To finish up our review of the Asustek nForce4-SLI hardware, we are going to explore what exactly the nVidia SLI technology offers you right now, the factors impacting SLI performance and then ask some questions to help those stuck in the rut of whether or not they should buy into nVidia SLI.
Make sure you check out Part 1 and 2 of the SLI article suite in which we reviewed the motherboard and graphics cards used in today’s article.
Birth of Multi-GPU Processing
If you don’t know already, nVidia SLI technology provides the ability to use multiple graphics cards to optimize performance in 3D applications by spreading the workload across several Graphics Processors (GPU’s). Although this technology, no matter how much the PR machine throws flashy ads and stellar performance numbers at you, it is not a new idea.
The first time that the computer enthusiasts industry heard of “SLI” was from a now gone graphics hardware company named 3Dfx. When they launched their Voodoo2 chip in 1999 the also brought a method of linking two identical Voodoo2 graphics cards together on a shared PCI bus in order to boost overall performance. They called this technology “SLI”, or “Scan Line Interleaving”. Like the name suggests, Scan Line Interleaving would take even screen lines and let them be rendered by one card, and take the odd lines and let them be rendered by the other card. This allowed for a much lighter load on each GPU and overall better performance. The technology rocked the industry.
Noticing the impact of multi-GPU performance gains several other companies tried their hands at developing their own SLI-esque technologies. One such company was Metabyte (Wicked 3D) and their “PGC”, or “Parallel Graphics Configuration”. PGC again used 3Dfx’s Voodoo2 cards or in some later cases an AGP card and a PCI card, but with a different method of splitting up the work. With PGC, one card renders the top half of the screen and the other card renders the bottom half. Again we see a large increase in performance as the load decreases at each GPU.
Although both of these technologies, 3Dfx-SLI and Metabyte-PGC, had pitfalls. Their biggest issue is not in the hardware but in the way they split up the work:
With 3Dfx-SLI, regardless of the fact that each card is only rendering half the number of lines on each frame, each card needs to know what is on each frame before they can render their lines. If a triangle edge has endpoints on an odd and even line, and that card is rendering even lines, it needs to know where the other card is putting the other endpoint before it can draw the triangle edge. This means both cards need the same data, they also need to do some calculation to know what is on some of the other lines or image quality suffers.
With Metabyte-PGC, it doesn’t take into account if the top or the bottom of the screen has more complex renderings. Therefore one card could be heavily loaded while the other is nearly idle. So the one card needs to wait for the other before it can move on to the next frame. In addition, when using PGC in the case of one card AGP and the other PCI, there can be significant image quality issues. Different cards render differently than others (Example: nVidia vs. ATI image quality), so the top half of the screen can look different than the bottom, and the screen tearing apparent from lines being improperly aligned.
Taking a different route, ATI introduced the Rage Fury MAXX graphics card in late 1999. This was a single card with two GPU’s that shared 128MB of memory (basically two 64MB cards on one PCB). ATI also implemented a brand new way to distribute the workload between GPU’s; they called it “AFR”, or “Alternate Frame Rendering”. With AFR one GPU would render the first frame, and the other the next frame. They would then continually alternate frames. This all-in-one hardware coupled with the AFR workload method provided the best alternative to date for multi-GPU 3D graphics. There were no image quality issues as each frame is uniform (rendered completely on one GPU, and any discrepancies are unnoticed by the eye as long as frame rates are high enough), and the workload from one frame to the next reasonably balanced.
Over time however, product competition, much better GPU’s, or tedious driver developments led to the demise of all of these products. Until now…
Part 3 – nVidia SLI
With the latest implementation of PCI Express on modern chipsets multi-GPU hardware has been brought to the forefront again. PCI Express offers an immense, scalable bandwidth. It runs on a full duplex (2way) operation instead of AGP’s half duplex (1way) meaning commands are given on the rise and fall of each clock. The inherent abilities of PCI Express make it ideal for high bandwidth accessory cards, such as graphics adapters.
nVidia has taken advantage of the PCI Express interface and introduced its own multi-GPU system. Using the term “SLI” which was acquired from their buyout of 3Dfx (late 2000), they have re-coined it to mean “Scalable Link Interface”. They claim massive performance increases, “…up to 1.9x” the performance using a single graphics card.To effectively use nVidia SLI you need three things: two “SLI-ready” nVidia graphics cards, an nVidia nForce4-SLI chipset based motherboard, and one of the latest nVidia Forceware drivers supporting SLI. Now first let’s set the record strait:
- “SLI” is not an industry term for multiple graphics cards working together. It is a name for nVidia’s proprietary multi-GPU technology. Other companies will likely use other names for their methods.
- nVidia’s multi-GPU system is not a rehashing of any single multi-GPU system (Example, re-using 3Dfx’s Scan Line Interleaving). It is more of a hybrid of several of the methods used in years past…
… and this is what we mean.
Most likely nVidia looked backwards to see what worked and what didn’t with previous multi-GPU systems. What they came up with (nVidia SLI) is a system that uses several different methods to split the work between the separate graphics cards depending on the situation; the situation being the game that you are running and the brain behind this selection being the nVidia graphics drivers. There are three methods of screen rendering in nVidia’s SLI solution:
- Alternate Frame Rendering (AFR) – Yes, a blast from the past; essentially exactly what ATI developed for the Rage Fury MAXX. Every other frame is rendered by one card, and the others are rendered by the other card. AFR is patent pending technology, so it is to be expected that there was a business agreement met between nVidia and ATI to avoid any intellectual property issues. This method provides the highest performance boost.
- Split Frame Rendering (SFR) – This method is very similar to Metabyte’s PGC technique, with a twist. The screen is initially split at the middle; one card renders the top 50%, the other the bottom 50%. As the content on the screen changes, algorithms determine the complexity of certain parts of the screen and change the % distribution so that both cards take roughly the same time to render their section. In addition, since the cards are identical you avoid differing image qualities and screen tearing if they operate properly.
- Compatibility Mode – In this mode only one card is used and the other is completely idle. This is the same thing as running without SLI at all.
Now since the drivers are selecting which mode to operate in, they need already know about the game you are running before you run it. So built into the drivers are a list of settings for games known as “Game Profiles”. There are many game profiles built into the current drivers but only some of the games effectively use AFR or SFR rendering mode, more on that later.
Setting up nVidia SLI hardware is pretty easy; follow your SLI motherboards instructions. In the case of our Asus A8N-SLI Deluxe, pullout the EZ-Selector card and insert it in the “Dual Video Cards” orientation, plug one card into to the blue primary PCI Express channel and the other card into the black secondary channel, then connect the “golden fingers” of both video cards using the supplied PCB adapter. Once finished it should look like this (and another good job to Asus on providing ample room between the cards, many SLI motherboards are very cramped):
After installing the hardware, check your BIOS for any setting to enable SLI (in most cases set it to AUTO). Boot into Windows and if you have the latest nVidia Forceware drivers installed you will be met with a pop-up bubble message.
Then you can right-click, go to Properties/Settings/Advanced/GeForce-Tab. Another menu should pop out to the left of the window and if your hardware is installed and recognized correctly you should see a “SLI multi-GPU” submenu. Click it and then click the “Enable SLI multi-GPU” selector. Reboot and in Windows XP you will be met with a pop-up message in the taskbar telling you have successfully enabled SLI.
Note: One thing I noticed right off the bat was that while dual graphics cards were enable for SLI, multi-monitor and TV-Out feature did not work properly. This was not only with the Forceware 66.93 drivers, but all the way up through the 71.20 Beta drivers. When TV-out was enabled there was no video output in windows. When you used the right-click “Play on my TV” feature, all monitor outputs would lock up requiring a system reboot. To get around this you must unclick “Enable SLI multi-GPU”, and then reboot into single card operation. After doing so, TV-out worked normally. Hopefully this feature gets fixed soon.
GPU Load Balancing
A neat feature included in the Forceware drivers is a little tick box titled “GPU Load Balancing”.
If you turn this feature on the drivers will plot a series of graphical lines on the screen when running SLI rendering modes in DirectX or OpenGL applications. The way the lines are used are different between AFR and SFR modes, so will easily be able to differentiate between them.
The first screenshot is of Doom 3 which runs in AFR. You can see the screen is split in half by a green line, and on the left there are two white lines with thick green bars coming out from the centerline. In AFR mode the thick green bars will extend outward towards the top and bottom of the screen to indicate the percentage of GPU usage in each card. One card being the top half and the other the bottom.
The second screenshot is of Far Cry which uses SFR. Here the thick green lines are totally filled in. The green center line however is skewed towards the bottom of the screen. This line will move up and down the screen indicating the approximate place where the current frame is being split for rendering. One card renders above the green line, the other below it.
Very nice feature to use if you need a quickly see if SLI is working correctly, what mode you are in or if you’re just curious to see how SLI is rendering your game, now let’s start testing.
To best test the Asus SLI setup we will compare using one Asus Extreme N6600GT (with motherboard in single graphics card mode) against using two EN6600GT’s (with the motherboard in dual graphics card mode, SLI). nVidia keeps a list on their website of which applications are currently compatible with SLI and have SLI profiles written for them. You can see the list here. We chose applications on this list for testing so we can compare our results against the nVidia claimed, “…up to 1.9x” the performance of a single card.
Asus A8N-SLI Deluxe (nForce4-SLI, BIOS 1003.06 Beta), AMD64 3200+ Winchester, Crucial Ballistix Tracer PC4000 2x512mb (DDR400 @ 2.0-2-2-5), Asus EN6600GT’s (Forceware 66.93), WD Raptor 360DG 36GB, WD 1600BB 160GB, Lite-On DVD-RW SHOW-1213S, Windows XP Pro (SP1 and DX9C)
Synthetic Performance
Synthetic benchmarks attempt to simulate and test real world 3D performance across a large set of hardware and programming features. These benchmarks generally run a series of pre-recorded sequences and measure frames per second in each sequence. Depending on the program and the proprietary formula they use, synthetic benchmarks will output either a number score or an average frames per second value to use for performance comparison against other systems. Synthetic testing is done using program default settings, as these are generally what the community uses to compare different systems to each other.
Futuremark 3DMark03 and 3DMark05 – Default
For both 3DMark03 and 3DMark05 Alternate Frame Rendering mode, the more efficient of the two SLI rendering modes, is used. Here you see almost a 62% increase in score when going from single card to SLI setup in 3DMark03. The scores get even better in 3DMark05 with almost a 69% increase. When you think about the fact that as the 3DMark series develops, they every increasingly focus on graphics hardware performance, the transition in scores from ’03 to ’05 makes sense. One impressive note is the fact that during one scene of the 3DMark03 “Wings of Fury” test, FPS broke 1000 for several seconds.
Massive Development Aquamark3 – Default
Aquamark3 also uses Alternate Frame Rendering mode. Here we see almost a 19% increase in average FPS when using the SLI setup. Not quite as high as the 3DMark series but still noteworthy.
If the 3DMark series is any indicator to how the Real World Performance should be, well then we would have a very impressive setup on our hands.
Real World Performance
Real World benchmarks are just that. They run a sequence recorded strait from game play (generally called a “timedemo”) from a game that you can run out to the store and buy. At the end of the timedemo an average frames per second value is reported. This value gives a good indication as to how a hardware setup with particular driver settings will perform with that specific game. Testing of Real World benchmarks is done using two different resolutions (1024×768 and 1600×1200) and two different image quality settings (2xAnisotropic Filtering/No Anti-Aliasing and 8xAniso/4xAA).
iD Software Doom 3 – OCmodshop Guardian Demo
Doom 3 uses Alternate Frame Rendering. You can see that SLI doesn’t do much good at low image quality and low resolution in this game. However when image quality is turned up at low resolution, or as soon as the resolution gets much larger, SLI shines with a 52% increase in frame rates at low resolution high image quality and a 47% increase in performance at high resolution and low image quality. Not an extra 90% in performance, but still good. Frame rates at high resolution and high image quality were still barely playable even on the SLI setup.
Valve Half Life 2 – OCmodshop Coastal Firefight Demo
Half Life 2 also uses Alternate Frame Rendering mode like in Doom 3. However you see a completely different scenario here. At low resolutions the SLI setup receives roughly a 3-6% performance deficit over the single EN6600GT. The same performance streak continues up until the high resolution and high image quality where the SLI setup breaks into the lead with a 16% increase in frame rates. These results were confusing at first however it should be noted that tests were repeated with the same results.
Ubisoft Far Cry – HardwareOC Steam Demo
Far Cry uses Split Frame Rendering mode. Here we see a similar situation to Half Life 2 where at low resolution and image quality the SLI setup trails by over 17%. However as soon as image quality settings are turned up on low resolution, SLI takes a commanding lead with over 68% performance increase. The same follows suit with a 38% increase at high resolution low image quality, but both setups seem overwhelmed at high resolution and high image quality. Both setups are unplayable that these settings.
Epic Games Unreal Tournament 2004 – Benchemall Recorded Demo
Unreal Tournament 2004 is yet another game to use the more efficient Alternate Frame Rendering mode. Here, again, we see the SLI setup stall a bit at low resolution and low image quality. Though when scaling resolution and image quality, this test seems to be what one would hope to expect from SLI performance. You see over an 8% increase at low resolution high image quality, over 24% increase at high resolution low image quality and an astounding 81% increase at high resolution high image quality! Although even with such a large increase both setups would be very playable at high resolution and high image quality.
Overall the real world gaming performance seems like a mixed bag, and not the shining star that the synthetic benchmarks would lead one to believe. I was quite disappointed in how in many cases the SLI setup seemed to be hindered in the low resolution tests. Curious, I tested a game that was not on the nVidia SLI approved list, Commanche4, which should not use any form of SLI rendering.
In this high test you can see that in both high and low image quality at high resolution, the SLI setup trailed the single card by roughly 1-2%. This may be due to the fact that there is more hardware/software overhead to control the SLI setup as opposed to the single card, even when only using one GPU.
What we find in the end is that some of the tests do offer good insight as to what SLI could offer, but many exploit some of the deficiencies of the techonology. With the minimum performance difference being a 17% deficit (but still playable) and the highest difference being 81% increase (very near the nVidia claim of up to 90%), it tends to perk some questions as to why exactly the performance may be so muddled at this point in time.
As was mention earlier in this article, for nVidia SLI to work, the drivers have to have a profile pre-written before it can recognize an application and then use one of the SLI methods. So for new games to be recognized, the drivers need to be updated before obtaining any SLI benefits. Here is the list of games that nVidia currently has multi-GPU modes enabled in the drivers for (taken when article was written, full list here, list excludes compatibility mode):
|
Aquamark3
|
Halo
|
|
Battlefield 1942 Desert Combat Mod
|
Painkiller
|
|
Battlefield Vietnam
|
Sims 2
|
|
City of Heroes
|
Star Wars: Knights of the Old Republic
|
|
Code Creatures Pro
|
Tomb Raider: Angel of Darkness
|
|
Colin McRae Rally 4 Demo
|
Tron 2.0
|
|
Command & Conquer Generals
|
Unreal Engine 3
|
|
Doom3
|
Warhammer Dawn of War
|
|
Far Cry
|
3DMark 2003 Gold
|
|
Ground Control 2
|
3DMark 2005 Gold
|
|
Half Life 2
|
You don’t need a masters degree to notice that’s a short list. On the date this article was written it only contains 16 games, 4 benchmarks, and 1 game engine that doesn’t have any marketed games yet. nVidia has stated however that they “…will update the list frequently as new games are released and as we complete more testing and tuning on currently released games.” Now since this hardware was received for review, there hasn’t been any additional games added, or new drivers released. Not a very good sign for a hardware that needs software support to be effective. In response to this, nVidia has released details on how to edit and add new profiles to your existing drivers, allowing you to add SLI support for applications not in the drivers, and new applications that come out. You can find nVidia’s details on how to edit profiles here, and our quick walkthrough below.
|
1. The first step is to find the file that contains all of the profiles. The file is called nvapps.xml and in Windows XP is located in the X:\WINDOWS\System32 folder. You can open the file up in WordPad or any other application that opens XML files. Inside you will find a list of profile information labeled using a tag system similar to HTML. A profile starts with a label tag that looks like : <PROFILE Label =”Game”> and ends with a </PROFILE> tag. Inside the profile label are the game executable identifier tags and settings for the multi-chip rendering mode. 2. To create a profile, copy and paste the following after any other profile in the nvapps.xml file: <PROFILE Label=”Profile_name“> 3. Replace the following values: 5. Test the application; use GPU Load Balancing in the display drivers to check the SLI mode. As an example I used the game Nexus – The Jupiter Incident, a brand new game with some killer graphics that can have some low frame rates especially on large resolutions. For this game I inserted the following profile for Split Frame Rendering mode: <PROFILE Label=”Nexus”> |
After rebooting I tested a generic flyby sequence of about 60 seconds of game play, recording the average FPS. Overall using SFR mode there was a gain of about 10 FPS. Not too shabby.
So overall it sounds pretty easy, so copy and paste, that’s all it takes to add a new application to the driver profiles, right? It’s too bad however that it didn’t turn out so simple.
Not So Cut/Paste
Anyone who has read through the nVidia guide to SLI profiles, and tried it for themselves knows that there is a lot more information regarding SLI profiles that nVidia is not telling us.
For one, the Nexus – The Jupiter Incident example used above, when using AFR mode, there was no change in average FPS over using a single card. Now AFR supposedly is most efficient mode that should logically produce the highest frame rates.
To give you another example, my curiosity hit the roof when looking at the list of nVidia Approved SLI applications. Seeing 3DMark03 and 3DMark05 on the list, where was the still popular 3DMark2001SE? I opened up the nvapps.xml file, and actually found a profile for it. However it wasn’t using any of the prescribed mode annotation in the nVidia Applications Profile Guide (1 – AFR, 2 – SFR, 4 – Compatibility mode). It had instead “C00001”, a 6 character string which resembles Hex values. To my surprise there were many other games using similar character strings to denote the rendering mode. After testing 3DMark01SE with GPU Load Balancing on, it seems that the value “C00001” denotes a compatibility mode, as no load balancing lines were present at all. The following is a list of possible SLI modes if you go off of what is contained in the profiles of the nvapps.xml file (game listed if used to identify rendering mode):
|
SLI Mode
|
Profile Designations
|
|
AFR
|
1, 2400401 (3DMark05), 3C00401 (Half Life 2), 400001 (3DMark03, UT2004)
|
|
SFR
|
2, 4C00002 (Far Cry)
|
|
Compatability
|
4, C00001 (3DMark2001SE)
|
|
Unknown
|
00020201, 00400001, 2431001, 2C10401
|
So now that I know that nVidia isn’t telling me the whole story on how to select rendering modes, I decided to trudge forward and alter the profile for 3DMark2001SE. I used the standard Compatibility Mode (“4”), AFR (“1”), and SFR (“2”) rand the benchmark and compared the scores.
As you can see, using AFR increased scores rather well, while SFR they dropped dramatically, almost 29%! More importantly, the image quality suffered greatly on both SLI rendering modes. Even though AFR produced higher scores, there were light artifacts in several game tests. Then when using SFR mode during one point in the Lobby – High Detail scene, I was shocked to see what looked like only one video card working during several scenes. The GPU load balancing lines were present during these scenes but anything below the Split Frame line was black. It was definitely a very odd situation to witness.
Yet Another Oddity
Recently I had a chance to beta test ArenaNet’s Guild Wars, a brand new game that should be launching in several months. I had been doing some profile testing on other games and had left the GPU Load Balancing setting on by accident. When I loaded up Guild Wars, to my surprise the green bar was twitching away in Split Frame Rendering mode. I quickly opened up the nvapps.xml file and found no profile for Guild Wars.
How could this be? According to nVidia, it couldn’t. So somehow Guild Wars was defaulting to SFR mode without the use of profiles within the drivers. Curious I decided to make a profile and try out several different mode settings.
At default with no profile the game would average around 30-35 FPS at 1600×1200 with all detail and 4xAA. After the profile was added and set to SFR mode, average FPS was the same. Using AFR the average FPS rose to 55-70, although the game would lockup randomly during game play. Next, the hex strings for AFR used for 3DMark03 (400001) produced frame rates of 60-80 (lockups after several minutes), and the hex string for 3DMark05 (2400401) produced frame rates of 65-140 (quick lockups). The 3DMark05 settings also produced much more GPU activity than any of the other settings. So why are the results so different if they are all supposedly using AFR? I believe the answer is locked inside of the short hex strings for those games nVidia has tested.
In one word – Infant.
The technology has proven it can shine under the right circumstances, but fall short in others. To me it just doesn’t have the rock solid stature that a product should have to be released to the public. I think nVidia may have been a bit premature with this one. It seems as if there are a lot of holes in the SLI equation as it stands right now: Poor low resolution performance, low compatibility with current and old games, incomplete profile creation settings, and unstable results with new games. In no way is this technology plug and play, unless the only applications you play fall on that list of 15 applications nVidia has approved for use with SLI.
So if you are one of the gaming community who enjoy older games not on that list, or if plan on playing new games as they come out and want to take advantage of nVidia SLI performance, one of two things needs to happen:
|
1. nVidia needs to step up to the plate. It has been quite a while since SLI hardware has hit the market and there has been no sign of a framework for new application profiles to be updated or installed. It’s not looking good in this area since history has proven that with nVidia you can wait many months before new drivers are released. 2. More information regarding the mysterious hex string SLI modes and any other miscellaneous settings needs to be released to the public. Since it doesn’t seem nVidia hasn’t been keen on releasing new drivers or profiles, the majority of profile creation has been left to tinkering in the nvapps.xml file. If new information regarding those mysterious hex string SLI mode settings is released, maybe the community can fill in where nVidia has dropped the ball. |
A novel short term idea would be a website where nVidia collaborates with the community for profile testing and releases. Stable profiles from nVidia could be released among profiles with which the community has developed. Their status could be labeled as either stable, buggy, or unstable. The profiles themselves could be boiled down to individual files for individual games and then downloaded installed separately.
But for now were stuck asking many questions, to the only ones who know the answers, nVidia:
|
|
nVidia, I and many others will be waiting for your answers.
A final question…
Should I buy it, and if not now then when? Sound advice to everyone reading this is, WAIT!
Multi-GPU technology shows a lot of promise. It clearly can be quite useful for helping GPU performance. nVidia has put a lot of effort into SLI and has even coupled it with their new server and workstation line of professional chipsets, so I doubt they will let this early infancy affect how they pursue their SLI technology developments. Though with other companies like ATI (ATI’s version is called AMR – ATI’s Multiple Rendering) and even chipset companies like Via (Dual GFx) showing interest in multi-GPU hardware, there should be some hefty competition in due time.
I have heard some say that using two 6600GT’s in SLI operation would be a good alternative to using a single 6800GT. I hope that this article helps to dissuade those thinking of going in that direction. The only situation in which I would feel fine suggesting an nVidia SLI setup to anyone now would be this:
If you are considering going SLI in the future, maybe once the software support is more full featured, it wont hurt to spend the extra money on an SLI motherboard now, but there’s no need to buy the second video card. Or… If you have money to blow, love being on the bleeding edge, and have no problem tweaking and testing to get SLI to work properly on your favorite games that nVidia has not supported yet.
Conclusion
In testing we hit close to the 1.9x performance against a single card on UT2004. However software support and strange bugs seem to plague those who want to write their own profiles for applications that nVidia has not supported yet. It’s a mixed bag of performance and issues that need to be worked out. However it shows a lot of promise, and if you don’t mind spending the extra dough to stay bleeding edge with your gamming performance in 15 games, nVidia SLI is the only route to go.
Overall from a product standpoint I feel Asus has provided a top notch setup of hardware. The Asus A8N-SLI Deluxe is a solid board with tons of features. Coupled with two rocking Extreme N6600GT’s and you have a top of the line SLI setup providing some extra performance over a single EN6600GT and keeping the frame rates rocking at high resolutions. The issues covered in this article pertain to the nVidia SLI technology itself, so I have no problem giving the Asus A8N-SLI Deluxe motherboard and Extreme N6600GT SLI setup 9 out of 10 and the OCmodshop seal of approval. I would like to Asustek for send us this SLI hardware setup for review.
Pros:
Cons:
|