Alienware Discuss New Video Array Technology For Gamers

Gaming Nexus writes "Over at Gaming Nexus, we've posted an interview with Alienware about their new video array technology, which 'will provide gamers with an expected 50% increase in gaming performance by utilizing two video cards.' The interview covers the creation of the technology, the problems they had in developing it, as well some more details on how it works." The short version is that it utilizes multiple cards to render one screen, similar to SLI, but with many more features added in as well. What Alienware has developed is a software layer that sits between the video drivers and the application, routing things to where they need to be.

Comments

AGP Slots

[SirChris]

Where Do i get the other AGP Slot from?

Re:AGP Slots

[Rubbersoul]

RTFA!

GamingNexus: Was this something that you couldn't do with AGP or had you considered doing something with AGP?

Brian Joyce: We actually had a working prototype with AGP. But as soon as it became clear that PCI Express was going to become the industry standard, we had to start re-working it for PCI Express.

Re:AGP Slots

[u-238]

Custom made motherboards with two PCI Express slots. Saw the actual board on TechTV a few weeks ago.

Still not sure whether they've patented it or not - hopefully not so we'll be able to but these mobo's from other vendors and build these rigs ourself without paying alienware an extra $1500 for unnecessary services.

Re:AGP Slots

[Paladin128]

RTFA: Multiple patents pending on the technology. Likely not on the concept of having multiple PCI Express x16 slots, just on the software and compositing/sync hardware.

Re:AGP Slots

[drinkypoo]

I don't know if anyone has actually developed a board like this but AGP8x supposedly supports dual slots. However, if no one has yet done it, probably no one will, because AGP and PCI are both going down the tubes in favor of PCI-Express, which scales in both cost and performance to levels both below PCI and above AGP. (The cost of a wide PCI-Express might well be below the price of AGP8x, actually.)

I... guess....

[Ieshan]

Doom 3: $60
Dr. Pepper and Potato Chips: $5
Alienware Super Extreme Gaming System: $10,000

Having the "Sorry, I'm broke" Excuse to Avoid Going out on Weekends and Playing Computer Games Instead: ...Priceless?

There are some things students can't afford. For everyone else, there's Alienware.

Re:I... guess....

[vmircea]

I'm going to have to agree with this... If you want the latest technology and awesome things you should go with Alienware, but you better be prepared to pay for it... as it doesn't come cheap. But with some of today's really good systems, you can get really good FPS on almost any game, for significantly less money than an Alienware. Which causes one to wonder... how much is too much...

Re:I... guess....

[kLaNk]

Which causes one to wonder... how much is too much...

You, my friend, are not a gamer.

It's PCI-X.

[stoneymonster]

They aren't using AGP. If you'd read the article, you'd know that ;)

Voodoo 2

[flok]

Multiple cards for one image?
Back to the days of Voodoo 2 cards :-)

You mean PCI-E !

[LordPixie]

PCI-X being the server standard.

No, not that it really matters. And yes, I'm being overly anal.


--LordPixie

Multi-cards vs multi-heads

[Kevin Burtch]

Didn't this only last about 1/2 of a "generation" the last two times it was attempted?

"Two?" you say?
Yes, the obvious one is the old Voodoo 1 & 2 cards, but I distictly remember at least one (I think 2 or 3) company(ies) making cards that used 3 S3 chips (one processing each color) for a performance boost.
They were all "really hot" (popularity, not thermally... well, ok both) for a very short period of time, since the next full generation of chips completely blew them away.

It was silly then, it's silly now.

Now what _I_ want, is a triple-headed system that you can play FPS games on with a front and two side views (peripheral vision, or at least just a wider landscape in 2 or 3 monitors). The hardware is there (well, for dual at least), but do any games support this?
It _can't_ be that off-the-wall, after all, the SPARC version of Doom supported triple-heads way back in version 1.2! (they dropped it after that)
OK, that wasn't *exactly* the same thing... that required a different box for each of the left and right displays, but they acted as a slave so you only operated the center system... it was _extremely_ cool!

Hmmm... I wonder how long it'll be before 16:9 displays are common, the only one I know of is the sweet monster made by apple that costs as much as a used car!

Re:Multi-cards vs multi-heads

[drinkypoo]

There is a three head video card right now, it's called matrox parhelia, and it's overpriced and underpowered like everything else matrox has ever made. Still, the visual quality is supposed to be quite good and if you don't pump the resolution too high (I wouldn't use more than about 800x600 per display tops) you should be able to get good performance out of most games. Games which support arbitrary resolutions are supposed to support it automatically (as they do with nvidia twin displays) and some games support it directly, like toca 2 and a number of flight sims.

Re:Multi-cards vs multi-heads

[chrismcdirty]

The Matrox Parhelia was advertised to use 150 degrees of vision in games, utilizing 3 monitors.

Go here and check out the TripleHead Desktop table.

Re:Multi-cards vs multi-heads

[gl4ss]

agp killed it.

plus that it really makes more sense have the power in one card anyways if you're getting it, so the market for these is a niche one.

cool software anyways, kudos to them and yadda yadda.

Re:Multi-cards vs multi-heads

[uhhhhhhh]

Don't have anything to do with video cards but... Star Wars Galaxies has a way to increase your field of vision. You can also change the size of the game window I believe. It requires some messing around with config type files I think but I know I saw someone say that it was possible to get a full 360 degree view. But it would either be really really quished or you would need multiple monitors to span it across.

Re:Multi-cards vs multi-heads

[jpmkm]

Matrox cards have never been about power(at least not 3d acceleration power). Their main focus is on image quality. Not everyone needs super-fast 3d acceleration.

Re:Multi-cards vs multi-heads

[Mr. Shiny And New]

I didn't read the article, but from what I understand the technology is somewhat more generic than the Voodoo SLI. If that's the case, then, eventually, as the technology matures, you'd be able to upgrade your two video cards to get better performance. Sure, the next generation of cards may be faster than two of today's cards. But two of the next generation's cards will be twice as fast as one. And eventually, maybe you'll be able to add as many video cards as you want, in order to make your system faster and faster. Video rendering is an inherently parallelizable problem, and if this was a generic parallelizer, it'd be worth the money.

Re:Multi-cards vs multi-heads

[drinkypoo]

This is the games section. We're talking about a card for gaming. Further I owned the original Matrox Mystique and I still have a Matrox Millenium 220 when I need a reliable and high-quality 2D card for a project, so not only is your comment irrelevant but it's also not saying anything I don't know.

Further, matrox claimed that the parhelia would have performance competitive with or superior to ATI and nVidia offerings which were out at the same time. Well, as it turns out, the card shipped late, but even if it had shipped on time this wouldn't have been true.

So: You are correct but it was not necessary to tell me and not germane to the conversation, and further Matrox had claimed that their focus WOULD be on performance. Since they discovered they couldn't manage to create a GPU which would compete with the Big Two, they had to settle for their fallback position and whine about how their card had better image quality because it couldn't pull the performance wagon. It's sad when they did it, and it's sad while you're doing it. If you need professional 3D graphics in a PC, you don't go to Matrox, you're using a nvidia quadro or some oxygen-based card. If you need high quality 2D graphics, you don't need a parhelia, though it will do the job - it's just vastly overpriced for it. You'd be better served by two or three Millenium 8MB cards, except of course that both Apple and Microsoft are going to end up requiring 3D acceleration for their GUI sooner or later.

Re:Multi-cards vs multi-heads

[Kevin Burtch]

"Video rendering is an inherently parallelizable problem..."

Um, no... not at all.

Think about how much bandwidth is needed just for the CPU of the system to feed the graphics hardware that is doing all the work (AGP 1x, 2x, 4x, now 8x and the new PCIx, etc.).

Rendering on two boards means _4x_ the traffic over the bus!
Don't believe me? Think about it.. your CPU has to feed card A _and_ card B (2x so far).
Then, since you're only displaying on ONE of them, card B has to xfer the rendered display back over the bus to the CPU, then the CPU has to send it back out to card A. Even if card B could directly talk to card A (maybe, my Bt878 card does), that's still 3x the traffic over the bus!

Maybe this is why they claim only a 50% improvement in performance by doubling the hardware.

Doesn't sound very "parallelizable" to me.

Re:Multi-cards vs multi-heads

[CityZen]

They are using gen-locked video compositing, so there's no pixel traffic at all. The video outputs from both cards go into a combiner card, which switches between the two at the horizontal retrace interval where the screen division is.

There was a picture of the setup at some other site where this system was mentioned a couple weeks ago.

Re:Multi-cards vs multi-heads

[CityZen]

Two cards are not twice as fast as one. The two cards are indeed faster than one, but not twice as fast. That's because screen-space subdivision cannot be 100% efficient; there's always duplication of work for geometry that falls across the border. And indeed, if your APIs don't support efficient culling, then perhaps all of the geometry may have to be processed twice, degrading your efficiency even more. As well, the more borders you have (by having more graphics cards), the worse your efficiency gets.

Rendering is indeed inherently parallelizable, but most of the "efficient" parallelization is already being done on a single graphics cards. They've got multiple processors working on the geometry vertices, and multiple processors working on the pixels. (Still probably just one tiler working in between those two steps, though.)

The further "up" the pipeline you go, the less efficient parallelization tends to be. Of course, it all really depends upon how you subdivide the work. People have written PhD theses on this problem, in fact. Just do a google search on "sort-first, sort-middle, sort-last" to see some ways of going about the problem.

Re:Multi-cards vs multi-heads

[Mr. Shiny And New]

Just because transferring data to two cards is not easily parallelizable doesn't mean that rendering a scene is also not parallelizable.

First of all, most devices can access other devices and memory without the CPU being involved. This is what DMA and its ilk are for. Secondly, and I don't know if this is possible or not, but it's possible that two devices on a bus could both be written to at the same time, since they both are listening to the bus at all times.

Anyway, the realities of implementing a solution to a problem always step in to impose limits on things; adding a second CPU to your computer doesn't double your performance either, except in rare circumstances. But my argument still stands: having two GPUs working in parallel will be faster than one GPU. Your description of all that bus traffic is over-simplified anyway; the data to the video cards is different than the data from video-card A to video-card B, and is potentially smaller. Besides, saturating the bus only matters if the bus is your bottleneck in the first place.

Re:Multi-cards vs multi-heads

[kLaNk]

I am not an expert on rendering theory, but:

To the best of my knowledge 3d rendering is all about matrix operations. Matrix operations are inherently "parallelizable". Period.

What the heck does the bus have to do with any of it? Yes, there are going to be bottlenecks, but that doesn't mean the the original problem can't be easily broken down into discrete chunks. Inter-operation communication is an issue to be sure, but it isn't the driving force behind whether or not a algorithm can be processed in a parallel manner.

Re:Multi-cards vs multi-heads

[Kevin Burtch]

"What the heck does the bus have to do with any of it?"

Try unplugging your AGP card and see how well it works.

If the bus were so unimportant, why the heavy focus on the BUS since the original VGA card?
ISA->EISA->VLB->PCI->AGP->AGP2x->AGP4x->AGP 8x->PCI x...

If you would actually READ what I posted, you'd understand that first you have to TRANSFER the data to be rendered, then you have to TRANSFER the rendered image BACK so you can TRANSFER it to the other card again.
OK, so another poster indicated that the re-merging of the data is done externally, that eliminates the issue with the secondary transfers, but you STILL have to transfer twice as much data s for a single card.
If you don't understand this, you REALLY DON'T understand how computers work.

IPC has NOTHING to do with hardware data transfers over a bus. Throwing a big sounding term around doesn't make you smarter.

Re:Multi-cards vs multi-heads

[Kevin Burtch]

"First of all, most devices can access other devices and memory without the CPU being involved. This is what DMA and its ilk are for."

Sorry. While it is true that DMA allows one device to talk to another device without the assistance of the CPU, it still requires the transfer take place over the bus that the devices are plugged into.
If the bus is a bottleneck with one card, two makes things much worse.

"Secondly, and I don't know if this is possible or not, but it's possible that two devices on a bus could both be written to at the same time, since they both are listening to the bus at all times."

Nope, not at all. That's not how a bus works... each device is addressed, 3-way calling is not available in this calling area.

Re:Multi-cards vs multi-heads

[kLaNk]

After a card has rendered a frame/scene, there is no data that needs to be return; it has performed its job. Furthermore, during rendering, there is no need for card A to communicate with card B and vice versa. Each works on the dataset assigned and then is done. Granted there will be some lighting/primitive overlap which is wasted cycles basically (since both cards will be computing the same information) but that really isn't here or now and has to do with how discrete you can break up the entire dataset NOT whether the problem can be worked on in a parallel manner.

I will touch briefly on the subject of the BUS traffic since you seem to have a certain infatuation with it. Certainly there is added information that gets transferred over the BUS when you do this. I haven't denied this and I don't think that anybody else has either for that matter. But I'm certain that number isn't doubled. I'm relatively certain that the amount of data pushed to a video card is linearly dependant on the complexity of the scene being rendered. This being the case, with each card rendering approximately half of the scene (based on complexity) I bet the actual BUS traffic isn't even close to doubled. But this isn't even really an issue as a platform could be made with a 1:1 relationship between video cards and BUSes.

But this doesn't touch on the real issue that started this entire thread. The question of whether or not 3d graphic rendering can be processed in a parallel nature easily. Like I mentioned in my previous message, the actual act of rendering a scene is nothing more than a ton of matrix transformations. You didn't disagree with me on this point so I will assume we at least share a mutual agreement on this data point. That being the case, matrix transformations are the textbook example of a problem that maps VERY well into a parallel computing paradigm. Different parts of the same matrix transformation can be computed at the exact same time without a dependency on any other parts, sounds perfect to run in parallel doesn't it!

So, in recap, if you are saying that with a standard PCI-X or AGP off the shelf motherboard there would very real bottleneck in the BUS causing very real diminished returns for large (maybe not even that large) number of video cards, you are certainly correct. But, if you are talking about graphic rendering in general (which is what I have been talking about), you are unfortunately wrong.

Throwing a big sounding term around doesn't make you smarter.

Awwww, shucks..I wassa sure hopin' it wud.

Re:Multi-cards vs multi-heads

[Kevin Burtch]

"After a card has rendered a frame/scene, there is no data that needs to be return; it has performed its job."

As I stated in at least 2 postings, if the video is not recombined with extra hardware downstream of the rendering (which as one poster indicated it is, making this a moot point), then the rendered data (which is drastically larger than the pre-rendered data) must be merged with the other card's rendered data in order for it to be displayed. This would require a data transfer over the only medium available to both cards: the bus to which they are both attached.
Again, this system (as another poster pointed out) does merge the data externally, so it does not have to make this transfer... so the point is moot.

"So, in recap, if you are saying that with a standard PCI-X or AGP off the shelf motherboard there would very real bottleneck in the BUS causing very real diminished returns for large (maybe not even that large) number of video cards, you are certainly correct."

That is precisely what I'm talking about, since that is what the article is about (y'know, the whole "on-topic" thing?).

"But, if you are talking about graphic rendering in general (which is what I have been talking about), you are unfortunately wrong."

Well yeah, if you have multiple chips doing parallel rendering in a manner in which they share the load without the bus being a factor (like the old 3-chip cards I mentioned in my first post), then it definitely helps... but that's not what the article was about.

ugh

[mwheeler01]

excuse me while I bang my head against my desk. I know it's important to keep innovating but this is why PC games are lagging behind consoles...because soon you'll need 3 video cards to run the latest ID game.

Re:ugh

[merdark]

Umm, you do know that the consoles use the same video hardware as the PC right? And you also know that ID makes games for both PC and console now right?

PC games are not lagging because of hardware. There are other factors, like price for instance (consoles are much cheaper for the same hardware), playing on a TV, controllers, less console warez... etc.

But it certainly is not due to the technology, since it's the same.

Re:ugh

[_Sexy_Pants_]

Um... wrong. ID develops PC Games exclusively, and develops them for PC hardware. Doom III for XBOX is being developed by Vicarious Visions, who did Tony Hawk ports, and work very closesly with ID. Carmack codes to push the limits of the systems available. Some games are reaching even further than that, such as Unreal 3, a game which couldn't be demonstrated until a working Nvidia 6800 was available, and it still runs at a terrible frame rate. The simple answer here is that PC is not lagging behind consoles. And when they are, it's because consoles have a very specific set of hardware, when on a PC you have to make sure EVERYTHING works. But I'll tell you this - on a high end machine, it'll be running at a higher resolution with more features and a higher frame rate.

Re:ugh

[merdark]

Vicarious Visions may be *porting* doom 3 to the console, but it's still the same game engine as the PC version and it still uses the same types of graphics hardware extensions, hence ID is developing games that target both systems. Not sure what you are trying to achieve here by playing the technical fact game.

Unreal 3 is not out yet, and next generate consoles will be using next generation graphics hardware as well, so there is a good chance that the new unreal engine will run on consoles as well.

I'm not sure why you are trying to be so agressive in your post considering you are agreeing with me that hardware is not limiting the PC in any way.

The consoles ARE getting more games than the PC these days, but it's not due to hardware, but rather a number of other factors including set hardware specs as you mention.

Re:ugh

[_Sexy_Pants_]

I'M BEING SO AGGRESSIVE BECAUSE I misinterpreted your post. *extends hand*

Re:ugh

[Scooter]

PC games are lagging behind consoles?? What causes you to think this?!

Console games are simple fun 5 minuters for playing on the sofa with your mates. They have neither the depth, or the eye candy of modern "PC" games. Sure they have lots of antialising, and are fairly smooth - but hell: they'd better be at TV resolution! If you saw that on your Pc's monitor you'd ask for your money back :)

Using multiple cards is a way of getting a "sneak preview" if you like at what the mainstream tech will do for you in a few months - like over clocking. Sure in 6 months time every PC/vid card will be just as quick, and yours will be the only one that requires a fridge strapped to it and it's own power plant - but hey - that's how stuff gets improved..

To get deep involving gameplay with good control, a decent multiplayer arrangement and a vast playing world you can't beat the PC gaming approach. That's why with every generation, consoles look more like PCs with all the boring unnecessary bits stripped out. Likewise, console hardware has always been built to shift bits for smooth play. PC's are heading that way with new breeds of video cards, better buses/memory architecture , CPU cache etc. This is largely caused by people starting to use their home machines for video realted applications (from editing home movies to ripping off commercial flicks, and integrating with the TV/stereo media centre style). Pretty soon they'll meet in the middle :) Or the video card will become a "console on a card" with it's own 1Gb ram that loads the entire game texture set onto the card at the start to save shovelling it down the PCI/AGP bus.

Re:ugh

[mwheeler01]

I should have been more clear. I meant in sales. They're lagging behind in sales, simply because consoles are cheaper and faster to develop for.

Re:ugh

[Scooter]

Ah - I see. Point taken

Correction

[Fiz Ocelot]

The article says:"Brian Joyce: SLI stood for Scan Line interface..."

In reality SLI stands for Scan Line Interleave.

Comparing SLI

[superpulpsicle]

Ok, using Voodoo2 term comparing SLI with dual ATI or Nvidia cards is just not fair. Things were simple then. Today the graphics drivers are far more complicated.

Murphy's law here.... too many things can go wrong SLI-ing ATI and Nvidia cards, more than any forums can handle I am sure. Christ, the PC gaming industry has already shot itself in the foot with years of driver problems.

Re:Comparing SLI

[Paladin128]

RTFA: It's not using SLI. The screen is divided in half; one card handles the top half, the other handles the bottom. Both cards have to use the same driver, and there's an additional PCI card that syncs things up. It's an expensive and relatively inefficient solution.

How I assume it's going to work: both cards keep the same geometry/texture/whatever information in RAM. They both try and render the same scene, only the software tells each to render only one half of the screenby "blue-screening" -- defining a polygon that's solid blue that covers one half of the screen. Since all modern video cards do an early Z, the polys under the blue poly will never be tessellated and filled. The pixels for the blue polygon are really easy to fill. Then, the compositor card uses the blue screens as masks.

Again, that's just a guess. My knowledge of OpenGL and other 3D programming API's is limited to my senior project. You may be able to define viewable windows and background colors or something.

Re:Comparing SLI

[Guspaz]

Isn't it also possible that they simply have the cards render at odd resolutions (ex. 1024x384) and rendering different POVs?

I must admit though, your proposed solution is elegant in it's simplicity.

Re:Comparing SLI

[Nicolai Haehnle]

You can always set a clipping rectangle for the graphics card, so the obvious solution is to mess with this rectangle.

This allows you to render at odd resolution *without* having to mess with transformation matrices.

Re:Comparing SLI

[CityZen]

I would wager it's done by changing the viewport setting on each card. The viewport controls the mapping of the viewing frustum onto the screen.

The compositor card is just a video switch. At the horizontal retrace interval where the subdivision is, it switches from one card's output to the other's. It's probably setup to count retrace intervals and do the switching itself, rather than being interrupt-driven. (I don't think horizontal retrace interrupts are supported by most video cards.)

Yes yes.

[stoneymonster]

PCI super duper eXtra Special.

Costs..

[Anonymous Coward]

Cost of extra video card to Alienware: 200$
Costs to you: +1,000$

Alienware NEEDS a new Director of Marketing!

[*BBC*PipTigger]

What the hell is up with this Brian Joyce guy?!?

Of course any "hardcore gamer" knows about the history of their "patents pending technology" as their Director of Marketing calls it. Too bad he doesn't.

In the article, this guy says: "SLI stood for Scan Line interface where each card drew every other line of the frame and my understanding was that the major challenge was to keep the image in sync. If one line's longer than another, then tearing, artifacts, and keeping the two cards in sync was a real issue. The benefits of doing it half and half is we can take advantage of the load balancing and the synchronization challenge can be overcome."

Alright... I'm sure the technology they've developed over there is some hot fscking shit. I'm sure they have a top R&D team that knows what they're doing && this custom motherboard + pre-driver thing is a good idea. Once developed fully, it could let you keep adding as many video cards as your case can hold, even potentially from different manufacturers, to improve total rendering capacity. That is great. Alienware has some very talented people to solve all the associated problems with accomplishing this. I respect their achievement.

That said, what the hell do they have a Director of Marketing for who doesn't know what he's talking about? He gets the SLI acronym wrong. How the fsck could one scan line be longer than the other resulting in tearing or cards getting out-of-sync? Come on! I know he's not a technical guy but then he should just stick to his hype buzzwords && patents && shit like that because he totally ruins Alienware's credibility when he shows no understanding of the most prominent attempt at this type of endeavor in the past. At least he said "my understanding" in there but he should've said "I don't know or understand the history so I'll just talk about what I do know."

Although I hold Alienware in high regard for making really fast gaming computers (that are arguably worth the premium price if you can't be bothered to build your own), I lose substantial respect for them when they allow their cool new technology to be represented by a marketing turd who couldn't be bothered to understand the history of what his company has done or what he's talking about. Buy a clue if you care to succeed. I want to like Alienware... I really do. TTFN.

-Pip

Re:Alienware NEEDS a new Director of Marketing!

[Anonymous Coward]

I think that's supposed to be "if one line takes longer than the other" as in rendering time and not length of the line.

Re:Alienware NEEDS a new Director of Marketing!

[*BBC*PipTigger]

I didn't read it that way but you have a much more sensible interpretation (albeit by adding a word that wasn't there). Regardless, interleaving scan lines is a lot less likely to have a load balancing problem than separating top && bottom halves of your resultant frame buffer. SLI does sacrifice the ability to load balance but interleaving is not likely to result in one card waiting on the other with any regularity so the complaint still doesn't make much sense.

-Pip

Re:Alienware NEEDS a new Director of Marketing!

[chromaphobic]

Although I hold Alienware in high regard for making really fast gaming computers (that are arguably worth the premium price if you can't be bothered to build your own), I lose substantial respect for them when they allow their cool new technology to be represented by a marketing turd who couldn't be bothered to understand the history of what his company has done or what he's talking about. Buy a clue if you care to succeed. I want to like Alienware... I really do. TTFN.

Try and actually order a system from them and you'll lose all respect, and stop wanting to like them and instead hate them.

Worst company I have ever dealt with in any market, computers or otherwise. I pray they blow all their cash on this technology (doomed to fail, IMHO) and send themselves to their grave.

Re:Alienware NEEDS a new Director of Marketing!

[Trepalium]

Actually, two identical video cards may not (completely) identical performance. Subtle variances can creep into the various timing crystals and other electronic components that can make them unsynchronized. Two video cards outputting at 60 Hz may start out in sync, but it's virtually guaranteed that over time they will drift out of sync with each other.

As an example, a primary rule of video capture is that you tie yourself to a single timing source. In other words, if you're capturing both video and audio, you don't assume that you're getting 44,100Hz from the sound card, and 29.976 frames per second from the video. If you're going to accept the framerate given to you by the video source, then you calculate the audio bitrate from that, or vice versa.

While you're capturing video output from two video cards, you can not assume the scanlines each produce will be of the same length, or that the cards will remain in sync with each other over any period of time. You can treat one source as a trusted, but the other can not be. You might be getting 60.00000Hz from one video card, and 60.00001Hz from the other. Over the course of 28 hours, the second video card would gain a whole frame over the first one. However, in half that time, the signal would've been 180 degrees out of phase. And that's only using a variance of 0.000017 percent!

Now, I'm no electrical engineer, and I might be incorrect about some of the things I've said, but I'm pretty sure that a 1MHz clock chip isn't always exactly 1,000,000 Hz. There's a couple article son timing crystals here and here. Apparently, age and tempurature can affect the oscillations of these crystals to an extent, as well.

Re:Alienware NEEDS a new Director of Marketing!

[CityZen]

The problem with SLI is that it only subdivides the rasterization work. Both cards must still process all the geometry (and lighting calculations). (Actually, the 3dfx cards which could do SLI did not have a geometry transform engine; regardless, the triangle-setup work is still all duplicated.)

Screen-space subdivision (ala Alienware) can subdivide the geometry work as well as the rasterization work. However, there will still be a lot of geometry work being duplicated, since you don't know where a polygon will end up until you've transformed it.

If the APIs supported it, you could cut down on some of this duplication (say by using bounding boxes on groups of geometry, and only sending the groups to the necessary cards based upon the bounding box tests). Even so, it's impossible to eliminate some duplication, and thus no system using screen-space subidivision will be 100% efficient.

Re:Alienware NEEDS a new Director of Marketing!

[CityZen]

That's why gen-locking or some other form of synchronization is needed. Essentially, there's just one master clock, and its signal is sent to both (or however many) cards.

Re:Alienware NEEDS a new Director of Marketing!

[Trepalium]

However, that's not possible to do with generic off-the-shelf hardware without modifying it. I understand that the Alienware technology uses off-the-shelf video cards, so there's no way to lock them to the same clock (unless they're hoping they'll all by synched with the bus clock, because I wouldn't count on it). With the old 3dfx voodoo cards, this was easy because you had a connector for the two cards. Perhaps the old VESA feature connector found on old PCI cards might've supported such features, but you don't find it on today's cards.

Now, the only way I can see this technology working would be for them to yank a copy of the framebuffer (or equivalent signal), and recompose the image before sending it out to the monitor. Either they could sit on the PCI Express bus, and try to access the framebuffer directly (SLOOOOOW), or they could sit on the analog or digital outputs, and try to grab it there. If they were sitting on the analog outputs, I have no idea how they would recompose the image. On the DVI-D output, they could either record only the data coming over the DVI link and simply replay it when it has the complete image. They'd still need to synchronize, but compared to the scanline problems the analog output would pose, it should be much easier.

Re:Alienware NEEDS a new Director of Marketing!

[CityZen]

First off, ATI and Nvidia have understood the concept of gen-locking for ages, and some of their ultra-high-end boards have provided this capability. They've been used in research and high-end visualization setups where you've got multiple projector output and gen-locking is required.

Second, Alienware is not exactly using generic off-the-shelf hardware. Can you find a PCI Express video card on any shelf? I'm sure they've worked with their vender in order to find out how to gen-lock the cards, in addition to discussing how to modify the display drivers.

Finally, did you look at the pictures that accompanied the article? If so, you'll see that there's an extra cable that goes from each video card to the combiner card. I'll wager that's the gen-lock cable.

(Actually, one of their setups appears to be running without a combiner card. It's driving two separate screens, so they may indeed have a sync problem with that setup.)

Re:Alienware NEEDS a new Director of Marketing!

[Trepalium]

I'll admit I did not look at the pictures. However, I was going by Alienware's FAQ on this technology for the information about 'off-the-shelf hardware' and such. This FAQ also claims they do not require any kind of custom driver support.

Until any of this actually makes it to market, it's all speculation. Perhaps NVIDIA and ATI are going to insist that PCI express cards have connectors for genlocking on even the lower-end gaming video cards. As it stands today, the only current NVIDIA chip that support genlocking or framelocking is the Quadro FX 3000G, and I can not determine if any ATI FireGL cards support it (I see the term referenced for the Radeon7500/8500, but only in terms of the video overlay).

I just hope they're not using something like FireGL or Quadro cards.

Interesting That Alienware

[TJ_Phazerhacki]

Why does it seem that Alienware is so far out on the bleeding edge of technology?

Oh, yeah, right. They are.

I mean, come on, with the kinda influence they have - they asked ATI and nVida for custom cards for the Area51m - is it any real suprise they are attempting to make themselves even better?

I suppose that the fact there are a number of other producers in this Niche - See the earlier slashdot story - might encourage the development. But the simple fact remains - They are on top, and if they can lock down this intelectual property till 2nd gen, then they can release it publically and become innovators in more than just overclocking and cool case mods.

MMMmmmm...Cool case mods.

It kinda does apply here...

[wickedj]

Imagine a beowulf cluster of these... *ducks*

Re:It kinda does apply here...

[Quarters]

NVidia already has.

Re:It kinda does apply here...

[CityZen]

Or, for a real-time version:

http://www.cs.princeton.edu/~rudro/sketch00.pdf

Hardware's old news, but availability is not

[FueledByRamen]

This kind of system has been around for quite some time, both the software and hardware to do it. SGI's Onyx4 uses their OpenGL Multipipe software kit (which works on unmodified OpenGL apps), a bunch of ATI FireGL cards, and some digital video compositing hardware to do both loadbalancing for complex data sets (what Alienware is doing) and realtime rendering at resolutions much higher than any one card could support (tiling).

The thing that's new about this implementation is that you won't have to run out and drop $40,000 on the base Onyx4 if you have an application that needs this (to some extent - SGI's solution will go to 16 cards, with the bandwidth to drive them all, while Alienware's is currently limited to 2). Only $4000 for the Alienware box.

Somehow I doubt that Alienware will get the patents that are 'pending' - I'd imagine that SGI probably already has a whole portfolio covering this area, since this kind of thing is their bread and butter. It's nice, though, to see a consumer-affordable implementation of this technology coming to market.

Re:Hardware's old news, but availability is not

[Paladin128]

Somehow I doubt that Alienware will get the patents that are 'pending' - I'd imagine that SGI probably already has a whole portfolio covering this area, since this kind of thing is their bread and butter.

Yes, because nobody is granted patents when there is a lot of prior art.

Re:Hardware's old news, but availability is not

[FueledByRamen]

Well, prior art aside, what I meant is that they won't get their patents because SGI has most likely already patented the technologies behind it. Of course, if they tack "on the internet" to the end, they could probably get all of their applications immediately granted...

Make your own video card?

[jameskojiro]

So does this mean if I buy 3 really cheap ATI radeon 9200's i can have the same performance as a Radeon 27600 ??

Or can I finally put to use those old ISA videos cards i used to have in my 386?

Load balancing? Not in their demo.

[Guspaz]

Anybody see the demo videos of this? If you did, you'd notice that when they're busy unplugging alternating video cables to show that only the top or bottom half of the screen is rendered, the size of the image never changes.

In other words, in their examples, which used quake 3, there was NO load balancing going on. If there was, when we saw, for example, the top half of the screen, the size of the top half should have been constantly changing.

I understand fully that we were seeing alpha or beta level stuff here, but perhaps they should have waited until they had a fully functional model before showing it off.

Re:Load balancing? Not in their demo.

[Drathos]

Um.. That's because the second video card is still in there rendering away. Most systems don't even pay any attention to whether or not the monitor is plugged in - they still keep rendering the desktop/console.

I've personally only seen one computer that cared if a monitor was plugged in, and that was all custom hardware.

Re:Load balancing? Not in their demo.

[Guspaz]

Yes. So, and your statement agrees with this, what we should see is the card that IS plugged in changing it's rendering size. We should have seen the visible part growing and shrinking. Which would indicate that that videocard was being given larger and smaller parts to render based on the load balancing.

Instead we saw a fixed size, which indicates the card was always rendering the same size, meaning NO load balancing was being done.

Re:Load balancing? Not in their demo.

[Anonymous Coward]

No, what you saw was them unplugging a monitor downstream of the array system which showed you that each one was drawing 1/2 the screen. They actually hooked each card up to a different monitor so you could see what was going on (they've got a picture in the article)

Re:Load balancing? Not in their demo.

[cgenman]

Instead we saw a fixed size, which indicates the card was always rendering the same size, meaning NO load balancing was being done.

Or that the amount of data being fed to the two cards to crunch was staying roughly the same for the two seconds of available grainy video from Tech TV. Geeze, for a multi-thousand dollar system which requires an 800 watt power supply, and two top-of-the-line graphics cards for a %50 percent increase in performance, the best complaint you can come up with is that the preview demo isn't obviously load balancing?

What about the futility of paying an extra thousand bucks for an upgrade that will need to be re-purchased every 6 months?

Re:Load balancing? Not in their demo.

[Guspaz]

I'm not complaining about it, I think it's an interesting new technology with a lot of promise. I'm simply making an observation about the presentation.

I didn't watch it on TechTV either, as I don't get TechTV. I watched it on the net, and the quality wasn't that poor, and there was significantly more than 2 seconds available.

Who said anything about upgrades every 6 months? Up until now (with the release of the Radeon X800 and GeForce 6800), there hasn't been a single videocard that has dramatically improved on the Radeon 9700 Pro, a card that is what, 18, 20 months old now? It took nearly two YEARS before a significant improvement was made. In short, had this technology been availale prior to the 9700's release, somebody with a system with dual 9700s would still be in business today, indeed they'd have no large reason to upgrade to a dual-X800 system. Their two year old dual-9700 Pros would still keep up with today's fastest. Who knows how long the X800 will last as a decent card.

Not quite practical

[S3D]

I did some devlopment for similar sytem - 4 videocard working in parallel, tiling the screen or time-dividing frames. To put it short - it's very difficalt to extract performance gain, require a lot of geometrical culling or synchronizations and other triks. Off the shelf game will not give 50% performance gain with such a system, 15% in the best case (and i doubt even that, and it would quite probably create artefacts) . To extract something similar to 50% would require a lot off efforts for developers, no develpers would want to do it to support a tiny market share.

Re:Not quite practical

[CityZen]

It really depends upon the amount of rasterization work compared to the amount of geometry work being done. If you take a game that uses lots of long pixel shaders and bump up the resolution, I'm sure you won't have that much trouble achieving a good speedup. On the other hand, if you have highly-detailed geometry, simple pixel shaders, and not-so-high resolution, this probably won't help you much (assuming that you don't have efficient culling in your APIs).

Also, a 4-way screen-subdivision system will typically be half as efficient (regarding the geometry work) as a 2-way system. It has twice the border length, and therefore twice the number of primitives that straddle the borders (on average, depending on the application).

Wow, another way to charge us.

[qoa]

Now instead of one 400 dollar video card, you can get two! I'm not to eager for this to become common place.

2 16x PCIe Slots....

[NoMercy]

Am I the only person who couln't give a damn about SLI and would rather have two dual-head cards in the system to power 4 flatpanels all with scrolling ccze'd logs so I can sit back in my huge leather chair and cackle with power while stroking my white cat?

Alienware on the downside

[Impy the Impiuos Imp]

Having just bought the top end 3.2 GHz Alienware in November, with 64 bit chips and this new design, I feel like I just bought a 486 right before the Pentium came out.

Tax Deductible

[catherder_finleyd]

If you are a US Computer Consultant, you can argue that your Alienware machine is what you need to develop and demo software for your clients. That can save you a fair amount come income tax time.