Up power meter. Load measurements were taken with Cool’n’Quiet disabled and the systems crunching a combined load of Cinebench 2003’s rendering test and the rthdribl high dynamic range rendering demo. Idle measurements were taken at the Windows desktop with and without Cool’n’Quiet enabled. All motherboard components and peripherals were enabled, with drivers installed, for each test system.
Keep in mind that because our AM2 platforms are using an underclocked Athlon 64 FX-62, their power consumption results aren’t directly comparable to our Socket 939 platforms.
The Radeon Xpress 3200 runs away with our power consumption tests at both idle and under load. However, it’s worth noting that on both AM2 and Socket 939 platforms, the nForce boards have a greater number of peripheral devices and expansion ports. The nForce 590 SLI board, for example, features one more Gigabit Ethernet port than the Xpress 3200 AM2 platform and an additional onboard Wi-Fi card. Those extras are likely only responsible for a fraction of the 20 W power consumption gap between the Xpress 3200 AM2 and nForce 590 SLI, though.
Although their performance is largely equivalent, the CrossFire Xpress 3200 AM2 and nForce 590 SLI really couldn’t be more different. ATI’s chipset is a rather simple affair, with a solid spec but few frills or extras, while NVIDIA’s latest nForce is jam-packed with peripherals and additional features, some of which are more gimmicky than others. Fortunately, both are easy to recommend, but for different reasons.
The CrossFire Xpress 3200 for AM2 is the first all-ATI chipset we have no qualms about recommending. ATI’s new SB600 south bridge appears to have resolved the I/O performance issues that afflicted the SB450. With a solid Native Command Queuing implementation, support for 300 MB/s transfer rates, and RAID 10, the chip’s feature set doesn’t leave us wanting. However, the SB600’s lack of integrated Gigabit Ethernet does leave the door open for motherboard manufacturers to use cheaper GigE chips with less appealing performance characteristics, specifically higher CPU utilization. Some will get it right, but as we’ve seen with Xpress 3200 platforms for Socket 939, others will almost certainly get it wrong.
Fortunately, motherboard makers shouldn’t be able to mess with the Xpress 3200 AM2’s exceptionally low power consumption and conservative cooling requirements. Those characteristics make this chipset ideal for low-noise applications, although you probably don’t need a dual x16 CrossFire board in a home theater PC.
While ATI has created a relatively lean high-performance chipset with the CrossFire Xpress 3200 AM2, NVIDIA has gone in the other direction with the nForce 590 SLI. This latest nForce packs loads of connectivity options, including an impressive six Serial ATA RAID ports and two Gigabit Ethernet controllers. Extra features are what really make the nForce 590 SLI stand out, though. Things like FirstPacket and hard drive performance profiles are particularly intriguing. LinkBoost comes across as little more than a gimmick, though, especially since its compatibility is currently limited to GeForce 7900 GTX graphics cards. NVIDIA’s much-improved nTune system utility has more potential, but as with earlier versions of the software, it’s up to motherboard manufacturers to take advantage of the app’s capabilities. Few mobo makers have fully exploited nTune’s capabilities in the past, leaving us doubtful that many will fully support the app this time around.
Features aside, the nForce 590 SLI’s peripheral and gaming performance are generally excellent. The chipset’s disk controller, Gigabit Ethernet, and USB components are particularly impressive, although our nForce 590 SLI test rig stumbled in a number of WorldBench application tests. We’re also not enthusiastic about the nForce 590 SLI’s higher power consumption and cooling requirements.
Ultimately, motherboard manufacturers must properly implement the latest AM2 core logic offerings from ATI and NVIDIA. Those using the CrossFire Xpress 3200 will have to be careful to include the right auxiliary Gigabit Ethernet chips, and they will probably want to throw in an additional Serial ATA controller with RAID 5 support. Others implementing the nForce 590 SLI will have to pay special attention to chipset cooling and would do well to include all the BIOS hooks necessary to support NVIDIA’s nTune system utility. Which chipset is right for you depends largely on whether you prefer extra peripheral connectivity or lower power consumption and heat output. I suppose if we view these chipsets as individual components, the nForce 590’s more extensive array of integrated features and peripherals make it a more impressive product. However, with the right mix of auxiliary peripherals, motherboards based on the Radeon Xpress 3200 will be very competitive.