Core 2 chipsets compared

The prospect of a Core 2 system build can seem a bit daunting for enthusiasts who have spent years focused solely on the Athlon 64. Core 2 processors need new motherboards for those switching from the Athlon 64, and that requires navigating a whole new world of core logic chipsets. Since the Core 2 processor relies on the chipset for its memory controller, one’s chipset choice can also have a much more profound impact on performance.

To guide you through the brave new world of Core 2 chipsets, we’ve rounded up the latest core logic that Intel and Nvidia have to offer. Read on to see which should be at the heart of your next Core 2 system.

Catching up with the Intel world
Before diving into some of the more interesting features you’ll find in these chipsets, I should take a moment to highlight a couple of contenders that are still missing in action. The most conspicuous absentee from today’s round-up is Nvidia’s nForce 590 SLI, a chipset that launched for the Athlon 64 months ago and was supposed to be available for the Intel platform by now. Nvidia initially intended for the 590 SLI Intel Edition to lift off this summer, but it pushed back the release date several times before finally asking reviewers not to publish articles about the Nvidia reference board at all. Our subsequent attempts to pry information about the nForce 590’s release from Nvidia have gone unanswered, and even motherboard makers have been unwilling to comment on the chipset. Rumor has it that they were unhappy with the chipset’s north bridge component, and that a replacement is coming in October.

ATI’s long-awaited RD600 chipset is also expected in October, complete with the SB600 south bridge that debuted in CrossFire Xpress 3200 chipsets for AM2 processors. However, the company’s recent acquisition by AMD has left RD600’s future in doubt.

Fortunately, there’s little doubt surrounding Intel’s chipset offerings. Several of Intel’s designs are compatible with its new Core 2 chips, including the older 975X Express. However, it’s the new P965 Express chipset that’s stealing the show. This “mainstream” chipset is being used on mid-range and high-end motherboards alike, and that puts it right in the sweet spot for enthusiast boards.

Although it’s classified as a mainstream chipset, the P965 sports a number of new memory controller features that Intel refers to as Fast Memory Access. These features mainly deal with the intelligent reordering of pending memory access requests to reduce latency and improve performance. Pending requests are monitored to determine which can be issued concurrently without interfering with a current memory request. Intel calls this feature just-in-time command scheduling. Also, out-of-order execution allows the memory controller to bump up pending requests that refer to memory pages that are already open.

Previous Intel memory controllers accumulated write requests in a pending queue that was only flushed when “certain watermarks were reached.” Unfortunately, flushing the write queue monopolized the memory controller, forcing pending read requests to wait. To improve the continuous flow of data, the P965 memory controller is capable of opportunistically issuing write requests during idle times and when they won’t interfere with read operations.

In addition to a bunch of swanky memory controller features, the P965 Express chipset comes with a new line of ICH8 south bridge chips. These chips add support for a couple of extra Serial ATA drives, but for the most part, they’re similar to what Intel had offered with the previous generation ICH7.

Speaking of similarities, Nvidia’s supposedly next-gen nForce 570 SLI chipset bears an uncanny resemblance to older nForce4 designs. The most obvious give-away is the filename of the chipset’s driver package: nForce4_intel_8.22_winxp2k_english_whql.exe. More surprising, however, is the fact that the chipset has a single integrated Gigabit Ethernet controller with no hardware TCP/IP acceleration. The nForce 570 SLI for AMD processors, on the other hand, sports dual hardware-accelerated GigE controllers. That chipset also boasts additional Serial ATA and USB ports, making us question why these two products are tagged with the same nForce 570 SLI moniker.

To be fair, though, the Intel version of the nForce 570 SLI does boast support for Nvidia’s FirstPacket networking quality of service feature. FirstPacket debuted with the nForce 590 SLI for AM2 processors, and allows users to prioritize outbound packets on an application-by-application basis. That should help keep BitTorrent uploads from lagging your Counter-Strike game, but since FirstPacket can’t prioritize incoming packets, downloads will still give you an excuse for dying.

Comparing the competition
Today we’ve assembled the four chipsets you’re most likely to find on enthusiast-oriented motherboards for Core 2 processors: the P965 Express and 975X Express chipsets from Intel, and the nForce4 SLI X16 and nForce 570 SLI from Nvidia. Here’s a quick look at their north bridge chips’ specifications.

Interestingly enough, only the P965 has explicit support for DDR2-800 memory. The rest top out at DDR2-667. That hasn’t stopped motherboard makers from building 975X, nForce4 SLI X16, and nForce 570 SLI boards with the necessary dividers to run their memory at 800MHz, though. We’ve even seen boards supporting memory speeds of 1066MHz.

Judging by the table of north bridge features above, Nvidia appears to have an edge when it comes to PCI Express lanes. Since some of these chipsets offer PCI-E lanes from their south bridge chips, though, that isn’t the whole picture. The nForce chipsets are, however, the only ones with Nvidia’s blessing to work with its SLI multi-GPU technology. The nForce4 SLI X16 even does so with a full 16 lanes of bandwidth to each graphics card. Multi-GPU graphics support on Intel chipsets is presently limited to ATI’s CrossFire, which is only fully implemented on the 975X. ATI just announced support for CrossFire on P965 boards with its latest Catalyst 6.9 drivers, but only for Direct3D applications. OpenGL support should be added by year’s end.

Chip-to-chip interconnect bandwidth is one area where the nForce chipsets have a definite advantage over their competition. HyperTransport serves up a cool 8GB/s of bandwidth between the nForce chips’ north and south bridge components, while the P965 and 975X are limited to only 2GB/s via Intel’s Direct Media Interface (DMI). DMI hasn’t proven to be a bottleneck in the past, so Nvidia’s advantage here may be theoretical at best.

By following the chipset interconnect, we find the 975X north bridge paired up with the ICH7R south bridge, the nForce SPPs matched with their respective MCPs, and the P965 paired with either the ICH8 or the ICH8R.

Intel prefers PCI Express peripherals hanging off the south bridge, so like the ICH7 before it, the ICH8 features six PCI-E lanes. The nForce 570 SLI, meanwhile, consolidates all its PCI Express on the north bridge. Because it hangs secondary graphics cards off the south bridge, the nForce4 SLI X16 chipset has an additional 20 lanes of connectivity in its MCP. Given its feature set, we suspect that the nForce4 SLI X16 MCP is in fact an nForce4 SLI chip (familiar from so many Athlon 64 motherboards) that’s been pressed into service as a south bridge chip.

Each of the chipsets supports 300MB/s Serial ATA transfer rates, but the ICH8 and ICH8R are the only ones to provide six SATA ports. (Interestingly, the nForce 570 SLI for AMD processors also sports six SATA ports, but its Intel cousin does not.) The ICH8 and ICH8R are differentiated from one another only by their support for RAID and AHCI. Intel implements SATA Native Command Queuing (NCQ) through AHCI, so the ICH8 also misses out on this worthwhile I/O performance enhancement. Interestingly, the nForce chipsets manage to support NCQ without AHCI.

Despite the fact that most south bridge chips support multiple RAID levels, there are a couple of key differences between the Intel and Nvidia storage controllers. Nvidia supports RAID 0+1 while Intel sides with RAID 10, for example. There’s little difference between the performance of each array type, but RAID 0+1 arrays can handle a single drive failure, while RAID 10 arrays are capable of surviving the failure of two drives. The ICH7R and ICH8R also support Matrix RAID, which allows users to combine separate RAID 0 and 1 array partitions using only two drives. Intel recently updated its Matrix RAID software to support three- and four-drive configurations, giving users the freedom to create freakish hybrids such as a four-drive array that combines RAID 0 and RAID 5.

With the hard drive world firmly embracing Serial ATA, the need for IDE ports is diminishing quickly. So quickly, in fact, that Intel has dropped IDE support completely from its ICH8 family. Unfortunately, Serial ATA optical drives are few and far between, forcing motherboard makers to use auxiliary ATA controllers on motherboards featuring ICH8 chipsets. Those chips aren’t necessary on boards with the ICH7R, nForce4 SLI X16, or nForce 570 SLI, all of which have at least one ATA channel.

Just about the only thing that suggests that the nForce 570 SLI is newer than the nForce4 SLI X16 is the fact that it joins Intel’s chipsets in supporting High Definition Audio. The nForce4 is still capable of producing eight-channel output, but it’s stuck with basic AC’97 audio and lacks support for HDA’s sampling rates and resolutions. To be fair, though, anyone serious about truly high-definition audio (as opposed to the Intel-spawned marketing name “High Definition Audio”) will probably want a discrete sound card with higher fidelity than motherboard audio implementations typically provide.

That leaves us with networking, and true to form, Intel prefers to let auxiliary chips handle Gigabit Ethernet rather than integrating it into the south bridge. That’s not necessarily a bad thing, especially given the throughput and CPU utilization we’ve seem from some of Marvell’s latest GigE chips on newer mobos. However, motherboard manufacturers don’t always use the best networking chips, so there’s some comfort in having a GigE controller integrated into Nvidia’s nForce chipsets. Well, at least there’s some comfort in having the nForce4 SLI X16’s hardware-accelerated Gigabit Ethernet controller. We’re not so keen on the nForce 570 SLI’s lack of hardware TCP/IP acceleration, and apparently neither are motherboard manufacturers. The nForce 570 SLI-based Asus P5NSLI that Nvidia sent us for testing forgoes the integrated networking controller in favor of a Marvell chip.

Our testing methods
We have an all-Asus lineup for testing today. You can find out more about the P5B Deluxe, P5W DH Deluxe, and P5N32-SLI SE Deluxe in our three-way Core 2 mobo round-up.

All tests were run at least twice, and their results were averaged, using the following test systems.

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.

Also, all of our test systems were powered by OCZ GameXStream 700W power supply units. Thanks to OCZ for providing these units for our use in testing.

We used the following versions of our test applications:

• SiSoft Sandra Standard 2005 SR3
• WorldBench 5.0
• TCD Labs HD Tach v3.01
• Futuremark 3DMark06 Build 1.02
• Far Cry v1.3
• Splinter Cell Chaos Theory v1.05
• RightMark 3D Sound 2.2
• NTttcp
• Cinebench 2003
• Sphinx 3.3
• trq4demo1 demo
• F.E.A.R. 1.04
• Intel IOMeter v2004.07.30
• Intel iPEAK Storage Performance Toolkit 3.0

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.

Memory performance

The nForce4 and 975X are faster than their more mainstream counterparts in Sandra and Cachemem’s memory bandwidth tests, but not always by much. Note that the P965’s write performance is particularly good in Cachemem, likely because of Intel’s new Fast Memory Access memory controller optimizations. The 570 SLI’s write bandwidth is particularly poor by comparison.

Moving to latency, the nForce4 and 975X lead the way again, followed closely by the P965. The 570 SLI is way off the pace with a memory latency that’s close to 50% higher than that of the P965.

Memory controllers don’t always perform well when all four DIMM slots are populated, so we added a couple of additional memory modules to our test systems. On the nForce systems, we had to raise the DRAM command rate from 1T to 2T to get the boards to boot. Command rate control isn’t available on the 975X or P965 boards we have, but they had no problem booting with four DIMMs without any fiddling.

The picture doesn’t change much when all four DIMM slots are populated. However, the nForce 570 SLI’s memory latency is greatly improved. 59.3 nanoseconds is much closer to what we’d expect from this chipset, despite the fact that we consistently observed much higher latencies with only two DIMMs installed.

Memory performance – con’t
The following Cachemem latency graphs are a little indulgent, but they do a good job of highlighting access latency across various block and step sizes. The Core 2 Duo runs out of on-chip cache after a block size of 4096KB, so you’ll want to pay more attention to the memory access latencies that follow with larger block sizes.

I’ve arranged the following graphs in order of highest to lowest latency with a common Z-axis to aid comparison. These access latency results are from our two-DIMM system configurations.

The nForce 570 SLI has all sorts of problems with memory access latency, especially when compared with the competition.

Cinebench rendering

Despite variance in memory bandwidth and latency, the chipsets are even in Cinebench’s rendering and shading tests.

Sphinx speech recognition

Sphinx is much more sensitive to memory subsystem performance. Here, the Intel chipsets lead the way, with the P965 just trailing the 975X. The nForce4 SLI X16 isn’t far off the pace, unlike the 570 SLI, which just can’t keep up.

WorldBench

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.

Only two points separates these chipsets in WorldBench. Nvidia’s nForce core logic fares a little better here, but the overall scores are very close.

Multimedia editing and encoding

MusicMatch Jukebox

Windows Media Encoder

Adobe Premiere

VideoWave Movie Creator

Scores are also bunched tightly throughout WorldBench’s multimedia editing and encoding tests, with the Intel chipsets running a little faster in Premiere.

3D rendering

3D Studio Max

nForce core logic proves faster in WorldBench’s OpenGL 3D Studio Max test, though.

Image processing

Adobe Photoshop

ACDSee PowerPack

WorldBench’s image processing tests don’t show one chipset finishing too far ahead of another.

Multitasking and office applications

Microsoft Office

Mozilla

Mozilla and Windows Media Encoder

The results for WordBench’s office and multitasking tests are a little more interesting, though. WorldBench’s Mozilla test tends to favor quicker memory subsystems. Although the nForce 570 SLI doesn’t suffer as much as one might expect, it trails the nForce4 SLI X16 and 975X Express, as does the P965. Surprisingly, the tables turn in the Office XP test, where the 570 SLI and P965 lead the pack.

Other applications

Winzip

Nero

Scores don’t vary much in Nero or WinZip.

Gaming

With the exception of Quake 4, gaming performance is pretty close. The nForce chipsets do especially well in Quake, at least at these relatively low detail levels and resolutions.

Multi-GPU gaming performance
Our first round of gaming tests was conducted with more modest in-game detail levels and display resolutions in order to remove the graphics subsystem as a performance-limiting factor, but we’ve cranked things up for a second round. These tests use high resolutions, high detail levels, and anisotropic filtering and antialiasing. We’ve tested each board with a single GeForce 7900 GTX. The nForce 570 SLI and nForce4 SLI X16 were also tested with a second 7900 GTX running in SLI. On the CrossFire front, the 975X was tested with a Radeon X1900 XTX in single-card and CrossFire configurations. Since the P965 only gained CrossFire support yesterday through ATI’s latest Catalyst graphics drivers, and then only for Direct3D applications, it will have to make do with a single GeForce 7900 GTX this time around.

Our purpose here is not to compare the merits of SLI versus CrossFire, or even the GeForce 7900 GTX with the Radeon X1900 XTX. Instead, we’re looking at how performance scales when a second graphics card is installed.

The performance of our multi-GPU systems scales predictably. Note that the nForce4 SLI X16 doesn’t have a clear advantage over the nForce 570 SLI, despite the fact that the 570 SLI is limited to a dual eight-lane PCI Express config rather than a true, dual-16-lane SLI implementation

Serial ATA performance
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’ve also included results from a Foxconn P9657AA-8KS2H motherboard to explore the P965 Express’ performance with a vanilla ICH8 south bridge.

IOMeter
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.

IOMeter is a best case scenario for Native Command Queuing (NCQ), and the vanilla ICH8 shows its weakness here. The nForce chipsets do support command queuing, but they’re quite a bit behind the other Intel chipsets.

For whatever reason, the nForce chipsets won’t run IOMeter with a load of 128 or 256 outstanding I/O requests. We’ve seen this behavior before from other chipset and RAID controllers—including older Intel designs—so it’s nothing to get too worked up about. Even 64 outstanding I/Os is hitting a desktop chipset pretty hard.

IOMeter response times track with transaction rates, with Intel’s 975X and ICH8R-equipped P965 out in the lead.

CPU utilization is below 1% for all the chipsets, but the P965/ICH8R combo consumes slightly more CPU cycles than the others across both test patterns and all load levels.

iPEAK multitasking
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.

We’ve seen iPEAK consistently favor command queuing with Maxtor and Seagate drives in the past, but that’s not always the case with our test system’s Western Digital Caviar RE2. The ICH8 does surprisingly well in our first wave of iPEAK tests, beating the NCQ-equipped ICH8R under multitasking loads that include file copy operations.

Despite a poorer showing in IOMeter, the nForce 570 SLI is consistently faster than the nForce4 SLI X16. In fact, the 570 SLI often beats the 975X and at least one of the P965 configurations.

iPEAK multitasking – con’t

Our second round of iPEAK tests swings things back into command queuing’s favor, with the ICH8R-equipped P965 turning in faster service times than the ICH8 with each multitasking scenario. The Intel chipsets do better than their nForce competition here, and in an interesting reversal, it’s the 570 SLI that lags behind the nForce4 SLI X16.

HDTach
We used HD Tach 3.01’s 8MB zone test.

With the exception of the write speed test, the chipsets’ Serial ATA performance is pretty similar in HD Tach. The P965/ICH8R combo only scores that high in HD Tach’s 8MB zone test, not in the longer variable zone test. That higher average is a little deceptive, too. Here’s what HD Tach’s output looks like with the P965 and ICH8R.

Write speeds are all over the map when they should be tracking smoothly like read speeds. According to Intel, the problem is specific to the Caviar RE2. However, we’ve observed similar performance from other command queuing-equipped drives on the ICH8R, suggesting that the problem may not be limited to the RE2, but instead related to the interaction of the ICH8R and HD Tach’s 8MB zone test.

Unfortunately, our Caviar RE2 doesn’t support 300MB/s Serial ATA transfer rates, so that’s holding back performance in the burst speed test. The nForce4 SLI X16’s SATA controller is still a little quicker than the rest of the pack, though.

ATA performance
ATA performance was tested with a Seagate Barracuda 7200.7 ATA/133 hard drive using HD Tach 3.01’s 8MB zone setting.

ATA burst speeds are a little faster on the nForce chipsets, but the only breakout performance is in the write speed test, where the P965 is more than 10MB/s faster than the competition. The chipset isn’t really doing the heavy lifting here, though. Since the ICH8 south bridge family lacks an integrated IDE controller, ATA is being handled by an auxiliary JMicron JMB363 storage controller.

USB performance
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.

Intel is pretty competitive with Nvidia when it comes to USB reads, but the Intel chipsets’ write performance is 6-8MB/s slower. There’s more variance when it comes to CPU utilization, where the P965 leads and the 975X brings up the rear. Keep in mind that HD Tach’s margin of error in the CPU utilization test is +/- 2%.

3D Audio performance

With all four chipsets relying on third-party codec drivers for 3D audio support, CPU utilization lines up along codec manufacturer lines. The Analog Devices codecs on the nForce 570 SLI and P965 Express boards perform better here. CPU utilization is a little higher for the others, which use Realtek codec chips. Maybe the crab isn’t so fearsome after all.

Note that nForce4 SLI X16 system doesn’t have a score with 32 buffers. Realtek’s AC’97 codec drivers only go up to 25 simultaneous voices, while the other codec drivers max out at 32. We declined to use Nvidia’s own audio drivers because we’ve found that they don’t always work with all nForce motherboards, especially when mobo makers specifically bundle drivers for the audio codec.

Ethernet performance
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 boards were tested with jumbo frames disabled.

The only chipset making use of its integrated Gigabit Ethernet controller here is the nForce4 SLI X16, which provides solid throughput and the lowest CPU utilization of the lot thanks to its hardware TCP/IP acceleration. Auxiliary Marvell controllers dominate the rest of the field, and nicely illustrate GigE done right—and done wrong. Right, in this case, is a PCI Express chip like the 88E8053, which offers great throughput and reasonable CPU utilization. Wrong is a PCI-based chip like the 88E8001, whose pokey bus interface constrains throughput by at least 200Mbps.

PCI Express performance
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 even across the board, although we consistently observed lower CPU utilization with the nForce4 SLI X16 system.

PCI performance
To test PCI performance, we used the same ntttcp test methods and a PCI VIA Velocity GigE NIC.

The throughput of our PCI GigE adapter is pretty close from chipset to chipset, but again, the nForce4 manages to consume fewer CPU cycles than the competition. If it weren’t essentially impossible, I’d almost suspect that the hardware acceleration engine for the chipset’s own networking controller was lending a hand here.

Power consumption
We measured system power consumption, sans monitor and speakers, at the wall outlet using a Watts Up power meter. Power consumption was measured at idle and under a load consisting of a multi-threaded Cinebench 2003 render running in parallel with the “rthdribl” high dynamic range lighting demo.

Measuring system power draw on four motherboards with different assortments of onboard peripherals and extras doesn’t give us the best picture of chipset power consumption, but it’s clear that Intel’s core logic tends to consume less power than Nvidia’s nForce designs. Differences in chipset power consumption are more apparent at idle, where they probably make up a greater percentage of the system’s overall power draw.

Conclusions
The landscape of Core 2-compatible chipsets is an interesting one. On one hand, you have tried and true chipsets like the 975X Express and nForce4 SLI X16 appearing on boards that have been updated to support Core 2 processors. These chipsets aren’t new or particularly flashy, but they’re proven designs that don’t sacrifice performance when compared with the latest and greatest core logic.

Of course, the Core 2 chipset market isn’t all retro refits. Nvidia’s nForce 570 SLI is new, at least in name. Unfortunately, the chip’s features make it look more like a minor update to the aging nForce4 SLI than a legitimate member of the nForce 500 series. The discrepancies between the chipset’s features and those offered by the nForce 570 SLI for AMD processors are striking, and for the life of me, I can’t understand why Nvidia would drop a unique feature like hardware TCP/IP acceleration. Extra features have long been a tenet of nForce chipsets, and on that front, the 570 SLI comes up well short.

Compounding the 570 SLI’s disappointing feature set is comparatively high memory latency with two-DIMM configurations. This particular problem might be a quirk of the Asus P5NSLI motherboard we used for testing, but it’s a retail board with a production BIOS—a board Nvidia itself provided for review.

Intel P965 Express September, 2006

Unlike the nForce 570 SLI, which feels like little more than an uninspired retread, Intel’s P965 Express still has that new chipset smell. It also has a number of new features, including Fast Memory Access optimizations and additional south bridge Serial ATA RAID ports. The P965 also has relatively low power consumption, competitive performance, more robust Matrix RAID options, and evolving CrossFire support that should be complete by the end of the year. No wonder this chipset has proven so popular with motherboard manufacturers, who are building everything from budget $100 wonders to high-end enthusiast boards based on it.

The Intel P965 Express chipset’s mix of features, performance, and power consumption make it the perfect partner for Intel’s new Core 2 processor and our Editor’s Choice. Really, it’s an easy call to make. The P965’s upcoming CrossFire support kills the one reason we might have recommended the 975X. If SLI’s your bag, the nForce4 SLI X16 is a solid—albeit power-hungry—option, but the more attractive nForce 590 SLI should be just around the corner. The nForce 570 SLI, meanwhile, doesn’t even measure up to its own name, let alone to the P965.