Running Video Cards in Parallel

G.A. Wells writes "Ars Technica has the scoop on a new, Alienware-developed graphics subsystem called Video Array that will let users run two PCI-Express graphics cards in parallel on special motherboards. The motherboard component was apparently developed in cooperation with Intel. Now if I could only win the lottery."

Comments

In other news...

[Unloaded]

...Microsoft announced that Clippy had broken the before unheard of 2,000 fps barrier.

Re:In other news...

[david.given]

Microsoft announced that Clippy had broken the before unheard of 2,000 fps barrier.

However, they went on to say that Clippy was still intact. They're going to try again using a bigger catapult and with a concrete-reinforced barrier.

Man am I out of the loop.

[Randolpho]

PCI-Express? What happened to AGP?

Seriously, I've been out of the PC market for too long. Alas, poor wallet. I had cash flow, Horatio.

Re:Man am I out of the loop.

[Laebshade]

PCI-Express is meant to replace AGP. From what little I've read into it, it will require lower voltages than AGP and has a wider bus.

Re:Man am I out of the loop.

[Randolpho]

After I posted that, I did a quick google:

http://www.pcworld.com/news/article/0,aid,94724,00 .asp

Re:Man am I out of the loop.

[Plutor]

You've been out of the PC market for about a decade then, if you've never heard of PCI-Express. It's been proposed and talked about and raved about for years, but it's just now finally coming to market. The best thing is that it's not limited to a single slot per board! That's why this parallel thing is even possible.

Re:Man am I out of the loop.

[Auntie Virus]

There's a White Paper on PCI Express from Dell: Here

Re:Man am I out of the loop.

[The_K4]

Since the PCI-Express spec defines switches (these are like P2P-bridges only they have 2 sub-buses) a mother board manufacturer could add 2 of 3 of these and get 4 PCI-Express Graphics ports (or 7 and get 8 ports) the problem is that every time you do this you have to share the total bandwidth at the highest level. Since PCI-Express does have more bandwidth the AGP 8x and 1/2 of that bandwidth is dedicated up the other 1/2 dedicated down. So the down-stream (where video cards use most of their bandwidth) is greater the AGP8x's TOTAL bandwidth. So this data path bottle next shouldn't be bad if you have 2 cards (might work well for 4 if they use the bus right).

Re:Man am I out of the loop.

[Geoff-with-a-G]

You've been out of the PC market for about a decade then, if you've never heard of PCI-Express. It's been proposed and talked about and raved about for years, but it's just now finally coming to market.

He must have been out of the market for a decade, to have never heard of something which is only just now in the market? What?

I've been half way out of the market for about five years, and I only recently heard of PCI-Express, and I didn't have many details about it. Researching new, not yet marketed technologies isn't the same as "being in the market".

Re:Man am I out of the loop.

[WuphonsReach]

You've been out of the PC market for about a decade then, if you've never heard of PCI-Express.

That's rather over-stating the case.

Roughly 10 years ago, PCI was finally just supplanting EISA/VESA and ISA boards were still common.

I build a few machines per year, and PCI-Express only just hit my radar screen in the past 12-18 months. Even today, I have yet to see mainstream motherboards or cards for it, so it's still rather ephemeral at this point.

It is an interesting design. Whether or not it will live up to it's promise remains to be seen.

Quad-screen?

[Vrallis]

Hell, I couldn't care less about parallel processing for the video cards.

I want tri-head or quad-head video, but with at least AGP speeds. You can do it now, but only with PCI cards getting involved.

Re:Quad-screen?

[houghi]

I want tri-head or quad-head video, but with at least AGP speeds

So order one now. They are available here at Matrox.

Re:Quad-screen?

[gr8_phk]

I'll second that. Flight simulation begs for 3 screens, as do some driving and other games.

On another note, I suspect the only way it will really accelerate single images is in cases where render-to-texture is used. i.e. per-frame generation of shadow or environment maps. The completed maps could then be passed to the card that actually has the active frame buffer to be used in regular rendering. Two cards could at BEST double performance and nothing ever scales optimally.

Re:Quad-screen?

[RhettLivingston]

As long as you can live with at least one of the screens not running quite as fast (maybe an informational type of screen as opposed to 3D scenery?), 3 screens ia really easy today. Almost all decent AGP cards these days support 2 screens at 1600x1200. Throw in a good PCI card and you've got 3. I've been running this way for years and it works well. Actually, the PCI card isn't shabby.

The only problem I encounter in Windows is an occasional tooltip coming up on the primary monitor instead of a secondary monitor. This is not the fault of the OS, rather the application is constraining the tooltip to be on the primary monitor by forcing it to be within the primary monitor's coordinates.

Note that Matrox's single board AGP solution does not compete with this. Using a high end NVidia for the main two screens provides too much of a performance advantage to give up for Matrox's slow cards. Matrox's cards, even though on AGP, run about like the PCI cards.

Regardless, when these systems become more available, I will be one of the first to put 2 video cards in and run 3 or 4 screens from my PCI Express system. But, though I like playing 3D games this way, I do it for the extra informational surface for programming. It greatly eases things to run your application on one screen and your development environment on all of the others so that you can see everything at once. And with 19" 1920x1440 monitors (which usually manage 1600x1200 with better focus than a 1600x1200) running around $250 a pop, its a very worthwhile investment.

Re:Quad-screen?

[Lumpy]

hows it feel to want? espicalyl about something that has been available for a decade now.

here

Matrox, because all the other cards are merely toys for the kids at home.

Press Release

[Anonymous Coward]

over here: clicky

Voodoo

[Eu4ria]

Didnt the early voodoo cards allow something similar to this ? I know they had a pass through from your 'normal' video card but i seem to remember the ability of running more and they would each do alternating scan lines.

Re:Voodoo

[scum-e-bag]

The company was 3dFx, and it was thier Voodoo II cards that allowed the use of two cards a few years back, sometime around 1998 IIRC.

Re:Voodoo

[Trigun]

Yes, they did. Unfortunately, at the time they were too expensive and took up all of your extra slots on your mobo. Now, with integrated everything, it's not so bad.

Good idea implemented too early. Such is life.

Re:Voodoo

[naoiseo]

without a special motherboard, yes.

I think you could string something like 4 voodoo rush cards together or something (who knows if you got 4x performance, but I'm sure it went up not down)

Problem was, by the time they put this out there, the tech it was running was months behind cutting edge. 4x something old is easily forgotten.

Re:Voodoo

[UnderScan]

SLI - scan line interleve, was available for 3dfx Voodoo IIs (maybe even Voodoo 1) where the first card would process all the odd lines & the second card would process all the even lines.

Re:Voodoo

[kamelkev]

Voodoo was basically the beginning of the performance PC market, with tons of wierd options and card types.

Benchmarks for the old 3dfx V2 SLI can be seen here:

http://www4.tomshardware.com/graphic/19980204/

I was (and still am, although its in the junk pile) a 3dfx V2 owner, the performance of that card was just amazing at the time. The Voodoo and the Voodoo2 definitely changed the world of 3d gaming.

Also of interest is an API that came out much later for the 3dfx chipsets that actually let you use your 3dfx chipset (they didn't call it a GPU back in the day) as another system processor. If you were an efficient coder you could actually offload geometric and linear calculations to the card for things other than rendering. I can't seem to find the link for that though, it may be gone forever.

Light on Info

[the morgawr]

The PR mess is light on information and I don't have flash to view their site. Can someone give some technical information? e.g. How does this work? What does it really do? What can a typical gamer actually expect (surely it doesn't just double your power by sending every other frame to each card)?

Re:Light on Info

[Caesar]

http://arstechnica.com/news/posts/1084398037.html

It splits the screen in half. Alienware claims a ~50% boost.

Re:what will this do?

[m4vrick]

Now you can play solitary and minesweaper at the same time on a maximized screen. :)

interesting technology

[cheese_wallet]

I think it is great that a company has the will to do something like this, even if it doesn't catch on. It's cool to try something new, instead of just hanging back and doing the tried and true.

I'll admit I haven't yet read the whole article, but even though it says that it isn't tied to any one video card, that doesn't say to me that it can have multiple disparate cards. If it is doing something along the lines of SLI, I would guess that the speeds would need to be matched between the two cards. And that would imply having two of the same card, whatever card the user chooses.

But maybe not... maybe it's the advent of asymetric multi video processing.

Re:interesting technology

[jonsmirl]

You can do this today with Chromium.

Chromium replaces your OpenGL library with one that farms the OpenGL drawing out to multiple machines. It's how display walls are built.

You can use the same technique for multiple card in the same box.

this isn't new

[f13nd]

Alienware didn't invent this
the PCI and PCI Express have had this written into spec
AGP does too, but when was the last time you saw dual AGP slots on a mobo? (they do exist)

Re:this isn't new

[BenBenBen]

The AGP port spec lays it out; AGP is a preferred slot on the PCI bus, with four main enhancements (pipeline depth etc) designed to... Accelerate Graphics. Therefore, if you had more than one PCI bus, you could technically have more than one AGP port. However, I cannot find a single motherboard that offers 2 AGP slots, including looking in numerous AV/editing specialists, where I'd expect this osrt of thing to tip up.

Big Deal - PCI Express. Any one can add two video

[liquidzero4]

So what technology did Alienware create here? None..

So they have one of the first MB's with two PCI Express slots. Big deal, soon MB's will contain many PCI-Express slots. Hopefully a lot more than 2.

Oh, come on!

[Short Circuit]

All you really need is some way to copy the data in memory from one card to another.

Easy solution? Several high-speed serial connections in parallel between the two cards. With a little bit of circuitry on the card dedicated to keeping the data identical.

Or, with a little bit of a performance hit, you could keep each section of RAM separate, and route misses over the cables.

Nice, A complete Vapor-article.

[Gr8Apes]

From the article: "The answers may have to wait until Q3/Q4". There are no performance numbers, no real statements of how it works, nothing much at all. Just wow, gee whiz, dual graphics cards in parallel. What exactly does "in parallel" mean? That's not even addressed.

Some things I thought of immediately reading this, great - two displays each driven by a separate card, or, better yet, quad displays driven by two cards. Nope, not a word about either possibility. The implication of the PR/article is that 3D graphics will be processed faster. How? Do they have some nifty way of combining two standard off the shelf graphics card signals into a single monitor? (Hint, it's hard enough getting the monitor to properly synch up with a single high performance graphics card!)

Since when does ArsTechnica merely regurgitate PRs? This was 99.999% vacuum.

Re:Nice, A complete Vapor-article.

[Caesar]

E3 is on this week. E3 is dominated by product announcements, including products that won't see the light of day for years. So, the "vapor" aspect of it is what pretty much what happens at most of the trade shows. By that measure, half of the news out this week is "vapor." Sure, some people might blow smoke and tell you about the performance stats, but we all know that when the products actually ship, their "performance stats" will probably have changed as well.

So, you have the press release to go on. And as you noted, I threw up a flag of caution. I didn't call "vapor" on it, however, because as I noted, they are saying Q3 and Q4 for VA and X2, which seems awfully soon to be pimping vapor. I also happen to know that this is in fact happening, so I didn't feel obliged to cast aspersions on it.

From the PR, it's rather clear to me that this is a single-monitor solution. It's also likely the case that this is not just a PCI-Express matter. Call me old fashioned, but Alienware says that they have patent-pending technology, and while I would usually take the language for marketing drivel, I have no reason not to trust Alienware on this matter. In fact, they have tried this before, but the underlying technology just wasn't doable.

Are we going to need this...

[CodeMonkey4Hire]

for Longhorn?

Intel's Chipset only supports One x16 PCIe

[hattig]

In fact all the first generation PCI-Express chipsets only support one x16 PCIe for graphics controller.

I doubt that Intel is going to make a 2 port one especially for Alienware.

So I expect it means that the second graphics card is plugged into a x4 or x1 PCIe connector.

Anyway, this is nothing special, it is all part of the specification. Hell, you could have two AGP v3 slots in a machine working at the same time - how do you think ATI's integrated graphics can work at the same time as an inserted AGP card's?

Re:Intel's Chipset only supports One x16 PCIe

[Short Circuit]

Anyone with that hardware could tell us.

Under Linux, run "lspci" as root, and see if the two cards are on different PCI buses.

You can do something similar under Windows XP:

Go to the device manager, and look at the Location field of your two video devices. The box I'm on only has one, but here's what an AGP card's location field looks like: "PCI bus 1, device 0, function 0"

Re:Is that new?

[lucas teh geek]

I belive you need to RTFA, this isnt about dual head setups. one head, two cards

Metabyte PGC

[Erwos]

It looks like the same thing as Metabyte PGC - and Alienware was supposed to be the roll-out partner for that.

Nothing wrong with it, though - PGC actually did work, and was previewed independently by several people (I think Sharky?).

-Erwos

Re:Next comes dual AGP graphics.

[DaHat]

Nay, the AGP standard is built around a single slot and a single graphics card. To permit two AGP cards running natively (via the AGP bus) in a single system would be quite difficult if not impossible, far easier to look to the future and a new technology to make it work better then any sort of hack job that could be done today.

Re:I don't think the author really got it there...

[Enrique1218]

Maybe they don't have to cooperate. Graphics card gennerally support the same standard (vga/directx/opengl). Perhaps, the video array will have its own driver/software component to receive the game data then parcel the data to each card.

But do you need multiple monitors?

[MtViewGuy]

I think the big question we need to ask is do we really need multiple monitor setups?

Besides the obvious issue of hardware cost of multiple graphics cards and multiple monitors, you also have to consider desktop space issues. Even with today's flat-panel LCD's, two monitors will hog a lot of desktop space, something that might not be desirable in many cases.

I think there is a far better case for a single widescreen display instead of multiple displays. Besides having a lot less impact on hogging desktop space widescreen displays allow you to see videos in the original aspect ratio more clearly and also allow for things like seeing more of a spreadsheet, clearer preview of work you do with a desktop publishing program and (in the case of a pivotable display) make the reading of web pages easier and/or single page work with a DTP program easier. Is it small wonder why people so much liked the Apple Cinema Display that uses a 1.85 to 1 (approximately) aspect ratio?

Re:But do you need multiple monitors?

[Jeff DeMaagd]

Two 4:3 displays can be bought at a lower cost than one widescreen display.

Re:Everything old is new again?

[skiflyer]

I'm sure this will be said a million times during this thread, but this has nothing to do with multiple heads... it has to do with multiple cards serving 1 head.

The real question

[241comp]

Is this compatible with Brook and other general-purpose GPU programming techniques? The use I see for it is this:

Imagine an openmosix cluster of dual-processor machines that run bioinformatic calculations and simulations. Lots of matrix math and such - pretty fast (and definitely a lot faster than a single researcher's machine).

Now imagine the same cluster but each machine has 2 or 4 dual-head graphics cards and each algorithm that can be created in Brook or similar is. That gives each machine up to 2 CPU's and maybe 8 GPU's that may be used for processing. The machines are clustered so a group of ~12 commodity machines (1 rack) could have 24 CPU's and 96 GPU's. Now that would be some serious computing power - and relatively cheap too (since 1-generation old dual-head cards are ~$100-$150).

By the way, does anyone know if there is any work going on to create toolkits for Octave and/or MatLab which would utilize the processing power of a GPU for matrix math or other common calculations?

The Return of Voodoo 2 SLI

[MoZ-RedShirt]

History repeating: Who can (or can't) remember

Power to them if they can pull it off!

[l33t-gu3lph1t3]

discrete parallel graphics processing has been around for a while. The most notable example of it is probably 3DFX and their Voodoo-2 cards. However, there's a problem with this tactic, namely, in the "diminishing gains" department.

So here's the question:

-How is pixel processing going to work? For a given frame, there is vertex, texture information, as well as the interesting little shader routines that work their magic on these pixels. How are you going to split up this workload between the 2 GPUs? you can't split a frame up between the GPUs, that would break all texture operations and there would be considerable overhead with the GPUs swapping data over the PCI bus. *MAYBE* having each gpu handle a frame in sequence would do the trick, but, again, it's a dicey issue.

It would appear to me that this dual-card graphics rendering is quite similiar to dual-gpu graphics cards. Except, where in a graphics card you can handle cache/memory coherency and logic arbiting easily due to the proximity of the GPUs, with this discrete solution you run the problem of having to use the PCI Express bus, which, as nice as it is, is certainly not that much faster than AGP.

So I say, power to you Alienware. If you can pull it off with Nvidia, ATi et all, great. It's too bad the cynical side of me thinks this idea reeks of those blue crystals marketing departments love :)

Re:How is this different than what I've been doing

[IAmAMacOSXAddict]

If you read the short article. they are talking about pushing all the data for a SINGLE monitor, through two PARALELL cards. Essentially allowing for twice the GPU power to crunch the graphics for the monitor.

You are running a bunch of video cards INDEPENDANT of each other. Clearly NOT THE SAME THING...

Re:Next comes dual AGP graphics.

[Jeremy Erwin]

It's in the AGP 3.0 spec.

AGP3.0 allows a core-logic implementation to provide multiple AGP3.0 Ports. Each AGP3.0 Port is a bridge device with multiple AGP3.0 devices hanging off the secondary bus. Each Port has a separate Graphics AGP aperture and GART that is independent and not shared with another AGP3.0 Port; however, these are shared across the devices within a single AGP3.0 Port.

too many standards

[jcostantino]

I'm going back to ISA...

But will they run?

[strredwolf]

Okay, we have a GeForce and a Radeon in parallel. What's the communication protocol that's running over PCI Express that allows them to do that?

Something tells me you need special drivers AND/OR a standardized graphic card accellerator protocol just to pull it off, otherwize you're stuck with two of the same cards.

Not really hard....

[Kjella]

How? Do they have some nifty way of combining two standard off the shelf graphics card signals into a single monitor? (Hint, it's hard enough getting the monitor to properly synch up with a single high performance graphics card!)

Duplicate data stream (should be doable in hardware), have them render half each (every 2nd scanline?) and merge them with a trivial buffer (keep two bools, one "firsthalf=done/not done, secondhalf=done/not done"). You'd limit yourself to the minimum of the two, but since they each paint nearly the same (one scanline off) the performance should be near a 100% doubling.

Basicly, it's going to be an expensive design involving some already expensive cards. But it's definately doable.

Kjella

Who else will obviously support this on mobos...

[Geiger581]

Let us not forget that nVidia and ATI both produce chipsets. Multiple graphics card purchases per system would be a dream for them, and they can help in a direct manner. Although there are not many (read as 'maybe a dozen worldwide') boards with dual AGP, the PCI-Express standard will lead to much easier multi-GPU setups. Also, the newest ATI chipsets with embedded GPUs support multi-monitor if an ATI card is used in the empty AGP slot, so you know that these guys already have to have agendas for PEG in mind.

ATI will be doing this it's Catalyst drivers soon

[89cents]

Ok, so they havn't explicitly said so, but have been hinting at it

Ati's Terry Makedon says: "Something big is coming for CATALYST in the next 2-3 months. It will take graphic drivers to a brand new level, and of course will be another ATI first. It will be interesting to see how long before other companies will copy the concept after we launch it."

Hmmm... just in time for PCI Express and it's not something specifc to Ati's hardware.

Re:Hello? Matrox, anyone?

[Niddix]

You've missed the point. Multi head is fine and has been around for ages.. 8 years ago I used to use a EGA monitor attached to a card for debugging. This is using the GPU of multiple cards to crunch the numbers for a single display. Very different.

Hardly a new concept

[default luser]

And no, I'm not referring to SLI, which was specifically designed to pair two Voodoo 2s together. I'm talking about technology that can bridge any two cards together. This is nothing more than the complex bridging involved in say Metabyte's TNT 'SLI' solution that consisted of a PCI bridge and software to split the framebuffers. It was never released for two reasons

1. GeForce 256 released shortly after this was announced.
2. PCI bridge required both the AGP and the PCI card to operate in PCI DMA mode. Unfortunately, there never was such a thing as an "AGP bridge".

In any case, other companies have now successfully implemented a simple framebuffer splitting concept on-card, where the bandwidth is more plentiful. The ATI Rage Fury MAXX and the 3dfx VSA-100 come to mind, these chips simply split the framebuffer rendering according to complexity. Beyond that, NOTHING was shared - triangle and texture data were replicated for each chip.

The key to this: on the software side in 3D mode their software automatically splits two framebuffers between the two cards. As for the "special" chipset, whatever scene data is sent to one video card, the same data is sent to the other video card. I can't imagine it being any more complex than this.

Re:3dfx Still Alive! Drivers Still Being Written

[Cutting_Crew]

i am a part of 2 teams that are still writing drivers for 3dfx so please dont give up on them just yet. the newest drivers support up to 8.0 and beta for 9.0. they also allows emulated T&L for playing some of the latest games at a decent FPS and some nice DVD movies with all the special effects. for instance, on my 3dfx Voodoo 3 i can play warcraft 3, UT 2003 and UT 2004, Need FOr speed HP2 and Underground(and of course the older NFS titles) and many more games! you can go to http://www.3dfx.com and click on the voodoo files link to get the updates and/or drivers for the current 3dfx card that you have or would like to buy. Unfortunately i think all the new drivers are compatible with WIN9x, 2k and XP..sorry linux 3dfx fans. anyway the cards are still alive so if you would like to help with GLide drivers or openGL drivers for the cards then please snoop around and ask how you can help. :) oh yes and to respond to one person, a lot of 3dfx fans did stay with NVidia b/c most of the 3dfx engineers went to NV. so youve got a lot of die-hard NV/ex-3dfx fans out there... including me :)