 | |||
|
|
|
|
|
 |
nForce4 SLI Performance Preview November 23, 2004 |
|
 |
|
| |
Summary: Ever since NVIDIA's SLI technology was announced earlier this summer, gamers everywhere have been eager to see how it performs. Well folks, today is that day! In this article we push GeForce 6800 GT SLI and GeForce 6600 GT SLI configurations through a battery of benchmarks, including overclocking. Do two GeForce 6600 GT cards equal one GeForce 6800 GT? Does SLI really provide double the performance? Find the answers inside!
 Introduction | Page:: ( 1 / 19 ) |
[image]
| <% print_image("01"); %> | <% print_image("02"); %> |
SLI technology appealed to both the hardcore and mainstream user alike. Enthusiasts who had the cash would simply drop in both Voodoo2 cards and enjoy the benefits right away, which not only included performance, but one new screen resolution as well: 1024x768. Voodoo2 gamers on a budget ended up buying one card immediately, and bought the second card at a later date, typically 6-12 months after the original card purchase. After all, keep in mind that these were $300 graphics cards at the time Voodoo2 was originally released.
No other card manufacturer had an immediate answer for SLI. Not NVIDIA, not Rendition, nor ATI or S3, both of whom were rather late to the 3D game, despite their dominant positions in 2D graphics. Eventually ATI released their MAXX technology to the market, but this didn’t provide the incremental upgrade option 3dfx’s SLI solution offered gamers on a budget, as both graphics cores were integrated on the same card. The performance of ATI’s Rage Fury MAXX card was also nothing to write home about. Metabyte also made an attempt with their Parallel Graphics Configuration technology (unofficially nicknamed Stepsister), but unfortunately for Metabyte, by the time PGC was ready for public consumption, newer AGP graphics cards were released, no one was interested in high-end PCI graphics cards anymore. PGC was also never entirely effective at load-balancing: the top segment of the screen isn’t always as complex as the lower half. PGC never truly addressed this. Alienware eventually bought the technology from Metabyte but a final product based on the technology was never released.
One fundamental problem seemed to doom all of these early attempts to recreate 3dfx SLI: the AGP interface was simply never designed to handle dual graphics cards.
[image]
| <% print_image("03"); %> | <% print_image("04"); %> |
PCI Express however has been designed with dual graphics cards in mind. NVIDIA – who bought many of 3dfx’s technology and assets a few years ago – is the first manufacturer to bring back dual graphics cards; and the name they chose to market their technology is none other than the legend that started it all: SLI!
| SLI theory | Page:: ( 2 / 19 ) |
While NVIDIA’s dual graphics card technology shares the same acronym as 3dfx, the underlying technology is completely different. 3dfx’s SLI technology worked by splitting the workload in half, one Voodoo2 card rendered the even lines on the screen, while the second card rendered the odd lines. The lines were then merged together to complete a full frame. The cards were linked together by an analog pass-through cable, with a second cable going to the 2D display. NVIDIA’s SLI, short for scaleable link interface is an entirely different animal from 3dfx’s scan-line interleaving technique.
NVIDIA SLI can work in one of two modes: alternate frame rendering, or split frame rendering. We’ll start with single frame rendering first.
Single frame rendering
Single frame rendering (SFR) mode in NVIDIA SLI works similarly to Metabyte’s PGC in the sense that the graphics card splits the workload horizontally across the screen. One card takes the upper portion while the second card takes the lower segment. The frame buffer data is then combined and sent to the monitor.
It’s important to note that SFR doesn’t necessarily split the screen directly down the middle; in some scenes the lower portion of the screen may be more complex than the upper portion, or vice versa. NVIDIA has developed custom load-balancing algorithms that are designed to take this into account, and split the screen appropriately – if one GPU takes longer to render, no problem, the driver just gives that GPU less work to do.
Alternate frame rendering
Up to now, ATI’s Rage Fury MAXX was arguably the most well known card to take advantage of alternate frame rendering. Rather than splitting the work up every frame like in SLI, with AFR, each graphics core handled alternate frames. Graphics core one would handle everything in frame 1, while graphics core two would then handle frame number 2. Each chip renders every other frame instead of alternating lines in the same frame, as 3dfx had done with scan-line interleave. If one GPU doesn’t finish drawing its frame, the remainder is pushed to the second GPU. This is the same concept NVIDIA has employed for their SLI technology, only we’re dealing with two distinct graphics cards, rather than two graphics cards on the same core.
The most well known downside with AFR has been perceived “lag” that may be felt by some twitch players in very fast-paced first-person shooters such as Quake 1 or Quake 3. The argument is that user inputs, such as a quick flick of your wrist to nail your opponent in mid-air with your rail gun may feel lagged. In theory, this could happen when the second AFR scenario we described occurs – GPU1 doesn’t finish its frame and the remainder is passed to GPU2 – there could be a lag between the key being pressed and the output being shown on the screen. Ironically enough, NVIDIA used this same argument to tout their GeForce 256 over Rage Fury MAXX five years ago.
We played a little Half-Life 2 (which uses AFR by default) to test out this theory, but didn’t perceive any additional lag during our “testing” sessions. Buffering kicks in when data is split between GPUs, and it appears to work well. Also keep in mind that today’s shooters aren’t as fast-paced as they were a few years ago.
Currently, AFR is NVIDIA’s preferred mode for SLI. In general, AFR uses less communications overhead (as no communications overhead between GPUs is used as long as frames are contained to each GPU), allowing it to scale better than SFR. NVIDIA also says that applications with heavy vertex loads benefit less from SFR.
| SLI hardware | Page:: ( 3 / 19 ) |
ASUS and NVIDIA made a big splash last week with their announcement that ASUS would be first to get an SLI product to market with their A8N-SLI motherboard. The A8N-SLI is a powerful motherboard, with dual Gigabit Ethernet networking (with NVIDIA firewall), dual IDE RAID with support for up to eight Serial ATA hard drives total, Firewire, 8-channel software-based audio with jack-sensing technology and an external S/PDIF-out, and of course the standard nForce4 features such as 10 USB 2.0 ports and 3Gb/sec Serial ATA.
ASUS also includes their standard features such as Q-Fan and AI NOS for overclocking, as well as POST Reporter and CrashFree BIOS2, but they’ve gone the extra mile by including an additional 4-pin power header for supplying extra power to the graphics cards, which can be helpful when running in SLI mode. ASUS refers to this as EZ Plug. Accompanying the EZ Plug connector is a red LED that will shine if the graphics card isn’t receiving enough power.
[image]
| <% print_image("05"); %> | <% print_image("06"); %> |
When running in SLI mode, we always used the EZ Plug connector to ensure optimal stability.
The board’s BIOS has everything a tweaker would want when it comes to overclocking. Bus speed options range from 200-400MHz in 1MHz increments, while the PCI Express interface can also be adjust from 100-200MHz in 1MHz increments. CPU voltages are plentiful, with increments of 0.0125V for precise overclocking.
The board also ships with an SLI selector card. If you recall our nForce4 performance preview article, you'll remember that NVIDIA found that for best performance, it was best to split the PCI Express lanes evenly among cards, with both running as x8 cards. When a single card is installed it runs at the full x16 spec. The SLI selector card is used to toggle between the two modes: with the selector set in dual mode, the PCI Express interfaces both run at x8, when the selector is flipped to run in single card mode, the master PCI Express interface (which is the upper connector that is colored blue on our board) runs at x16.
[image]
| <% print_image("07"); %> | <% print_image("08"); %> |
Once the selector is set for dual mode, drop in your graphics cards, plug in the EZ Plug connector and power connections for the cards, and connect the two cards to each other with the provided SLI connector. NVIDIA has recently adopted a newer SLI connector that’s reminiscent of the old 3dfx SLI ribbon cable that they’ve dubbed the Flexfit connector.
| SLI hardware (cont’d) | Page:: ( 4 / 19 ) |
The reference GeForce 6800 GT cards were similar to the reference 6800 GT board we received from NVIDIA a few months ago, only they were based on a newer A04 board revision. Externally the board revisions look identical, and continue to sport dual DVI connectors.
[image]
| <% print_image("09"); %> | <% print_image("10"); %> |
Power
For supplying power to the system we used OCZ’s Powerstream 600W. This PSU is capable of delivering up to 700W under peak load, and features adjustable power rails in case you’d like to tweak the amount of power being delivered to the 12V, 3.3V, and 5V rails. OCZ has integrated a nifty adapter for converting from 24-pin to 20-pin operation. If you have a newer 24-pin ATX motherboard such as the ASUS A8N-SLI, simply slide the extra four pins into place on the power connector. If you’re using an older, 20-pin motherboard, simply slide out the extra pins and you’re done.
[image]
| <% print_image("11"); %> | <% print_image("12"); %> |
The Powerstream 600W also includes a 6-pin power adapter for PCI Express cards, as well as two specially sheathed Molex connectors which OCZ claims are designed to supply cleaner power to sensitive cards and disk drives. The other power connectors are neatly twisted to minimize cable clutter inside your case. OCZ even includes zip ties and Velcro strips for bundling things together.
NVIDIA initially recommended a 550-watt PSU during our first nForce4 briefing a few months ago, but we also ran some tests with a 500W unit and didn’t encounter any problems. Of course, at the same time it also doesn’t hurt to play it safe; that’s why we used the OCZ unit for the bulk of our testing.
| SLI software | Page:: ( 5 / 19 ) |
For SLI testing, we used NVIDIA’s ForceWare 66.93 driver which is currently available on NVIDIA’s website. This is the driver NVIDIA has certified for use with SLI, which is the reason why we decided to go with it rather than the newer ForceWare 67.02 driver.
Once you’ve got all the hardware needed for SLI up and running correctly, the ForceWare driver will prompt you that your system is SLI-capable. At that point, all you have to do is click the checkbox to Enable SLI and reboot the system, once that’s complete the system will come up again in SLI mode.
It’s important to note that NVIDIA’s cooling solutions for their GeForce 6 cards (including the 6600 GT and 6800 GT) are ducted designs that rely on nearby air for additional cooling. Because of this, the “Slave” card, which is located on the bottom in an SLI configuration (nearest to the PCI slots), will run cooler than the “Master” primary card, we even noticed this when the bottom PCI slot was populated with a sound card. This is because the slave card has access to more fresh air than the master card. Remember, sitting between the master and slave cards is the nForce4 chip.
ASUS also placed both PCI Express x1 slots between the two cards, so if you planned on running SLI with two GeForce 6800 Ultras, it probably wouldn’t be a good idea to populate either of those slots with an x1 PCI-E device.
NVIDIA continues to provide hardware monitoring functionality for SLI users. Just select the card you want to look at, and the driver will tell you its temperature. You can’t however overclock the graphics card’s individually; they must be overclocked as one unit, just as if you had a single card installed.
Unfortunately, NVIDIA doesn’t allow users to toggle between AFR and SFR modes. As you probably know, for some time NVIDIA has had custom profiles for many popular games on the market today. The driver uses these game profiles to determine which game you’re launching, and automatically selects the appropriate SLI mode NVIDIA has decided for that game.
For those who do want to see SLI in action, NVIDIA provides a load-balancing option in the driver. Once checked, you’ll notice a green horizontal bar running across your screen and a second vertical bar that runs from the top to the bottom of your monitor.
[image]
| <% print_image("13"); %> | <% print_image("14"); %> | <% print_image("15"); %> |
In SFR mode, the thin horizontal bar will dynamically move up and down your screen. This shows you in real time the split between the master and slave cards, as the upper portion above the line depicts the master card, while the area underneath the line is the slave card.
[image]
| <% print_image("16"); %> | <% print_image("17"); %> |
In AFR mode, the horizontal bar is static and located across the middle of your screen. Instead the vertical bar adjusts up and down your screen, indicating the current graphics load. When the vertical bar is filled, the cards are running at their optimal load for peak performance.
| How we tested | Page:: ( 6 / 19 ) |
System Setup
AMD Athlon 64 4000+
ASUS S8N Deluxe
1GB OCZ EL DDR400 Platinum Edition Rev 2
eVGA e-GeForce 6600 GT
NVIDIA GeForce 6800 GT reference card
Driver version 66.93
250GB Maxtor Hard Drive Maxline III SATA Hard Drive w/16MB Cache
Windows XP Professional SP1
DirectX 9.0c
Benchmarks
Lock On: Modern Air Combat (Mig-29 custom demo)
IL-2 Sturmovik: Forgotten Battles (The Black Death track)
Halo: Combat Evolved (stock benchmark)
Far Cry 1.2 (custom demo for mp_jungle, SM2.0b and SM3.0 paths used)
DOOM 3 (gameplay custom demo)
| Lock On: Modern Air Combat | Page:: ( 7 / 19 ) |
Lock On: Modern Air Combat – Direct3D
| IL-2 Sturmovik: Forgotten Battles | Page:: ( 8 / 19 ) |
IL-2 Sturmovik: FB - OpenGL
| Halo | Page:: ( 9 / 19 ) |
Halo – Direct3D
| Far Cry Volcano 4xAA/16xAF | Page:: ( 10 / 19 ) |
Far Cry – Direct3D
| Volcano HDR | Page:: ( 11 / 19 ) |
Far Cry – Direct3D
| Far Cry Training | Page:: ( 12 / 19 ) |
Far Cry – Direct3D
| Training HDR | Page:: ( 13 / 19 ) |
Far Cry – Direct3D
| DOOM 3 High Quality 8xAF | Page:: ( 14 / 19 ) |
DOOM 3 – OpenGL
| Canals_01a | Page:: ( 15 / 19 ) |
Half-Life 2 – Direct3D
| Canals_09 | Page:: ( 16 / 19 ) |
Half-Life 2 – Direct3D
| 6600 GT Overclocking | Page:: ( 17 / 19 ) |
DOOM 3 – OpenGL
Half-Life 2 – Direct3D
| 6800 GT Overclocking | Page:: ( 18 / 19 ) |
DOOM 3 – OpenGL
Half-Life 2 – Direct3D
| Conclusion | Page:: ( 19 / 19 ) |
In some games and applications, SLI can have a profound impact on performance. NVIDIA’s claims of nearly a 2X performance improvement were no lie – we witnessed similar gains in many applications, including cutting-edge titles such as DOOM 3 and Half-Life 2.
Our performance results with flight sims was mixed. We saw impressive gains with IL-2 Sturmovik: Forgotten Battles, with the GeForce 6800 GTs SLI’ed delivering very playable frame rates at 1600x1200 with 4xAA and 8xAF while the GeForce 6600 GTs running in an SLI configuration nearly matched the performance of a single GeForce 6800 GT. But in Lock On: Modern Air Combat, a more demanding, newer flight sim performance was unchanged. We have a feeling this had nothing to do with the SLI technology itself, rather the culprit is likely NVIDIA’s driver – the driver team simply hasn’t had enough time to optimize SLI for LOMAC. This is a shame, as LOMAC is the only modern air combat sim out there and has a loyal following of enthusiasts, but was probably passed over due to its relative obscurity in relation to other mainstream titles such as Microsoft’s Flight Simulator series. Also, while we didn't include official benchmarks in this article, we did test UT 2004 and Splinter Cell and found no performance improvement from SLI in either 6600 GT or 6800 GT configurations.
As long as the resolution is kept in check, two GeForce 6600 GT cards can deliver equal, if not better performance than one GeForce 6800 GT card, even with AA and AF cranked up. Once the screen resolution increases though, the GeForce 6800 GT’s superior memory subsystem kicks in and the single GeForce 6800 GT outpaces two GeForce 6600 GT’s SLI’ed together.
One additional side affect SLI provides besides performance is that new screen resolutions open up. For instance, 2048x1536 actually becomes a playable resolution. 2048x1536 has always been there for 256MB cards, only with SLI, 2048x1536 actually runs with fluid frame rates. And of course, once you’re at 2048x1536, you don’t need AA: jaggies are gone. While we didn’t run vanilla 2048x1536 benchmarks without AA, we have no doubt that many SLI users will want to give this resolution a try.
And now comes the hard part, price and availability. Purchasing two GeForce 6800 GTs, or even 6600 GTs isn’t cheap. But fortunately, one of the beauties of SLI is that you can purchase one card now, and a second identical later for nearly double the performance. GeForce 6800 GT SLI owners get a taste of tomorrow’s high-end card’s today, while GeForce 6600 GT SLI users will see what all the fuss surrounding GeForce 6800 GT is all about. It’s a win-win scenario for everyone but your bank account.
The second question, availability, is a bit more difficult to answer. We’ve been told that ASUS will be the first to have boards out, with Gigabyte and MSI to follow shortly thereafter and the smaller motherboard manufacturers shipping in Q1 2005, but many of the early motherboards will go to system builders such as Alienware, ABS, Falcon Northwest, and VoodooPC. Therefore, finding motherboards at retail will be difficult at first. The real problem is, we don’t know when retail supply will catch up with demand.
Fortunately, SLI is almost here. A couple of reviewers ran into stability problems with their preproduction A8N-SLI boards (our board corrupted our Serial ATA hard drive on two different occasions just recently), but most of the teething problems should be sorted out by the time final products hit retail. When that happens, we have no doubt that SLI will turn lots of heads.