Dual Video Cards Return

Kez writes "I'm sure many Slashdot readers fondly remember the era of 3dfx. SLI'd Voodoo 2's were a force to reckoned with. Sadly, that era ended a long time ago (although somebody has managed to get Doom III to play on a pair of Voodoo 2's.) However, Nvidia have revived SLI with their GeForce 6600 and 6800 cards. SLI works differently this time around, but the basic concept of using two cards to get the rendering work done is the same. Hexus.net has taken a look at how the new SLI works, how to set it up (and how not to,) along with benchmarks using both of the rendering modes available in the new SLI." And reader Oh'Boy writes "VIA on its latest press tour stopped by and visited in the UK and TrustedReviews have some new information on VIA's latest chipsets for AMD Athlon 64, the K8T890 and the K8T890 Pro which supports DualGFX. But what has emerged is that DualGFX after all doesn't support SLI, at least not for the time being, since it seems like nVidia some how has managed to lock out other manufacturers chipsets from working properly with SLI. VIA did on the other hand have two ATI cards up and running, although not in SLI mode."

Comments

New trend ?

[FiReaNGeL]

Dual video cards... soon dual-core CPUs, is it a sign that we're slowly approaching the Moores Law limit? The 'dual' strategy allow for further performance gains.... but I can't see myself using more than 2 video cards (hell, I can't even see myself using more than 1), so that will be a very temporary solution.

And we're not even speaking of how much power (wattage) these 'dual solutions' consume...

Re:New trend ?

[Scoria]

The 'dual' strategy allow for further performance gains....

Eventually, barring any further technological advances, perhaps we'll even result to modular clustering. Once again, the enthusiast's computer will be larger than a refrigerator! ;-)

Re:New trend ?

[krymsin01]

Once again? heh.. yeah...

Re:New trend ?

[LWATCDR]

Duel video cards are not the same as duel-core CPUs. It is more like dare I say it a beowolf cluster. When will we see a true duel gpu card or maybe a duel core gpu?
What I am more interested in is will we see smp boards supporting duel-core AMD-64 cpus. It could be interesting since from what I read the AMD64 is NUMA when using more than one cpu but I would guess that the duel core would be more of an UMA.

Re:New trend ?

[Anonymous Coward]

Multiple GPUs will be good to have as there are lots of uses for GPUs additionally to pretty pictures.

The Folding@home (http://folding.stanford.edu/) is about to enter the GPU based Folding:
http://forum.folding-community.org/viewtopic.php?p =75287#75287

Interesting times ahead...

Re:New trend ?

[GreatDrok]

Dual video cards... soon dual-core CPUs, is it a sign that we're slowly approaching the Moores Law limit? The 'dual' strategy allow for further performance gains

I don't think so. Quoting from Intel's web site: "Moore observed an exponential growth in the number of transistors per integrated circuit and predicted that this trend would continue." Many people assume Moores Law states that speed of processors will double every 18 months and that the fact that it is becoming difficult to increase clock speed now means that Moores Law is finished. However, increasing speed is a consequence of higher clock speeds and higher transistor counts. Dual cores means you can increase the number of transistors per IC more and actually use them to do real work rather than simply adding a huge cache (as was done with the latest Itanic). End result, more speed, higher transistor count, and Moores Law still fine. In fact, dual cores could mean that the transistor count increases at greater than Moores Law in the near term. Of course some might question whether a siamesed pair of processors actually constitutes a single IC.....

Re:New trend ?

[Neward Rylet]

what about Dual Dual video cards?

Re:New trend ?

[LabRat007]

My personal opinion is that this really just boils down to marketing. They want a new product for the high end consumers to buy and the dual video card thing hasn't been used in a while. Given time we will see it fade and return again.

Re:New trend ?

[Masami Eiri]

Voodoo 4 and 5. 4 had 2, I believe, and 5 had 4 (!). The 5 needed an external power brick. They still were both outgunned (at least in bang-for-the-buck) by the GeForce.

Re:New trend ?

[EpsCylonB]

Duel video cards are not the same as duel-core CPUs. It is more like dare I say it a beowolf cluster. When will we see a true duel gpu card or maybe a duel core gpu?

I'm no expert but I was under the impression that the bottleneck in current graphics cards is the amount of memory and the speed of the bus that the texture data has to travel down. Aparrently the actual 3d geometry is very easy to process, its the rendering and its associated problems that slow things down.

Re:New trend ?

[cannon fodder 0109]

The XGI Volare has two GPUs on one card already. It's been out for some time. It's performance is nothing special. Try this http://www.xbitlabs.com/articles/video/display/xgi -volari.html for a review.

Re:New trend ?

[Surt]

Dual video cards is just a poor engineering solution to how to render using two GPUs. Soon enough they'll be making multi-core and multi-gpu cards (there is some motivation to go to multi-gpu before multi-core to avoid manufacturing losses).

Re:New trend ?

[Surt]

You will be able to buy multi-cpu multi-core setups pretty much immediately after the multi-core cpus are available.

Re:New trend ?

[Nahor]

The new Moore's law: the number of CPU double every 18 month.

Until now, nobody never really had a "full" CPU, so the double was always less than two.
But today, CPUs are more efficient that we can say that most transistors are used, so now we have more than 0.5 CPU in our machine. You double that, so you get a bit more than 1 CPU. And since you need an integer number of CPU, it must be rounded to 2.

And now, if you'll excuse me, I'll return to coding my revolutionary tool for developping new CPUs.

Re:New trend ?

[Carnildo]

Duel video cards are not the same as duel-core CPUs. It is more like dare I say it a beowolf cluster. When will we see a true duel gpu card or maybe a duel core gpu?

We've seen dual-core GPUs already. What do you think a multi-pipeline GPU is?

Re:New trend ?

[sethstorm]

It didnt stop government contractors from using 3DFX before they were cursed by Nvidia.

Re:New trend ?

[cplusplus]

And we're not even speaking of how much power (wattage) these 'dual solutions' consume...

A long time ago I had an Obsidian X-24 graphics card, which was basically an SLI Voodoo2 on one card that drew its power from a single PCI slot. It used so much power that that my computer would just power off without warning quite frequently.
A 350 watt power supply fixed the problem (I had a 250 watt), and that was a LOT of power back then. Now I have a 400 watt Antec power supply which was the recommended solution for my AMD. I think in a year or so that 500 watts will be commonplace.

Re:New trend ?

[csimpkin]

What about the dual gpu card from ati. I had one. It was the rage fury maxx. It had dual rage 128 chips. It was only ever fully supported under win9x. It couldn't do better that 640x480x256 in win2k and xp. Linux only supports the first gpu on the card and only half of the memory, so the second gpu is just decoration. At a time when a good card had 16megs of memory this thing had 64 megs and was AGP.

Re:New trend ?

[MarkGriz]

"Duel video cards"

Isn't that what Zell Miller runs on his PC?

Re:New trend ?

[king-manic]

moores laws truly only regaurds transitors density. CPU speed or clock speed does not follow moores law.

Re:New trend ?

[oliverthered]

Well, it a pair of adders can be on a single IC then yes, so long as the bus between the CPU's is build into the IC.

Re:New trend ?

[adler187]

No the Voodoo 6 6000 needed the external power brick because it had 4 chips. The 5500 had only 2 and the 4500 had 1.

From wikipedia

My memory differs from the wikipedia, I seem to remember there being a Voodoo 4 4000 and I believe the 3dfx site listed it as the Voodoo 6 6000 and not a Voodoo 5 6000. Although most of my information was looking around on the 3dfx site back in the day so it may be they listed cards that weren't actually released (like the 5000 which I remember seeing there too). you can find pictures of it here

Also according to the wikipedia entry, the Voodoo 6 6000 would have beat the GeForce 2 GTS but not the Ultra or the GeForce 3.

Re:New trend ?

[owlstead]

Of course some might question whether a siamesed pair of processors actually constitutes a single IC.....

As long as the software developers do (with their crazy per-processor schemes it doesn't matter. Microsoft got that right in one go (I still don't like them, but they seem to do more right lately). Others will probably follow suite, at least for the PC/small server market.

And the rest is academic. Call it what you like, as long as it speeds up my PC and gives me better response time? Since the processors do share resources on the IC - and it's almost impossible to not share some resources - lets call it one IC (and anyway, IC means integrated circuit, does say nothing on what the circuit constitutes).

Re:New trend ?

[OblongPlatypus]

My computer is already larger than a refrigerator, thank you very much :)

Re:New trend ?

[Anonymous Coward]

I fully expeceted a "3 |\| |-| /-\ |\| [ 3 y0R P3N1s n0w!" joke - about higher performance, clustering, dual-members and what not....

Re:New trend ?

[timeOday]

Of course some might question whether a siamesed pair of processors actually constitutes a single IC.....

That's the rub right there. Moore's law under its broad interpretation - "computers get exponentially faster" - was great because new processors could run the same old programs with exponentially increasing speed. (Moore's law under its narrow interpretation - transistor count - is quite useless, since nobody cares about transistor counts per se).

N parallel processors are never as good as a single processor N times faster. Even if you're lucky and your algorithm is perfectly parallelizable (which it never is), the parallel implementation is more complex - meaning higher cost and more bugs for the same algorithm.

That said, I guess it's job security. It will take a whole different mindset to program for 1024 processors. We might even start dumping efficient algorithms for alternates that take more cycles overall but are more parallelizable.

Re:New trend ?

[kcb93x]

Just was at AMD's 64bit technote presentation (or whatever they called it) They've got dual-core CPUs in the works right now...I want to say it was like Q2 or Q3 of 2005. I might be wrong, I can't quite remember. At that point I was getting a little ansy to go look at all the cool displays.

Barring BIOS support, you'll be able to drop the dual-core CPUs into existing boards (assuming the board itself supports it - which from the sounds of things, some will just require a BIOS update)

Re:New trend ?

[Bitmanhome]

Moore's law says double the number of transistors every 18 months. So a year and a half after Opteron, a dual-core would be par, so AMD is actually a little late. But a quad-core a year and a half after the dual would also be par.

Re:New trend ?

[scribblej]

The "integrated" in "integrated circuit" surely means the circuit is integrated into a single chip. That's what it always meant when I was a kid. We had plenty of systems where processing power was distributed over multiple processors, but a logical circuit made up of NON-integrated circuits is not an IC.

Re:New trend ?

[Anonymous Coward]

... what? That made absolutely no sense.

Re:New trend ?

[TyrranzzX]

Uhhh, The pentium 1 had 2 fused 486 cores in it. Technically, you can say "well, each pipeline is a new proc". We're sure to see new and interesting techs come out, probably clustering pipelines together or something. Afterall, there's only so many ways you can cut glass.

Re:New trend ?

[Anonymous Coward]

Erm, Moore's Law refers to the number of transistors doubling on a chip.

Two cores on one chip certainly satisfies this requirement.

Re:New trend ?

[Anonymous Coward]

I guess Fragfest 2015 will be held at the Boeing's main assembly plant in Everett, Washington with an expected jam packed crowd of 5 whole computers fit into the hanger!

(measuring 473 million cu. ft. it's the worlds most capacious building :)

Re:New trend ?

[OAB_X]

A dual-core gpu card would offer much more performance then just videocards in SLI mode. They would use nasty amounts of power (and would certainly need to be in PCIX setups), but they would offer a much higer performance then SLI mode cards. Of any type.

Re:New trend ?

[Guignol]

you mean dual,
if you have duel video cards/CPUS, it's like saying your car has 200 hp, but 100 of them are running backward :)

Re:New trend ?

[Jynx97]

I, for one, would like to see a beowulf cluster of these in action....

Re:New trend ?

[Solosoft]

Im still one of the suckers who is running a 3Dfx Voodoo 5 5500 PCI on my Dual Pentium Pro 200MHz.

The card is simply perfect for what I do with it but I do wish that the SLI would of been completed for this card so it's not just seen as a 32mb Voodoo 4 with 1 166MHz GPU but a Voodoo 5 with 64mb of RAM and 2 166MHz GPU. Plus the AA would look pretty I bet on some games and such.

Oh well, I spent some extensive time reading on the internet and alot of people took note that this card's linux drivers will never be finished.

Quick Question tho: Does nvidia's new SLI use the same technology as 3Dfx's. I remember reading this somewhere that they where going to use the same SLI as the voodoo 5 (nvidia did buy out 3Dfx's technology).

Re:New trend ?

[Ohreally_factor]

Are you suggesting pixels at dawn?

Re:New trend ?

[minus9]

"Once again, the enthusiast's computer will be larger than a refrigerator!

Well, sure, the Frinkiac-7 looks impressive, don't touch it! But I predict that within 100 years computers will be twice as powerful, 10,000 times larger, and so expensive that only the five richest kings of Europe will own them.

Re:New trend ?

[Anonymous Coward]

Once again, the enthusiast's computer will be larger than a refrigerator! ;-)

.. and require one.

Re:New trend ?

[Anonymous Coward]

There were four models based on the 3dfx VSA-100 (Voodoo Scalable Architecture) chip:

Voodoo 4 4500 (1 VSA-100 chip, AGP or PCI)
Voodoo 5 5000 (2 VSA-100 chips, PCI)
Voodoo 5 5500 (2 VSA-100 chips, AGP)
Voodoo 5 6000 (4 VSA-100 chips, AGP)

There might have been a "Voodoo 6" designation published initially, but 3dfx dropped it long before any of these appeared in the market, so it was just the "4" and "5". (The next generation would have been "Rampage", of course.) The V5 5000 didn't sell in large numbers, and the gigantic V5 6000 card never made it into the retail channel (due to constant small problems with the AGP interface, or so rumors had it).

Actually already the 5500 was surprisingly close to a GF2 in game benchmarks and in people's experiences, maybe thanks to the good memory bandwidth (dual-chip SLI). The 6000 would have given the GF2U some hard competition (and the GF3 was rarely if ever faster than a GF2U except at the handful of PS1.1 games) -- it actually had more raw single-texturing fillrate than the GF2U, and more memory bandwidth.

*sigh* I got the GF2U anyway, 3dfx was visibly dying, and a single-chip solution was better for me.

Re:New trend ?

[benjaminchoate]

I think with all those processors and GPUs more likely it will be in the refrigerator ;)

Looks better on SLI Voodoo2's than on my Rad 9800.

[Lostie]

...because you can see more than 1 colour (black).

Who to Trust

[superpulpsicle]

Every review I have seen has claim SLI to be the wave of the future giving you ridiculous speed boost. But don't all video card reviews do that now? Last I checked on some of the older Tom's hardware, anantech reviews, my hardware should be polling in 70 fps for some games. I'd be lucky to hit 35 consistently... that's reality.

Re:Who to Trust

[Hoi Polloi]

But with SLI you'll be able to use two boards to reach the fps you were promised to reach with your single board!

See, it works out in the end.

Re:Who to Trust

[ArsonSmith]

I don't know they give an example of a single G256 compared to dual G256 in SLI mode and it is almost 10x faster. With the addition of one card I get 10x faster that sounds great to me.

-Both logic and RTFA will hurt me

Re:Who to Trust

[mikael]

Every review I have seen has claim SLI to be the wave of the future giving you ridiculous speed boost. But don't all video card reviews do that now?

If you look at the history of video cards, you will see that whenever they succeed in reaching the limit in one particular technology, they will continue to move on something else. First it was screen resolution, then pixel depth, followed closely by 2D pixblitting, then 3D acceleration, multi-texturing, then programmable vertex and finally fragment programs, with the hardware then supported dual DVI output. Since vertex and fragment programs are already written in high level languages and accelerated using multiple pipelines to the limit of clock speed and die size, SLI is the only path left.

Intel & SLI

[DeadBugs]

It is worth noting that NVIDIA will be bringing SLI to the Intel platform according to this press release:

http://www.nvidia.com/object/IO_17070.html

I'm looking forward to a P4 NForce board.

Re:Intel & SLI

[Jeff DeMaagd]

There is already an Intel-based PCIe board that has dual graphics slots, I think based on the 925 chipset.

The only thing different here is that nVidia might introduce a cheaper way to get such a board.

Re:Intel & SLI

[OmniVector]

i'm really not trying to be a troll or anything, but why would you go intel right now when the amd 64 chips have so many advantages (price, performance, power, heat)?

Re:Intel & SLI

[Anonymous Coward]

"I'm really not trying to be a troll, but here's some opinion totally unrelated to anything you said that tries to put down one company without evidence."

Re:Intel & SLI

[OmniVector]

i don't feel like posting links to prove every point, but each one is true. amd chips right now are cheaper at a price per performance level. their top end (4000) is faster than the top end pentium (3.8ghz). their top end chip consumes far less power than the 3.8ghz, which is around 100W or more. they're better chips in general with 64bit support and the NX bit.

Re:Intel & SLI

[obeythefist]

Probably because ATI is doing an intel chipset, which seems all the rage these days.

One begs the question, however, as to why one might want to purchase the premiere gaming chipset on a mainboard, and couple it with the least cost effective gaming CPU on the market (the only worse CPU you could buy for gaming would be an itanic, again from intel).

Intel chips are really best for repetitive processing tasks like video and audio encoding. Gamers prefer Athlon. If you're not sure, check the prices, check the benchmarks.

A plus...

[ReeprFlame]

Dual video cards are a plus. Not only wil lthere be one GPU processing your requests, but you have the better ability to use trwo displays with more performance. As the future goes, boxes will be able to process more and will need to display more graphics intense software. Also, I believe that it is a good idea to have one box to power several stations and that is where several video cards come into play.... One will not be acceptable several years into the future especially for performance power users [gamers, multimedia, etc].

AlienWare

[Spez]

You can already buy from the alienware luxury collection some gaming systems featuring SLI

http://www.alienware.com/ALX_pages/choose_alx.aspx

SLI?

[Anonymous Coward]

What is SLI?

Re:SLI?

[sjaskow]

SLI = Scan Line Interleave, the cards alternate drawing lines on the monitor.

Re:SLI?

[stratjakt]

Scan Line Interleaving.

Two video cards, one draws all the even scan lines for the final display, and one draws all the odd ones.

Re:SLI?

[Gates82]

Not true anymore. The new SLI format does not stand for the scanned lin..... Rather then having one video card take odd lines and the other even the cards actually break apart the frame to be rendered and determine where more detail (thus processing power) is need and the concentrate on those areas, so card one may be calculating more pixels, and card to might be concentrating on a detailed area, but they are both doing about the same amount of computation.

--
so really, who is hotter? Alley or Alley's sister?

Re:SLI?

[Gates82]

Hate to reply to my own post but here is a link. Nvidia SLI SLI = Scalable Link Interface

--
So really, who is hotter? Alley or Alley's sister?

SLI != SLI

[Jahf]

First it is mildly interesting to note that SLI from Voodoo was "scan-line interleaving", as in every other line was alternated between the 2 cards. Nvidia SLI is "scalable link interface" and instead renders the top half of the image on one and the bottom on the other.

It does make me wonder if the technology is capable of truly scaling ... ie ... more than 2 cards? Could be useful for scientific simulations or even getting closer to the idea of "ToyStory in realtime" (and no arguments here ... using the same shaders as Pixar used in the movies in realtime is not feasible today ... cheap tricks to get close, maybe).

However, given the cost, and looking at what the 6800 can handle by itself, and comparing -those- to the evolution of games it appears to me that it will be no more costly to simply upgrade to a 6900/7000/whatever when it is required, as I can easily get by for the next year or two on a 6800 Ultra especially if including the fact that I would need a new computer to run it on since I don't have PCI-E (though I do have PCI-X, but not for gaming needs). And will be saving on electricity and mean time to failure (though that doesn't seem to be an issue much with video cards).

Not saying I don't see the attraction, but I don't get anywhere NEAR interested in 3D gaming enough to be spending that kind of dough.

Re:SLI != SLI

[Linuxthess]

Could be useful for scientific simulations or even getting closer to the idea of "ToyStory in realtime"...

I propose a new acronym, TSIRT which will be the standard of rendering performance, similar to the "LOC" (Library Of Congress) reference when comparing download speeds.

Re:SLI != SLI

[Anonymous Coward]

I see this as nVidia buying the wrong assets from 3Dfx. The riva series and early GeForce cards were able to make design choices that greatly increased the capacity of the cards. With the latest cards I see the same behavior that 3Dfx had just before going under. Don't innovate and make better, just throw more GPU's at the problem. Pretty soon I expect to see an GeForce 7700 that is two 6900 GT's on one card, with the option of SLI for the processiong power of 4 GPU's. With a little bit of ducting your computer will be able to hover.

Re:SLI != SLI

[Fortun L'Escrot]

think of it like this. with the option to go dual gfx core you can still continue to upgrade your gfx card as normal.

BUT if you do not want to spend that much dough, you can always buy a cheaper card and stick it in.

UNLESS this Nvidia SLI thing requires you to have two identical cards in which case many gamers will just get budget to mid-range cards and when newer games come out double up on their cards. either way it means you can upgrade for cheaper but at a price ie power consumption.

Re:SLI != SLI

[Anonymous Coward]

First it is mildly interesting to note that SLI from Voodoo was "scan-line interleaving", as in every other line was alternated between the 2 cards. Nvidia SLI is "scalable link interface" and instead renders the top half of the image on one and the bottom on the other.

Why don't they render the left and right side instead of top and bottom? Is it because it's easier to sync with the beginning of the horizontal sweep of a CRT?

Re:SLI != SLI

[ArsonSmith]

I created a top notch 3d rendering engine and all I got was this TSIRT.

Re:SLI != SLI

[Anonymous Coward]

Throw 2 TSIRT PC's in a lake, compare benchmarks ==
Wet TSIRT Contest!
sorry.

Re:SLI != SLI

[JoeNiner]

First it is mildly interesting to note that SLI from Voodoo was "scan-line interleaving", as in every other line was alternated between the 2 cards. Nvidia SLI is "scalable link interface" and instead renders the top half of the image on one and the bottom on the other.

Actually, nvidia's solution does either, based on their own testing of which performs better for a given game. The drivers include profiles of the 100 most popular 3D titles which state which technique to use.

Re:SLI != SLI

[man_ls]

Vertical sweep is measured in Hz.

Horizontal sweep is measured in kHz.

That, and the fact that CRT monitors lend themselves to horizontal divisions (top/bottom, not left/right) since they sweep top to bottom during refresh.

Re:SLI != SLI

[Charcharodon]

3d gaming? Wouldn't you like 2-4 monitors for a desktop or for video or photo editing (drooling at the thought.)

Everyone keeps getting caught up in the idea that you would have to use this with just one monitor. I see it's potential much like a raid controller. Sure you can have two drives run together twice as fast but you can also use it to control the drives individually and increase the number you have. I can run either 2-3 RAIDs with 4 IDE devices on my computer or control 10 IDE/SATA devices without the raid.

Run the cards together to get some screaming 3d performance or seperately to control a 4 monitor desktop array with absolutley loads of space to work on. I can't wait.

Re:SLI != SLI

[Jahf]

Ummm ... there are MUCH MUCH MUCH cheaper ways to get 2 or more monitors.

Video and Photo editing don't -touch- hardware 3D acceleration functions, which means that you are a couple of hundred w of electricity and a few hundred $ of hard earned cash if you were to use SLI for that.

Nvidia even recognizes this by their NVS280 card ... essentially a 6x00 series card that will drive dual monitors but with none of the 3D acceleration tech. A current generation Nvidia dual-DVI card for $160 without the need for external power.

I'm quite sure (can't say why, unless they've released it) that there will be a PCI-E NVS280 which means you can add multiple dual-DVI cards for much less than dual-DVI 6x00 cards.

And ya know ... between virtual desktops and a KVM, I've been -fairly- happy with my monitor for a couple of years now. I am anxiously waiting to be able to afford a 24" widescreen LCD but until I can I'll stick with what I have. The cost of a 4 screen system with even mid-range (say 25ms 17") LCDs (unless you want to have 300lbs of monitor on your desk) is going to cost a ton.

I guess if you have the money to waste on multiple top of the line 3D cards and the electricity over their lifetime, you might have the money to spend on those LCDs ... in which case lucky you ... but I don't think most people miss your point so much as have other things to pay for first.

Re:SLI != SLI

[Jahf]

Not completely certain, but yes I do believe you're correct about the horizontal-ness of CRTs.

Even with LCDs, most drivers if not all -expect- the data to flow that way.

I think I see where you are going ... an advantage of L->R (or R->L) would be the ability to split the screen into stereo images. There have actually been APIs and I think even hardware specific pieces (I remember glasses that did this, but you of course had to have the software split the image to the lenses) that did this, so it really wouldn't address much. Especially with the ability to do multiple monitors, since you can now render to individual monitors instead of splitting up a single.

Then again, I'm way out of my field at this point :)

Re:SLI != SLI

[alexo]

AnantTech has a nice article about the Nvidia SLI.
It includes an explanation on how it works, power consumption and benchmarks of several games.

Re:SLI != SLI

[Jahf]

Not sure about the identical bit. In theory you're supposed to be able to use an Nvidia 6600 PCI-E from brand A with one from brand B, but we all know how that can go.

If you mean can use plop a 6600 in with a 6800 or a future card, I don't believe so (maybe ... doubt it though since then you'd have to compensate for one card being better than the other).

As for the option to upgrade, yes, if you already have a PCI-E capable machine. If I remember correctly it took about 2 years for AGP to become the defacto standard (as in Best Buy can assume if you are buying a video card, they can assume safely that it is AGP in almost every case). My point simply being that very many people do not have the ability to run these SLI cards. It is those people (me :) that I was pointing out will probably just find it cheaper to wait until there is a faster card and upgrade it.

In my case for instance I have a dual Opteron 246 box with AGP. That is going to be plenty fast for me for a long time (I was using a P4 2.4 laptop for gaming happily until this season's games came out). I am willing to bet that my CPU capacity will be within the minimum recommended specs for games for 3 years or more (and after that who knows, maybe I won't want to game as much or maybe I'll buy a gaming system). My video card, however, will likely only do for 18-24 months. It will be much easier for me to go from my 6800-GT to a next-gen card than to upgrade to a PCI-E system so that I can use SLI.

Not to mention the fact that cards and games tend to move to new APIs every year or two. What happens if a new API becomes the power-gaming standard and you are now stuck with either buying a second SLI card with the old standard (consider that -2- GeForce 5200's have less power than 1 6800 Ultra on DirectX 9 benchmarks) or upgrading to a new card anyway.

For people who have a PCI-E system, assuming the next gen cards and games don't move to DirectX11 / OpenGL3, upgrading by adding a second card is great ... that's alot of "if"ing though.

So my opinion (and that's all any of this is) is that if you want SLI and have the money to spend, get it now. If not, AGP cards will likely be produced for a few more years and so wait until you need to upgrade your entire system and -then- if you have the money for SLI buy it all at once. I understand the incremental upgrade, but having tried it in the past (do I buy a faster 486 or do I upgrade to a Pentium? Do I buy my 2nd Voodoo card or do I go ahead and get that cool-fast ATI card that requires me to get an AGP mobo? etc) I just don't see it being as truly practical as it seems in theory.

Re:SLI != SLI

[Jahf]

It is the trend of computing in general right now to go with dual-core.

Sun (UltraSPARC), AMD (Athlon64/Opteron) and Intel (not sure which one they are planning on ... guess where I work *lol*) are all planning on multi-core CPUs within the next year or two.

Not just dual core CPUs (that will be the first round) but 4-way and maybe later 8-way CPUs. At the same time the CPUs will become marginally faster ... but not too much. It is the speed blocks they are hitting that is causing this.

And that doesn't count SMP machines or Hyperthreading. All of these groups are talking about 2, 4, 8 and more-way machines. Imagine an 8-way 8-core CPU machine ... literally 64 CPUs (64 CPUs has been possible for a long time but not in such a compact form ... and imaging a 64-way 8-core machine ... maybe 5 years but still ... that's a lot of CPU). I've seen slides from all 3 talking about this basic direction.

I think you'll see that dual-core GPU in an SLI configuration soon, probably around the time you see a dual-core dual-CPU (64bit of course).

My opinion is that 3D gaming is KeWl but -good- games need logic and heart, not more polygons. I would really like to see developers take more of their resources and place them into those areas ... we've got enough horsepower today to last me for a long time :)

Re:SLI != SLI

[menkhaura]

I was thinking another thing: one card renders one frame, the other cards renders the next frame, the first card renders the next, and so on, with both cards alternating the frames. This could double the frame rate as compared to a single card, and/or improve the image quality by processing it while the other card spits out its frame.

Re:SLI != SLI

[Thaelon]

IF you'd RTFA you'd know that it doesn't render halves of the screen. It uses load balancing similar to that found in multiprocessor systems. While most applications aren't written to use two cpus simultaneously, the system can distribute the total system load very evenly provided no single process uses 100% of a CPU.

So it's not rendering "halves" of the screen, each GPU is doing an equal share of the rendering work so that while the work of one card may only be a small portion of the screen they both did the same amount of mathematical work. I read about this four days ago on http://www.hardocp.com/.

Re:SLI != SLI

[Charcharodon]

Yeah I see your point, but hey I like using the 3d accelleration too. I had a pair of 19" CRTs before and yes it was way too heavy. The table I used for a computer desk began to bow in the middle. :)

I see the multi-monitor systems becoming more popular once the high end $400 cards of today become the $50 cards of tomorrow, or simply being able to continue to use your "slow" graphics card for an extended desktops with your latest and the greatest running your games.

It'll be interesting to see how things go since the next version of windows is supposed to dump, or at least add to, the 2d desktops in favor of 3d desktops. (You can still have your 2d by simply showing one side of a square/shape.)

Re:SLI != SLI

[man_ls]

This would only double your framerate if the bottleneck framerate is 1/2 your vertical refresh.

"Visible" framerate is limited to the vertical refresh of your monitor, anyway: the electron gun will physically not draw more frames on your screen in 1 second than it is swept across. (This is where vsync in games comes into play, to prevent "tearing", where the image is being redrawn so quickly in the GPU that the scene doesn't remain constant even between vertical sweeps)

It would probably be about as efficient as scan-line or split-scene interleaving, since most hardware of the caliber you would run in SLI/SSI mode wouldn't be bottlenecked at half the vsync in the first place.

Re:SLI != SLI

[lendude]

Slight correction - Nvidia's SLI solution doesn't do the old Voodoo scan-line interleaving: it either does SFR (split frame rendering i.e. a portion of each frame is rendered by each gpu) or AFR (alternate frame rendering i.e each gpu renders each alternate frame completely), as per the driver profile you mention.

Re:SLI != SLI

[Howzer]

Get TSIRT into print, so you can be the originator! I've already used it twice in conversation today...

Double The Money

[Squeebee]

So the guys at Nvidia were sitting around when in walk the PHB and says "Guys, we need to make more money". And flunkie one says "Hey, let's release a new card, all the fanboys will rush out and buy it!" PHB says "Well that's ok, but we do that enough already". Flunkie two says "I know, let's convince the users that the one overkill video card they buy is not enough, let's convince them that the need to buy TWO!" And the rest my friends, is history! Stay tuned for the new quad-card cash-vacuum, coming soon.

Re:Double The Money

[Thought Harvester]

And I went out and bought one.

You know what? Comments like yours are worthless. Thanks for your opinion that you think gaming isn't worth spending money on. The fact of the matter is, I am a gaming hobbyist. I like games, and I really like games running well on my rig. Setups like this push the dollar envelope, true, but how is it any worse than spending $1000 on a new golf driver?

Come to think on it, SLI is better than a driver because the improvements are evident and more dramatic compared to more inexpensive solutions. It improves my overall gaming experience and in my mind is worth every penny.

Why don't you tell us what your hobbies are, so the collective group can crap all over them.

Re:Double The Money

[ArsonSmith]

Doesn't this imply almost the oposite of what you say. Isn't it more like pay $400 for a video card and 6 months later a new twice as fast $400 card comes out, well the one I bought 6 months ago is only $40 now, sli twice as fast as my old card and only $40. By the time I needed quad cards they'd be in the bargin bin $10 with $10 rebate.

Re:Double The Money

[hooqqa]

This is absolutely retarded. With the voodoo sli you get to buy a new piece of hardware without throwing away your old card. This is more like, "Would you like to buy /half/ a video card?" - not that it's suprising or anything...

Re:Double The Money

[Anonymous Coward]

Apparently his hobby is making jokes, something you arn't terribly familiar with. Learn to laugh at yourself, it does wonders.

Re:Double The Money

[Ann Coulter]

I would gladly buy two nVidia based PNY Quadro FX 4400 cards with a dual Opteron motherboard that supports SLI. I would use these two graphics cards in non-SLI mode most of the time so that they can drive 4 1600x1200 (or 2 3840x2400) screens at the same time, and use SLI only to simulate very detailed environments. I would also buy Matrox QID Pro cards that handle 4 monitors per card, a total of eight monitors. This setup would cost at least $5700 for the video cards alone ($900 for the QID Pro and about $2400 for one Quadro).

Quadros have 128 bit IEEE compliant floating point processors. This translates to extremely precise numerical analysis that occurs in a super-parallel (or doubly super-parallel) platform. I thought I was dreaming when I heard that the GeForce 6800 GPUs had 64 bit floating point processors. Now these Quadros are godsends as I can develop algorithms that have not even dreamed of planning on because of the lack of a parallel architecture that supports the kind of precision I need.

For simulation purposes, I have not worked with much 3D objects (or projections onto 3D space) because of the lack of precision of most desktop cards. With workstation cards, I can finally analyze 3D point sets and the like.

In conclusion, the newer Quadros are definitely worth the money to me.

Re:Double The Money

[Raptor CK]

More like, "Hey, the last generation videocard is now obsolete, and no one wants it! How do we fix this next time?"

"I know, let's make it so that if you buy a second one a year later, it'll work WITH the first one!"

No one needs to buy two right off the bat. One is usually more than enough for any modern game. But one for a few hundred now, and the other for less than $100 later? That's a bargain basement upgrade, and one that's far more sensible than getting the new mid-range card now, and the new mid-range card a year from now.

Now, if someone *wants* to buy two top of the line cards today, more power to them. They want the ultra-high-resolution games with all the effects cranked up, and they have the money. It makes their games look nicer, while my games run well enough. We both win, and Nvidia no longer sits on piles of unused chips.

Re:Double The Money

[Anonymous Coward]

Thanks... forget all the bullshit and pretention, it makes me so happy to see someone who knows what they want, and above all how to use it.

Don't lose that.

Re:Double The Money

[freakmn]

I've not had the opportunity to try out this new hardware, but as I understand it, the SLI requires a connecter to connect the video cards together. I would assume that this is how the switch is done between dual cards and SLI, so you may need to open up the case to get the results you are looking for. Just a heads-up, so you aren't disappointed.

Re:Double The Money

[afidel]

Why not buy a $120 card now which will play any game at decent resolution and framerate and another newer card in 18 months or so when the next round of demanding games come out? You're still spending way less than one top of the line card and the only thing you can't do is turn on all of the eye candy or play at insane resolutions. (the one counterargument that I can think of is needing good framerates for a large lcd's native resolution, but I play turn based strategy games so what do I know =)

Re:Double The Money

[Anonymous Coward]

Yes, learn to laugh at yourself, I know I am laughing at you as you blow that much money on a card that will be outdated in 2 years. Here's my hobbies...pc tech..gamer..console gamer..baseball...paintball....now try crapping on them. Note that I wouldn't spend 500 let alone 1 grand on ANY single item for those hobbies. That's the difference you "spenders" miss. It's not what your buying, or your hobby..its the actual use you get out of it. If you think playing a game, with the tiny difference the more expensive card makes, is worth 1000 bux then you need to go get a life...same for people who spend 1 grand on a bat, a paintball gun, etc etc...the ONLY time this isn't thought to be totally WACKO is if you are using it to make MORE money such as tournaments etc. And don't think your local piss on town tournament counts either asshat.

Re:Double The Money

[Anonymous Coward]

IHBT

Re:Double The Money

[10Ghz]

What makes you think that current vid-cards are "overkill"? I mean, it seems to me that they are already struggling with Doom3 and Half-Life2 when using higher resolutions together with FSAA and AF.

Re:Double The Money

[Raptor CK]

Funny you should mention that. I *do* plan on moving on to an LCD, and will need all the processing power I can get my grubby paws on.

I just figure that if someone's going to spend $300 on the latest and greatest card every two generations, you might as well hook them in for buying a dupe of that card for $100 in 18 months.

And, well, the target market for a $300 card does want all that eyecandy turned on. I know that I can't live without running the highest resolution textures available. :)

this is sweet

[the_2nd_coming]

because the current crop of high end cards are physically incapable of rendering Doom 3 on the best settings. that requires 512 MB or VRAM. with SLI, you put 2 256 MB video cards in and you can!!!

Re:this is sweet

[stratjakt]

No, it doesn't work that way.

Each card renders half of the same image. So each card needs access to the full texture set.

So 2x256 cards still only gives you 256 megs for your textures.

Re:this is sweet

[LordKronos]

Each card renders half of the same image. So each card needs access to the full texture set

Well, not always. That is definitely true when using the 3dfx style SLI, where both cards are rendering pretty much the entire screen. However, in the case of this type of SLI, where the first card renders the top, the second one the bottom, you can end up with optimizations. Assuming the sky only appears at the top of the screen, your skydome textures only need to be loaded on one card. Sure, it's not 100% efficient, but it's not 100% redundant either.

Re:this is sweet

[stratjakt]

I didn't know it worked like that.

I suppose it's not properly termed SLI, since that means scan line interleaving, which isn't what you describe.

If they put their thinking caps on, I'm sure they could come up with some funky card-to-card bus that would let each GPU access both cards video RAM as one contiguous unit.

power consumption???

[Hackura]

My question is whos got the 1100watt power supply that running 2 6800's is going to "require"?

Re:power consumption???

[Anonymous Coward]

The same person that has 2 motherboards, 2 cpus, 2 harddrives, etc to go with the 2 6800s.

1 6800 will not take up all 550w of a power supply.

(And yes, I know the post was a joke)

Re:power consumption???

[Mr.Radar]

You don't need a 1100 watt PSU, just a 500+ watt one that can output the amount it claims to (like the Antec TrueControl 550, the Fortron/Sparkle* FSP550-60PLN, the OCZ PowerStream 520, or the PC Power & Cooling Turbo-Cool 510 Deluxe). Most PSUs don't live up to the "ratings" quoted on their stickers in real-world scenarios.

*Sparkle products are rebranded Fortron ones with the only difference being the name on the sticker and the price (Sparkle branded Fortrons tend to sell for less than Fortron branded Fortrons).

SLIing other GeForces

[shadowzero313]

Is the SLI only compatible with the new GeForce 6x00 series or can you use an older GeForce set?

Re:SLIing other GeForces

[mesach]

its only compatible with like cards capable of SLI, you cannot just throw 2 cards in your box and run the latest drivers and get SLI.

there is a bridge adapter for the cards, if you look around they apparently come in PCB and Ribbon styles, and connect to a funky new cutout on the PCB on top of the Card.

Re:SLIing other GeForces

[shadowzero313]

damn, i can't get a new board and a second GF4Ti. Time to get a job.

Re:SLIing other GeForces

[teg]

Is the SLI only compatible with the new GeForce 6x00 series or can you use an older GeForce set?

Only some of the new GeForce 6x00 cards (not all) can be used for SLI. You need a special connector on the cards.... also, there are no PCI express versions of older GeForce cards anyway AFAIK.

Re:SLIing other GeForces

[afidel]

That's because the work is split dynamically between the cards based on scene complexity, so if you have a simple skymap taking up the top 1/4 of the screen the cards will probably split the screen 5/8-3/8 so that they are both doing the same amount of work. This is vastly superior to limiting performance to the slowest card.

Note to self

[Scoria]

Please substitute resort for result, and never program a database query manager while commenting here.

Should have invested in...

[vasqzr]

Dual webservers. Would have delayed the Slashdotting.

Re:Should have invested in...

[Anonymous Coward]

Writing your comment in bold made it twice as funny. Next time try italics.

Re:Should have invested in...

[grazzy]

"Did not know how they was fucking with."

Re:Should have invested in...

[Anonymous Coward]

...until the webserver gets dual-slashdotted!

Re:Should have invested in...

[Anonymous Coward]

oh slashdotting jokes are so funny oh oh omfg!!!!!11111!!!1oneone!!!

Re:TWO Cards!

[Rew190]

TWO Cards! (Score:-1, Redundant)

I think that's the first time the actual moderation of a post has made me laugh more than the post itself.

Ironic?

[goldspider]

I find it funny that some of the people who lamented the $15/mo. for WoW in the last article are probably the same people who will go out and drop $600 for a top-notch SLI video setup.

Re:Ironic?

[Surt]

Situational ethics is the only acceptable kind of ethics to use, no matter what the situation!

Re:Ironic?

[pyite]

That's probably because daddy's more willing to tolerate a one time "gift" to shut up his son than a monthly recurring reminder that his child is spoiled.

Re:Ironic?

[Anonymous Coward]

I don't know how this gets "Insightful" mods without even one example of the hilarious irony the poster smirkingly "points out".

Baseless speculation is not insightful.

So get to it, goldspider. Find someone who lamented the WoW fee and is in here posting about how they're going to buy an SLI setup.

why so little support for gamers?

[sowdog81]

Why aren't there more quad/dual processor motherboards built with gamers in mind? It seemed like an inevitable choice for the must-have-it gamer? There is a big enough market that there are 500usd graphic cards and 700 usd processors. It be nifty if this sort of thing had become mainstream :D

Re:why so little support for gamers?

[dead sun]

What's the last well parallelized game you played? I had a dual Pentium 3 setup and while it was certainly nice for other things, any gaming I did was only really helped by being able to run system processes on the second cpu.

I would absolutely love to have a dual desktop again, but mostly just because it was more responsive and handled multitasking under load so much more gracefully. Little better in computing than having a processor running at 100% and still having a usable desktop running on the other processor.

Re:why so little support for gamers?

[meestaplu]

Right now, the answer is pretty simple. If you want a game to use multiple processors at the same time, you need to include more than one execution thread--the programmer has to divide the work in such a way that two or more processors can do it. It's quite hard to build a multithreaded game; there was some SMP support in Quake III, but it wasn't very stable and didn't provide a huge performance boost.

With a multithreaded application, you have to guard against strange bugs that are very, very hard to fix. If your multithreaded application runs into a deadlock every hundred thousand frames or so, it will be next to impossible to isolate, and production will end up being slower than it already is. While I'm sure that writing multithreaded games will happen in the near future, I don't think it will catch on very quickly.

Re:why so little support for gamers?

[tygerstripes]

The answer's pretty recursive; there aren't enough gamers with dual/quad processor platforms out there to warrant coding big-name games to make use of multiple-processors. So, there aren't many gamer-oriented multi-processor platforms out there... Besides, there really isn't much in terms of processing power that people demand at the moment. There aren't any games that would benefit more from a twice-the-price processor than from a twice-the-price graphics card. When it comes to gaming, we don't want clever - we just want pretty. Sad, but true.

Re:why so little support for gamers?

[Enrique1218]

What games actually use the second processor? Quake 3 was the only game I knew of that could use two processors. I believe the gain was minimal. This is probably why ID elected to drop the feature from Doom 3. Moreover, the gpu is more imporatant than the processor when it come to performance. Alot of decent gaming systems will use a mid-range processor and put extra money inro a high end video card. One thing, I would like to know is whether games would benefit from being design for 64-bit processors .

Re:why so little support for gamers?

[sowdog81]

Thanks for all the replies. I had the impression that many games where built as multi threaded apps already. If this is the case how will the upcoming multicore processors affect gaming?

Re:why so little support for gamers?

[lp_bugman]

The new Playstation if I remember correcly will be dual core based. So spect great advances in multithreaded gaming.

We only need to wait for some IA, Rendering Engine or Physics Engine to start using multiple cores.

Re:why so little support for gamers?

[dreadlock9]

Excuse me, I beg to differ. I've heard a lot of people say that Doom 3 does not use SMP, so I decided to find out for myself.

I have a dual Athlon 1900+ box with XP, and dual displays. This allowed me to perform an experiment with Doom 3 to see if it uses both CPUs. I put the task manager on my non gaming monitor, killed all extra tasks, and fired up Doom 3. It did indeed use both processors, about 20-30% in the menu screens. During gameplay it stayed at about a constant 50% per CPU. I wonder if it would use more CPU if I had a faster video card (I have a GeForce Ti 4400).

Re:why so little support for gamers?

[John Courtland]

I think you'll see a good boost if a game was being written/compiled specifically for AMD's 64-bit extensions, solely because of the availability of more *visible* registers (there are masked ones on the current IA-32 processors, but you can't, as a programmer, access them, AFAIK). I think it quadruples the amount of general purpose regs (some can argue the IA-32 has 8, I say 4, because 4 of those "GP's" aren't really general) and adds quite a few floating point regs.

Re:why so little support for gamers?

[adamdeprince]

FPSs are not the sort of applications to sit back and say "I've had my fill of CPU time (select, yield, whatever)" for as soon as one frame is done, it is going to start up with the next one. Doom3 consumed 100% of the virtual processor that it requested on your machine. Your silly OS seems to have punted the process back and forth between CPU's.

If anything, you are losing performance to your second CPU. Bouncing between CPU's means that each time you bounce, your contend with a stale cache and the possiability of redundant loads. Not to mention the sudden presence of locks that can be so conveniently compiled out of single cpu versions of OS code.

Now if you can convince your OS to keep Doom on one CPU, you might find that the other picks up the lions share of house hold chores such as networking, responding to keystrokes, spyware, etc etc .. but that wouldn't be fair to make one do all of the chores while the other plays.

Re:why so little support for gamers?

[Anonymous Coward]

The next Xbox is also "multi core" at least in the sense that it will try and run multiple threads at a time (some sort of bastardization of hyperthreading?).

most famous acronym?

[Anonymous Coward]

Come on, how useful can this article be, consider that the author opens with the statement that SLI is "that most famous of three letter acronyms." Huh? How about ABC, CPU, etc? Hell, even GPU is a better known acronym, and that's in the 3d world itself! The article was interesting, but the author needs to learn to moderate his or her prose.

That giant sucking sound...

[Aggrazel]

LOL a link off the front page to a page filled with hundreds of screenshots?

I weep for that man's router.

Ouch on Costs!

[Evil W1zard]

I can't imagine shelling out another couple hundred bucks for another XT pro and then shelling out even more money for a more robust power supply and better cooling as well. Its prolly great for those who can afford it, but I know I won't be doubling up anytime soon.

What I'd like to see..

[Smidge204]

I'd like to see something set up so onboard video hardware can take advantage of this. It's difficult to get a motherboard that doesn't have onboard video anyway, and if you buy the right video card (ie: same manufacturer) they can both run to get an added performance boost. (You should, of course, be able to install any graphics card, but won't get anything extra for it)

=Smidge=

Re:What I'd like to see..

[macklin01]

Nice idea.

On a related note, I'd like to be able to use SLI with two cheap video cards. Perhaps two cheap vid cards could be work together to give performance comparable to that of a much more expensive single card. -- Paul

Re:What I'd like to see..

[The Other White Boy]

the 6600GT's are in the $150-200 range. two of them in SLI according to HardOCP keeps up with and sometimes surpasses a single 6800 Ultra, which would run you $500-600 last i checked.

Re:What I'd like to see..

[macklin01]

Thanks for the heads-up. That sounds like a great trick for the future ...

Alas, $150-$200 is still out of the "cheap" range for a married graduate student ... but a very nice thing to keep an eye out for on ebay! :) -- Paul

Re:What I'd like to see..

[Anonymous Coward]

"Alas, $150-$200 is still out of the "cheap" range for a married graduate student"

• You made your bed...

Re:What I'd like to see..

[Eraser_]

"It's difficult to get a motherboard that doesn't have onboard video anyway"

Just to be an ass, it is hard to get a $30 motherboard w/o onboard video, and all sorts of other crappyness. Those boards aren't meant for any sort of real throughput or use. Word processing? No problem. MP3's? OK. Spyware out the wazzo? Preinstalled! High end hardware accelerated 3d rendering? Hahahhahaha.

But for $65 you can get a MSI Delta something other other, nForce chipset, no onboard video junk. Thats pretty dirt cheap. Comes with onboard audio and NIC, but who cares? They also make them w/o audio, and they make server boards w/o nics.

Re:TWO Cards!

[Anonymous Coward]

I think you stumbled onto something.

From now on, every article about SMP, clustering, etc. will have troll posts for the express purpose of getting Redundant mods.

See also the UK "PC Pro" magazine

[Pete (big-pete)]

This month the UK "PC Pro" magazine has a review of the Scan White Cobra gaming machine.

This is a fine example of SLI running with jaw dropping performance...a quote from the review puts Doom 3 running at 98fps!

Now I know what I want for Christmas, just not a snowball's chance in hell of getting one! :)

-- Pete.

Re:See also the UK "PC Pro" magazine

[Elwood P Dowd]

Now I know what I want for Christmas, just not a snowball's chance in hell of getting one!

A snowball would fare worse in a Scan White Cobra.

Re:See also the UK "PC Pro" magazine

[Jormundgandr]

Any modern gaming rig can run Doom 3 at 100 fps... please post the video settings next time, so we can be duly impressed.

Ummm....huh?

[Anonymous Coward]

Would it be difficult to tell the unwashed masses (me included) what an 'SLI' is??

Yes, I could eventually figure it out, but if it's the point of the article....

Re:Ummm....huh?

[affliction]

Scan Line Interleave. Each card renders half of the screen for a given frame.

quad-card cash-vacuum

[Tumbleweed]

Stay tuned for the new quad-card cash-vacuum, coming soon.

Interestingly, that might even work. According to the tests I saw (Anandtech or TechReport, can't remember), the PCIe videocards are only using about 4x of the available 16x anyway, so even with dual cards, they're only using half of the available PCIe lanes, so if they can figure out how to do it, quad cards _could_ work, in theory.

Not that you'd find enough suckers with enough money to make it worthwile, I bet. :)

I just wish my recently-purchased 5900XT wasn't so bad at DirectX 9. I only (currently) play an OpenGL came (BZFlag!), anyway, but I'd like to have the option of playing DX9 games at reasonable framefrates in the future. I guess a 6600GT is in my future, somewhere.

Re:quad-card cash-vacuum

[afidel]

I haven't seen any boards with more than two x4+ slots yet, are there any? If only there was some way to split the x16 slot, since according to this article not even an ATI X800 really stresses a x4 slot.

Re:quad-card cash-vacuum

[Tumbleweed]

The problem is that there aren't going to be any chipsets with any more than 20 PCIe lanes, so once you split an x16 off from that, that leaves only 4 left. If you have one x4 slot and 1 x16, that's it, you're done.

I don't know if the x16 graphics slot is handled differently than other PCIe slots. If so, you'd have to manage that somehow, as the upcoming SLI mobos have done. Really, I'd rather see chipsets with more than 20 PCIe lanes. I'd like a mobo with an x16 for graphics, say 2 x4 slots, and 2 1x slots. That adds up to 26 lanes, not counting bandwidth for anything else like onboard LAN, audio, etc.

Maybe nForce5? :)

Hercules?

[michaelmalak]

And here I thought the story was about once again running a debugger on a Hercules Monographics card while the app being debugged runs on the color card.

Re:Hercules?

[AndroidCat]

Me too. And we liked it too! (It sure beat running debug traces out the serial port to a terminal.)

Re:Hercules?

[Anonymous Coward]

I do that in a way, but instead of a hercules card I use an ati tv tuner card that I cant get working (but makes a fine 1024x768 head for my crappy 17" monitor) which can have a full screen copy of windbg, ollydebugger, or anything else while a full screen app sits on my 19" right next to it served by my fx5600
Also good for custom made cheats that just sniff traffic and display info on my left monitor, which wont show up in any form of detection (unlike overlays).

-irc.goatse.cxtroll

Buy the second a year later

[PIPBoy3000]

The real benefit, from my perspective is that it's a low-cost way to upgrade your video card in between new computers. I bought my first Voodoo 2 for $300. My second cost $30.

Re:Buy the second a year later

[tygerstripes]

...and, of course, the convenience of being able to fry your breakfast on the case!

Re:Buy the second a year later

[LoudMusic]

The real benefit, from my perspective is that it's a low-cost way to upgrade your video card in between new computers. I bought my first Voodoo 2 for $300. My second cost $30.

A friend of mine did the same with CPUs back in the day. He bought a dual Pentium 2 board with one processor - together it ran about $400. Then a year or so later he bought two matched processors for less than $80 (combined) and gave the old one to his brother. We're still using the dual P2 system as a file server.

Re:Buy the second a year later

[dnoyeb]

This only works if the feature sets / API don't change. But video cards today always try to show what NEW technology they are implementing like vertex shaders and whatnot.

So 1 year later you can add more brute force, but chances are that you will be behind the technology curve.

They didn't work before. Nothing has changed since. They wont work now.

Dual cards are just a way for the card sellers to make more money. We see them now ONLY because its easy to do on the PCI Express bus.

What I would like form a dual card setup is this. For me to buy a a MB with built in video, and add an external card, and still be able to use both. I would love that, but otherwise I wont be doing dual.

Re:Buy the second a year later

[space_jake]

Not as easy as the voodoo2 days when 3d accelerators were add-in cards that worked in PCI slots. Now its not so easy to just buy a 2nd one and toss it in your box. You need to plan ahead and buy a motherboard with 2 pci-e slots on it. Not too bad if you're looking to build a computer now, sucks ass if you built one in July for Doom 3. *shakes fist*

Re:Buy the second a year later

[Anonymous Coward]

ATI's integrated solutions support this. You end up with a tri-head system. Unless you happen to want a tri-head system I cant actually see the point. The video card you buy will provide better performance and not impact on your main memory bandwidth. Before most cards came as dual head I thought it would have been cool. But times have changed. Of course, could still be handy in a slim case using low profile cards.

quadro

[Anonymous Coward]

High-end Quadros have 512 MB RAM. plus, they're dirt cheap ;)

Re:quadro

[accelleron]

http://www.ntsi.com/order/video.asp "Call for pricing" I don't know your definition of dirt cheap, but I sure could use your dayjob...

6073

[Anonymous Coward]

dollars....

Tom's Hardware also has a test

[Anonymous Coward]

Tom's Hardware also did review the SLI setup.

Already Slashdotted...

[PrintedChickenQuack]

Good work team!

32x

[Ann Coulter]

The PCI Express standard allows for 32x lanes. The nVidia SLI uses two 8x lanes. Wouldn't it be nice if a motherboard supported two (or more) 32x lanes and 32x graphics cards working in parallel? Think ray tracing because at those bandwidths, and the fact that there is a ergonomic limit on how small a pixel on a display can be, one can have the average size of a triangle be smaller than a pixel. This isn't true ray tracing but the effect is there.

On a similar note, are GPUs a good platform for genuine ray tracing?

Re:32x

[harrkev]

Ummmm... Ray tracing does NOT depend on the video card. If all you are doing is ray tracing, get an old Voodoo 2 or something for $10 from eBay.

Ray Tracing uses the CPU to do all of the work. Video chips are optimized to do a lot of "shortcuts" and "tricks" to render a scene, and the math is completely different. Trying to make them do something else is like trying to strap fins on a donkey and turn it into a fish.

A dual-core CPU, on the other hand, would work wonders on a ray tracing.

Re:32x

[amorsen]

Wait and see. Sometime soon I bet a graphics card with "64-bit support" or something will be announced, and it will turn out to be able to work with double precision floats. High-end GPU's are generic vector FPU engines hidden between fancy drivers anyway, they just tend to be stuck in fixed-point or 32-bit floating point.

Re:32x

[Ann Coulter]

Uhhhhh, the Quadro 4000 and 4400 boards have "128-bit support". The Geforce 6800 has "64-bit support". With the exception of the Quadro 4400 boards, they are already available. Read the specifications on them.

Re:32x

[afidel]

As another poster pointed out the current generation of consumer cards are 64bit already. The Quadro line is 128bit, which is what you'd really want for raytracing. Btw with the C like shader languages already around it should be relativly easy to write a raytracer for modern (DX9 level) cards.

Re:32x

[leathered]

The difference between AGP cards running at 2x and 4x is small. The difference between 4x and 8x is practically non-existant. So I seriously doubt that extra PCIe bandwidth would increase performance. Also note that SLI'd cards talk to each other via the SLI cable.

Re:32x

[Anonymous Coward]

"So I seriously doubt that extra PCIe bandwidth would increase performance"

I think I speak for everyone when I say:

HAHAHAHAHAHA Point and laugh....just point and laugh..

STORK!

Re:32x

[renoX]

Depends on what you call 64bit: is-it per component? Or is-it the RGBA who fits on a 64bit word?

I believe that currently it is the latter and what we need for scientific computation/ray tracing on videocard is the former.

Re:32x

[amorsen]

And as it was replied to the other poster, you want 64-bit per component. No card does that, as far as I know.

Power consumption

[Hoi Polloi]

And we're not even speaking of how much power (wattage) these 'dual solutions' consume...

SLI power consumption can be significant!

Re:Power consumption

[Tim C]

According to those figures, two 6800 Ultras in SLI consume 35% more power than a single 6800 Ultra, when the system is under load.

That's a lot, but still a damn sight better than double.

Re:Power consumption

[gordyf]

That's comparing the entire system's loads, though, and not just the video cards themselves. That means that adding another video card and changing nothing else raised their entire computer's power usage by 35%. That's a fair amount.

GPGPU

[ryanmfw]

This is actually a very interesting possibility for general purpose GPU programming, which aims to offload as much easily parallelizable operations off to the video card. If you can have two, running off of PCIe, you could get a big return in speed, allowing some very cool stuff to be done much quicker.

Check out http://www.gpgpu.org/ for cool stuff. And if I'm not mistaken, it is already possible to use SLI.

Cheers,

There's something I don't understand here.

[Anonymous Coward]

According to your sig, you should be smoking way too much pot to know that.

What about the nforce4?

[lrwx]

Other than the fact that this is old news. I would have figured that the focus would be more on the new nforce4 chipset http://www.nvidia.com/page/nforce4_family.html familiy. There are three board types in this family The Nforce4 Standard, the Nforce4 Ultra, and the Nforce SLI. As a matter of fact Asus is releaseing an sli board based on this right now called the A8N-SLI with a slew of added features that you could expect out of and asus board including dual gigabit ethernet ports! Why the via board is even being covered is beyond me the nforce is a much more better chipset. Here is a [H]ardOCP benchmark page here http://www.hardocp.com/article.html?art=Njk2. Enjoy. ;)

May the GeForce be with you!

[asliarun]

nVidia some how has managed to lock out other manufacturers chipsets from working properly with SLI
A case of nVidia acting on the SLI?

VIA SLI pictures Houston AMD Tech Tour October

[ruiner5000]

We posted pictures here and here of the VIA SLI last month from AMD's tech tour in Houston. More interesting is our pics of the Tyan dual nForce 4 chipset board. That is two nForce 4 chipsets, two full 16X PCI Express slots, and two CPU sockets for Opteron.

SLI is a rip off.

[Anonymous Coward]

I'll wait for the dual GPU on a single card solution. You gain nothing from having 2 cards, the dual PCI express boards still have the same bandwidth the lanes are just split between the two.

This simply forces you to get a new motherboard. Which I guess is a win for intel and nvidia eh?

Let's see, get dual cards which requires a new motherboards, or wait and get a new video card that has gual GPU"s which takes about 10 minutes to install at most.

I bet you ATI will do the dual GPU solution first and nvidia will go "fuck we should have learned from 3dFX's voodoo 5500"

I had a 5000 series card, dual Gpu's on the SAME card amazing concept!

The dual voodoo cards made sense in a day when you had a lot of spare pci slots. But ever since we've gone to the methodolgy of a single graphic slot it's not simply a matter of slapping in a new video card and connecting an sli connector, you have to get a whole new motherboard.

I DO agree with a previous statement made that is if we could go up to 4 cards and 4 cpu's on a system. that kind of flexibility would be awesome.

Re:SLI is a rip off.

[SorcererX]

PCI-E isn't just there for the SLI, it is there to finally replace AGP & PCI. PCI-E has less connectors and is therefore cheaper to manufacture. PCI-E 1x is much faster than PCI. It's all good. I didn't see anyone complain when we phased out ISA for PCI, how is this different?

SLI confuses me.

[Wescotte]

Doesn't SLI work by rendering basically every other line (or half the screen) for each card and then combining them? Is there any work being done with more complex situations?

Like ok card 1 you render this building, card 2 you render this tree. Ok done card 1? Now go ahead and render that scary monster! Card 2 get your ass in gear and finish rendering that tree! Now combining all these completed elements would be much more complex than just merging pixel data but I think for significantly more complex geometry it would be the way to go.

We are getting better 3D enviroments with alot more detail but it seems like we're just hacking our way thru. For example (and I forget what the actual term is) but I saw a tech demo for the Unreal 2 engine where they take original high poly models and create a special texture that allows the lighting to produce the acurate depth. Sure you get amazing visuals but other things suffer. Like accurate collision detection or just a ball bouncing off the wall. It's simply not mathmatically possible to accurately simulate how the ball would react in these situations.

Re:SLI confuses me.

[gzunk]

Scan Line Interleave works the way you describe, by interleaving the scan lines. This is the way the Voodoo 2 did it.

Scalable Link Interface, on the other hand, gives the top half of the image to be rendered by one card, and the bottom half to the other half.

Each card shares the ulimate framebuffer (to place the results into), and each card has a complete copy of the texture / model so that it can perform the rendering.

Splitting the work according to the 3D model would be a complete nightmare to write drivers for, because of the variety of the 3D models involved. Much simpler to just split it later in the 3D pipeline when it gets rendered into the 2D framebuffer.

Re:SLI confuses me.

[Anonymous Coward]

The main reason the "new" SLI dosent render alternate lines is pretty simple, it would break early Z rejection and some other block orientated bandwidth saving technologies.

Spliting the screen in two halfs works neatly with early Z and may (statisticaly speaking) improve the on-chip cache hit ratio for texel lookups.

Rendering alternate objects as you proposed wont work because you would have to merge the color buffer and the z buffer of both cards. This means you would have to copy the entire color and z buffer from one card to the other and "blend" them. As long as you only use opaque stuff it would work but as soon as you add transparency into the mix you run into serious problems because the order in which stuff is blended into the colorbuffer is important and for some blending modes you disable z buffer writes all together.

The "new" SLI simply does a bitblt for the lower half screen and is done (for the sake of simpliciticy I wont go into FSAA).

Sorry, for the pisspoor English. I'm not a native speaker plus my mind is in an altered state :D

Hope this helps,
Cheers!

Re:SLI confuses me.

[Maagma]

On the same point the Scalable Link Interface actualy is not a direct 50/50 ratio of top and bottom. It's more like 60/40 since sky is taking up the top half of your screen in most games and does not take much to render compared to a complex landscape.

Re:SLI confuses me.

[Anonymous Coward]

I believe the driver can change the ratio depending on the game running. In Doom3 I am pretty sure it's 50:50.

Re:SLI confuses me.

[lendude]

Actually no - as the article says, DOOM3 is profiled in the driver set used in the testing as an AFR (alternate frame rendering) user, hence screen splitting isn't applicable to it.

6 months ago....

[Tongue In A Box]

Did I hit the Slashdot archeological nostalgia site? This was news 6 months ago.

Modern CPU's cannot handle this...

[Omniscientist]

One NVIDIA 6800 GT bottlenecks by most CPU's out today. The GPU has 222 million transistors, more than most CPU's available. In fact I'm not even aware of any CPU that exceeds that (I'm talking about home use processors).

If you get two 6800 GT's working together, well if one GT is bottlenecked from most CPU's (the GPU has to actually wait a little bit more for the CPU to catch up), how can that CPU possibly catch up to two?

I say that we should wait to buy SLI technology until better CPU's come out, or if you have a dual CPU setup, or even until dual core CPU's come out.
Well, that sounds expensive to me, better start saving...

Re:Modern CPU's cannot handle this...

[DaPhoenix]

Its not about a CPU bottleneck. Its about being able to run pixel and vertex shaders on the card faster and more efficiently thereby reducing the load on the CPU.

Old article on human eyes

[Anonymous Coward]

For some reason what you said made me look up this article, it's 2-3 years old but good read about the misconceptions about human eye fps.
http://amo.net/NT/02-21-01FPS.html

On framerates...

[temojen]

Why do you want more than 30fps? At 30 fps the motion is smooth, and there is no interference with the light flicker (on North American 60Hz AC, Incandescent & CF Lights). Any extra GPU cycles can go to improving the quality of those 30 frames.

Re:On framerates...

[molotov303]

Why do you want more than 30fps?

The benchmarks quoted are average framerates, but what you really care about is the minimum. When you have 4 enemies on the screen and are running and shooting wildly is the time you most care about framerate, and it's also usually the time when your card is the most strained.

Re:On framerates...

[airjrdn]

30 is an extreme minimum. Actually, you want it to match your refresh rate (or be higher) at all times. That way you can enable Vsync and get rid of screen tearing.

Also, just because you get 30fps staring a wall, doesn't mean you'll get that when there are 20 rockets, 10 explosions, and 12 players on the screen at one time.

decent explanation here

Re:On framerates...

[Woody77]

Incandescants don't flicker at 60Hz, they take too long to heat up. CF and cheap flourescents, however, will flicker (drives me up the wall).

Re:On framerates...

[king-manic]

Like always, 30 fps isn't the key. You might see a smooth movie at 30 or even 25 fps but for interative 3d enviroments 60 is better, and for some really really anal retentive visual processing people, no number is too high.

Re:On framerates...

[fireduck]

also keep in mind 30 fps walking down a hallway in Half-Life 2 could easily turn into 5 fps the minute you enter the outdoors and 8 marines start lobbing grenades at you, if the game wasn't programmed to compensate...

Re:On framerates...

[accelleron]

It seems to me we need a solution that will detect framerate and adjust detail automatically.
This way, when we've the desire to gawk at some doodad in the game world for three minutes at a time, we can enjoy it in full detail, but when you're being bumrushed by five beasties, your first reaction isn't to bask in the per-pixel lit glory. That is when the engine can crank down the detail and turn up the FPS (and potentially the amount of carbohydrates being pumped into your bloodstream.)

Re:On framerates...

[blaine]

There's a couple good reasons you need more than 30FPS. For one thing, you never want the framerate to drop below smooth. So, even if you can average 30FPS, that doesn't mean you aren't going to sometimes drop to 15FPS or lower.

However, there's a much more important factor at work here that confounds the film-vs-video-card comparison: video game frames are not the same as film frames. The biggest problem in this regard is motion blur. Here's a little exercise. Try it out in real life if you have the equipment, or just think along through it:

Let's say you were to use a video camera and capture 30 frames in 1 second. The subject is your own hand, waving up and down quickly.

Now let's say you rendered a 1 second video using the 3D engine du jour, also 30 frames, of a hand waving up and down quickly.

If you were to look at the 30 film frames, they would not be crisp. Each one of them would likely exhibit motion blur. However, when played at a rate of 30fps, to the human eye, that motion blur looks smooth.

If you were to look at the 30 rendered frames, there is no motion blur. Each frame is rendered crisply. The problem with this is, when played at 30fps, instead of smoothly moving from one frame to the next, the hand appears to jump between frames. There is no intermediate data to allow a smooth flow from frame to frame.

There are two ways around this: first, you could simulate motion blur in the engine. Second, you can pump the FPS up high enough that there is intermediate data for your eye to take in, and do the motion blur on it's own. The former of these options seems much more likely.

Re:On framerates...

[scribblej]

The former of these options seems much more likely.
---

Nitpick? I think you meant the latter. I can think of only a few games which have tried to emulate motion blue (Grand Theft Auto being the biggest one that comes to mind) and it's so annoying it's the first option to turn off.

Re:On framerates...

[scribblej]

Why must you match your refresh rate or higher to enable vsync? When I was into all this stuff, which admittedly was around the time Wolfenstein 3D came out, refreshing during the vsync was the only way to avoid nasty page tearing. Back then, I was lucky to pull 12fps, and you can bet that page tearing looked horrible.

On the contrary, I'd say just the opposite; if you can crank your framerate high enough, you can *turn off* vsync.

But then I'm outdated. Why is it the opposite of what I'd think? at 75fps and 75 cycles of refresh per second, a page tear will hardly be visible, it'll be on the screen fo nearly no time at all -- and the difference between the rendered frame pre and post tear will be minimal.

Since I've gotten a "hot shit" video card, I've turned off page tearing. Before, I always had to keep it on.

Re:On framerates...

[airjrdn]

Honestly, I probably couldn't explain it well enough for you to understand it, because I'm not sure I fully understand it myself.

I found this which does an ok job of explaining it: http://www.d-silence.com/feature.php?id=255

I do play a lot of Unreal Tournament 2004 though (about 3 hours a night) and for whatever reason, I seem to aim better with hitscan weapons (instant hit, unlike rockets, etc.) such as the lightning gun and shock rifle with vsync off than I do with it on. I'm not sure why that is, but it became pretty apparent when I turned it on for a few days.

Since I play with vsync off, I do get tearing from time to time, but I'd rather deal with a little tearing here and there than getting blown to bits 5x more often.

Offtopic....

[Anonymous Coward]

SLI Stands for 2 things: Scan Line Interleave and Scalable Link Interface. The first was available on the Voodoo series of graphics cards. This should not be confused with the latter, which is an entirly different concept (and acronym), and is used on the Nvidia line of graphics cards.

Interesting upgrade options

[asliarun]

SLI offers some interesting upgrade paths. Say you plop $200 for a 6600GT today. Maybe, you can squeeze enough performance out of this baby to get decent frame rates for the current and next season's FPS games. However, going by past trends, i really doubt that you would be able to get adequate performance 2 years down the line. On top of it, your 6600GT would probably fetch you less than a hundred bucks if you do decide to upgrade it.

However, say you decide to buy a relatively cheap-ass motherboard/processor today that does NOT support SLI but fit it with a 6600GT. Note that for a decent gaming system that offers you a good price/performance, you're spending more on your video card than your processor and motherboard combined. 2 years down the line, simply upgrade your motherboard to an SLI enabled one (which would have become really cheap by then), buy yourself another 6600GT from eBay for a hundred bucks or less, and hopefully, you should be able to play FarCry2 by then!

This make any sense? Basically, my rationale here is on 2 points:-
1. I dont really care about my CPU/motherboard as it's way cheaper than my video card
2. With SLI, i can simply buy today's high-end card at tomorrow's prices instead of throwing a perfectly good but outdated video card. This way, i can hold off a major upgrade for another year or two.

Re:Interesting upgrade options

[Anonymous Coward]

The downside of SLI is that you need two identical cards (same vendor, same bios version).

Chances to get the same version of your card in two years are slim to none.

Some clever hackers may circumvent this limitation but I would not bet on it.

I personly dont think that SLI was "invented" to give you an cheap upgrade path in two years, last time I checked Nvidia was still a revenue based company.

Re:Interesting upgrade options

[Anonymous Coward]

The downside of SLI is that you need to identical cards (vendor and bios version)

Chances to get 2nd card that is identical to your old one are slim to none.

Some clever hackers may find a way to circumvent this limitation but I would not bet on it.

I personaly dont think that SLI was "invented" to give you a cheap upgrade path, last time I checked Nvidia was still a revenue driven company.

Just what I needed

[BollocksToThis]

I was sitting around yesterday wondering to myself "How can I make the inside of my computer hotter than the fires of hell?".

Re:Just what I needed

[obeythefist]

I have a plain 6800 with a massive heatsink on it, and whew, it generates a lot of heat in my case.

Sweet card though. It's a shame that ATI didn't have anything to compare with at the time and the price.

Of course now that I've bought it, the 6600GT and the X700's are starting to hit the market at lower price points. Which is always the way, isn't it?

At least I didn't buy a pair of 6800GT's for SLI just as the 6900GT comes out... which is bound to happen. Right? Sigh. They saw me coming.

I Feel Another Commercial Coming On

[Greyfox]

3.4 GhZ P4 with a gig of RAM: $3000
2 nvidia SLI cards: $600
Getting 4 FPS anyway because 40,000 people are on the same server as you: Priceless.

Re:I Feel Another Commercial Coming On

[Anonymous Coward]

"3.4 GhZ P4 with a gig of RAM: $3000"

Dude you got screwed!

Re:I Feel Another Commercial Coming On

[obeythefist]

But... who buys P4 for gaming? AMD CPUs cost less and perform better in gaming situations at this point in time. Given the next round of CPUs to compare will be dual core, and intel is lagging a good solid year behind AMD on the technology front...

Read the articles. Look at the prices. Compare the benchmarks. Make the right decision.

Vsync

[Anonymous Coward]

Sounds like you have vsync enabled. Vsync drops frames so that the number of frames per second evenly divides the refresh rate of your monitor. Disabling it will get you back lots of performance.

Re:Vsync

[kin_korn_karn]

Disabling vsync can cause lag problems with USB input devices. If I turn vsync off on Madden 2005 my logitech gamepad picks up about half a second of lag - not exactly good for gaming.

(yeah, I know, the "you should be playing sports games on a console" speech - I'm a franchise mode guy, I just use the gamepad for the training camp drills)

not useful.

[Penguinoflight]

Of all the reviews detailing dual/quad cpu mb's I've seen just as many gaming benchmarks come out slower as faster. The boards cost so much to make because of limited audience, most people couldn't even understand a dual MB if they had a 1 semester class, so lack of knowlegable customers really hurts here.

Possible reason...

[apharov]

Would you happen to be using a motherboard with a VIA chipset? My old MB used a KT400 chipset. I didn't notice anything strange when using a Radeon 7200 on it but when I upgraded to a 9800Pro the speed I got was way slower than what it should have been. A couple of nights on tweaking and googling and I came to the conclusion that KT400 AGP support was s**t, especially with ATI video cards.

One more night of examining other motherboards and I decided to buy a mb based on nForce2Ultra chipset. After installing the new mb my actual FPS's almost doubled. Bying a new mb might seem a bit drastic but considering that it cost "only" ~120e vs. the 250e of the R9800Pro it seemed quite reasonable to me and I haven't regretted it.

Personally I will never use VIA's chipsets again if I have any other choices.

Err you aren't trying hard enough.

[Charcharodon]

You obviously don't have much imagination if you can't think of a use of more than two video cards/monitors.

As a lover of flight sims I'll be first in line to buy a mother board that can support 10 video cards. Along with an array of cheap monitors I will finally have a wrap around view of the sim world. This can apply easily to any game.

First person shooters could finally have peripheral vision (one center and two on the sides) along with a inventory and map screen. Brings the grand total to five.

Driving games could finally have a true perspective instead of the stupid 3rd person or 1/3 screen in car view. So at least three monitors.

RTS resource monitors, sat view, and ground maps. Well that could become quite the array depending on how much you wanted covered. Say anywhere from 3-12 monitors.

Same for Massive Multiplayer Online Games. I could see a use without trying hard that would require at least six monitors.

You could double, tripple or even quadruple up on the number of required cards for any one monitor that would require higher end graphics. There are always those twisted monkeys that come up with graphics that won't run on any one GPU these days. For example those lovely to the horizon maps that show up in various games that add about 100meters of high detail every year. I see another scenario where people boost their systems performance by picking up cheaper versions of cards they own to keep their graphics improving without breaking the bank. (We can all remember when GF 2 cards cost $400 each, that'll buy you 50 of them these days.

Who could afford all this you ask? Well just about anyone these days. I've got a stack of 17inch CRT monitors in the garage I picked up for $5 a piece that are just begging to be used. With the advent of sub $100 video cards and CRT monitors, and the fact that not every output would have to be super hi rez. Perpheral views, 2d maps, and inventory lists would be just fine on something to the equivalent to a GeForce 4 MX ($32 new). You could seriously enhance your gaming machine for the price of one top of the line latest and the greatest video card from ATI/Nvidia.

So you keep your two monitor display, for me I'm going to check to see if the wiring in my computer room can handle the extra 10 monitors I plan on adding.

Re:Err you aren't trying hard enough.

[Anonymous Coward]

First person shooters could finally have peripheral vision (one center and two on the sides) along with a inventory and map screen. Brings the grand total to five.

Matrox Parahelia card could drive up to 3 monitors to do this, with the right game drivers. It didn't take off so well.

So your idea is a solution looking for a big enough userbase.

Of course, they do use multiple video cards to drive new medical LCD displays (those 250dpi ones we hear about every once in awhile), as sold by ViewSonic and IBM.

and the 30" Cinema Display for the Mac uses two video cards...

Re:Err you aren't trying hard enough.

[G-funk]

First person shooters could finally have peripheral vision (one center and two on the sides) along with a inventory and map screen. Brings the grand total to five.

Doom (1.666) did this, but I think you needed three networked pcs.

Re:Err you aren't trying hard enough.

[Anonymous Coward]

D0h! that should be (<1.666)

Re:Err you aren't trying hard enough.

[Nurgled]

Having three monitors to give you "peripheral vision" in a first-person shooter doesn't really work, because when you are looking at the center monitor and you see something move in the left, your natural instinct is to move your head or eyes to look at that screen, not to move your mouse. When you do this, the centre monitor becomes peripheral, the left becomes main vision and the right becomes useless. A few weeks ago I saw a friend playing some game where it deliberately blurred the outer edges of the display to simulate peripheral vision, but it had no way to stop you looking at those parts of the screen with your central vision, and my friend said that it made his eyes hurt after a while.

The only way this would work is if the three monitors were strapped to your head and the game got feedback about your head movements, but lots of people seem to get motion sickness with such arrangements, and you've still got to find some way to handle the discrepency between the view angle and the weapon angle, since in most FPS games these are considered to be the same thing. You'd need some kind of gun-shaped wand to hold in your hand and have the computer detect where it is pointing, or something like that.

Re:Err you aren't trying hard enough.

[Anonymous Coward]

As a lover of flight sims I'll be first in line to buy a mother board that can support 10 video cards. Along with an array of cheap monitors I will finally have a wrap around view of the sim world. This can apply easily to any game.

As a lover of interactive porn I will finally have a way to experience my exhibitionist fantasies.

Re:Err you aren't trying hard enough.

[Palarran]

A redundant array of inexpensive displays, perhaps?

Re:Err you aren't trying hard enough.

[Charcharodon]

That's because the Matrox did nothing more than just stretch the picture over three monitors it didn't actaully give you any more degrees of view than with a single monitor, that and it had very poor stats compaired to other cards (Nvidia/ATI).

Re:Err you aren't trying hard enough.

[broohaha]

You obviously don't have much imagination if you can't think of a use of more than two video cards/monitors.

...and on the business end, dual monitors are becoming more and more common. Especially in the financial trading industry. In fact quad cards are actually just as popular at the company I work, with the four flat screens hoisted up by arms.

Re:Err you aren't trying hard enough.

[dago]

So, if you'll be first in line to have a solution with 10 video cards, I guess you'll jump to buy one of Magma's PCI extension. The 7-ports extension cost just above 1000$, fill it with Radeon9200 (70 each), use the AGP port and 2 PCI slot of your main motherboard and you'll be fine.

If

Re:Err you aren't trying hard enough.

[Slashamatic]

Yep, but the Matrox was cool for 2d and it ran fast enough too. If all the graphics you do are standard graphs then it was easily fast enough.

Re:Err you aren't trying hard enough.

[Slashamatic]

Real flight simulators work a little like this with multiple GPUs to give front and peripheral vision. The peripheral stuff isn't rendered to the same detail.

You only ever 'fly' using the front windscreen so the peripheral stuff doesn't confuse (especially, as the cockpit windows are physically represented between you and the screens.

Re:Err you aren't trying hard enough.

[Nurgled]

A flight sim is one thing which does work very well with multiple monitors; in a real plane, you are fixed in relation to your controls and the windows and you can look around you and see the world from your current POV at any angle your head can turn to so long as there is a window there. In this case, the monitors represent the windows of the plane.

In the FPS case, though, I suppose you could argue that looking at the left monitor is like turning your head to the left without changing your aim or body direction. A 360-degree display (or one of these sphere things) Is really needed here since in this kind of environment in the "real world" you'd have a full field of view; in a plane, you're restricted by where there are windows anyway. FPS with only three monitors would be like walking around with a big black partition attached to your back so you can't see outside a certain view angle, but I suppose that's no different to playing with only one monitor.

You'd still have to get used to rotating your "body" with the mouse when reacting to stimulus, though, and it would probably make lots of people motion-sick because their vision makes it look like the body is turning but the balance and so on suggest that the body is standing still. This isn't normally such an issue since without the ability to "move your head" you don't have to do the head attitude adjustment when you turn your body, and this automatic adjustment must require lots of feedback to keep you locked on to your target while turning, and the turning will generally be very fast, especially in quick-fire FPS games like Q3A and UT.

Re:Err you aren't trying hard enough.

[Slashamatic]

It really depends on how immersive you want to make the experience. As you rightly say, motion sickness can become a serious risk. I don't know about wraparound screens, but I know about the experiments with VR glasses that were done earlier by NASA amongst others and there were a lot of problems (some of which may be solved by faster GPUs).

I still think that a basic three screen arrangement with the left and right screen not showing any dashboard or weapon, then it should be reasonably clear that you can only do something out of the front screen and you should centre before firing.

Re:Err you aren't trying hard enough.

[klmth]

You can, of course, to that to some extent now already.
http://members.chello.nl/~s.ferris/ is a photo of one flightsim enthusiast that has 12 monitors lined up for the purpose. Granted, this takes up a total of 8 computers due to the limitations in card technology.
Some people have also gotten Grand Prix Legends to work with three monitors.

3d modelling is another endeveour which benefits from increased real estate. In addition to that, ATC installation need some kind of overview displays. Since the blips are small, and the area covered is big, projection display simply do not cut the mustard. These days, the deed is done by using either tens of small displays or a couple of absurdly expensive Barco displays, running on a multitude of machines.

When the need arises for a 42" screen real estate with a high resolution, using several computers and several displays to render it is a neglible cost compared to the price of large high-resolution projection displays.

Re:Err you aren't trying hard enough.

[Charcharodon]

true but how many people run just 2d anymore? (Outside the business world)

Re:Err you aren't trying hard enough.

[Slashamatic]

You have a good point, but in the building shere I'm currently working, my client has several hundred dual screen PCs all running straight 2D and within a mile or so, you can multiply that many, many times.

Dual mania!

[freeze128]

but I can't see myself using more than 2 video cards.

That's because you need to use Dual Webcams!

Slots?

[HexRei]

So, are you limited to PCI, are do you need to have two AGP ports to do this?

NO NO NOOOOO!!!!!

[Dac Vin]

That doesn't do it! Why the hell would i want two nvidia when one do me good? Give me dual cards, but this time, ONE nvidia and ONE ATI - The best of both worlds, damnit!

According to my nvidia tech rep,

[Anonymous Coward]

this isn't really for squeezing extra $$ out of consumers, since it's way beyond the price reservation point for even most high-end users.

The real point was that for a game developer with a 2-3 year developent cycle, a SLI combination of 2 top-of-the-line cards today roughly approximates the speed (in raw # crunching) of a mainstream card from the next generation of chipsets. So you can get a meaningful approximation of the performance capabiblties for your state-of-the-art-pushing-in-3-years 3D engine today, instead of just guessing whether or not you're optimally exploiting the future "modern" hardware at the time you release your game.

PCI limits

[Anonymous Coward]

I'm sure many Slashdot readers fondly remember the era of 3dfx. SLI'd Voodoo 2's were a force to reckoned with.

Except when texture sizes were too detailed for the onboard memory to handle (consider that the memory on each Voodoo2 had to store the same textures, thereby the texture memory did not effectively double). In this case, a single Voodoo2 or SLI'd Voodoo2's become dependent on the PCI bus.

As texture sizes went up, the performance would very suddenly drop. Drop so badly, that even a Matrox G200 could be seen to run very much faster than an SLI Voodoo2 setup, given high resolution textures.

3Dfx were wrong to ignore the benefits of AGP for so long. Even their first AGP card was working in PCI mode, not getting benefit from AGP performance. Where are they now?

Two video cards = 2 x resolution?

[UnknowingFool]

Does this mean that if I spend $600, my pr0n will be twice as detailed?

Give it to me straight.

[Legato895]

what are the chances that apple will ever utilize this in any way? im going into 3d animation and modeling and this could help me immensely. ill be getting my college computer (a top of the line g5) in a half a year, and knowing that something like this was assured would cause me to wait for it. couldn't you do this using pci express? and will apple abandon pci x? i don't want to have people flaming me or whatever, i love apple, and my college has macs at 80%, there also giving me maya to do homework on, so just building my own pc is out of the question.

Re:Give it to me straight.

[tolldog]

My animation you do in college doesn't need top of the line computers. Just animate rough models and swap in hero models for rendering. Unless you are doing complicated simulations or need to work in highly detailed scenes, you can get by animiating on some relativley average hardware.

The real bottleneck usually comes in rendering.

Re:Give it to me straight.

[Legato895]

thanks a bunch, im pretty noobish at all this still. i tend to always want the best i can get w/o even thinking of an alternative

Italians Pulled it off First!

[StarWreck]

The Italians were the first to pull off playing Doom 3 on a Voodoo card! Although I can't tell if the card they used was SLI'd, the screen shots are a good deal better. Here's the link along with the screen shots: Forum Zone

"Return"? When did they go away?

[LoudMusic]

Using two cards to generate one image for one monitor may be a return, but I've been using three video cards in my computer for a while.

Eh, sorry, just being nit-picky about the post title.

Doom3 on Voodoo 2 SLI

[Naito]

I dare say it looks better on the Voodoo2 setup than it does on the DX9 cards. I mean, I can actually SEE!

Dual-core is the future, not dual video cards!

[wikinerd]

I believe that video card manufacturers should embrace dual-core GPU designs. I don't think that dual video cards will ever become mainstream and will always remain an expensive high-end solution. A dual-core GPU, on the other hand, may even turn all single-core designs obsolete. Read more in my blog.

Re:Dual-core is the future, not dual video cards!

[Intocabile]

They're already dual core, you could even say they're quad core or octal core etc. Nvidia and ATI both use multiple "core" GPUs to cut down the cost on developement and increase yeilds of highend cards.

The 6800 Ultra has 16 pixel pipelines and 6 vertex shaders, the 6800 non-Ultra has 12 pixel pipelines and 5 vertex shaders. The only difference between their chip is that the non-Ultra has 4 pixel pipelines and one vertex shader disabled through hardware. These disabled "cores" can even be reenabled through software but they are often broken, hence the reason they were disabled and in the lowerend card in the first place.

Re:Dual-core is the future, not dual video cards!

[wikinerd]

I don't think a pipeline can be described as a processing core. Pipelines are features of cores and a dual-core GPU would have 2x16 pipelines.

Just another gimmick...

[pandrijeczko]

...to get you to part with your hard-earned money.

How about all those lazy PC games programmers actually stick with a single game engine for a while and spend time optimising that rather than recreating the wheel for every new game.

What happened to all the slick programmers that were able to push "fixed" machines like the Amiga and Commodore 64 beyond their capabilities through neat programming tricks and constant code improvements?

I've really had it now with constant upgrades and with hardware that is never tried and tested long enough to see what can be done with it.

Re:Just another gimmick...

[cr_nucleus]

Maybe you should try getting a look on the console side. That's for the fixed platform and optimising issue.

PC programmers simply can't do the same since the platform is nothing near fixed. This is strange since you seem to be aware of this issue, i don't understand what you're bitching about.

This is why I stopped following Slashdot regularly

[mightyfoo]

We're just now talking about SLI? SLI has been available for the past 6 months from a number of motherboard manufacturers, matter of fact this is in the technology department, 'ancient history'.

Slashdot is much better at discussing what color socks ESR is wearing than what technology is 'hot'.

AlienWare Sucks

[Anonymous Coward]

$60 for a power cable
$29 for a 10 foot net cable

Multimonitor gaming

[neuroking]

I think this I the first step toward truely usable multimonitor gaming. Most cards have a hard time reaching 1600x1200 with nearly full detail, and stretching the screen to cover 3200x1200 means a severe reduction in quality. On top of that it allows for 2 DVIs per card, for the average consumer, and 3+ for others.

Personally, I would love to see two vpus working in tandem to render 3 monitors. I just got 2 Dell 2001FPs, and FPSes can only use one (if for no ther reason that with two, you have your weapons and crosshair lingering in limbo-esque 'bevel land'). The only logical step up is to go to 3 monitors, but there's nothing game worthy with 4 DVI connections on 1 card.

Moore's Law limits?

[bs_02_06_02]

Is it just me, or are there a few people out there that simply wanna claim, "See, I told you so? Remember, back in Nov. '04, I pointed out the end of Moore's Law. See! I was the one that figured it out."

And what happens? It keeps going, and going. When we do hit the end, you'll hear about it, but it's not going to be on /. first.
The issue with dual video is bandwidth. It's not microprocessor speed. The bus can't keep up. Bottlenecks exist.

Correction on doom version history.

[zokum]

Actually, it was removed in 1.2 i think. It's a feature that was only present in versions around 0.98-0.99 to 1.1 or so. Lee Killough's page at http://www.rome.ro/lee_killough/versions/index.htm l even has a way to get 1.1 doom. I've been thinking about doing some dosbox-experimenting with this version. Imagine plaing 3-monitor doom!

I wonder, is doom the first example of a multi-monitor game. Has to be the first fps to have such a feature. (yes, i know, we're talking about 2 extra networked pcs used as dumb terminals, but at the time, noone had the cpu for rendering the gameworld three times.)

Your example makes sense, but it sucks

[Ohreally_factor]

Unless you're shooting your video in frame or progressive mode, all* home video cameras are going to give you 2 interlaced fields for every frame. A better practical experiment would be film, where you can actually see the motion blur without needing to de-interlace. However, motion blur can be controlled by the shutter angle on film cameras. (It's the equivalent of shutter speed on a still film camera.) Variable shutter angle is a feature on high speed film cameras.

Motion blur makes motion look smoother. Lack of motion blur looks jerky, unless you compensate with a high frame rate.

Anyway, you're on the right track in how you're thinking about this. I wish I could explain better, but I'm really sleepy at the moment, and I'm not at my most coherent.

*not sure about hdv

Ultra cards and LCD

[Anonymous Coward]

If you have an LCD monitor(s) then there is no point getting a top of the range graphics card. The 6800 and X800 are very fast and really shine in ultra high resolutions 1600x1200 etc, but if the native res of your LCD is 1024x768 then there is no point getting them. The fill rates of the mid-range cards will provide 4xAA 8xAF etc and still be fine at your native res and your cards will look great at on.
A 9800pro is still quick for these resolutions while the new 6600GT agp is about the same price but even faster and some sweet Dx9c tricks.

Uhhhh, no.

[Anonymous Coward]

(Every graphics-specialized CS collectively groans in unison.)

Modern graphics cards will never be good at raytracing for a very simple conceptual reason. Raytracers and polygon* renderers are opposites of each other. Raytracers take a bunch of rays and intersect them with a scene to produce an image. Polygon renderers take a bunch of polygons (a scene) and intersect them with an image plane. Fundamentally different. One is feed-forward, the other goes backwards.

* - I say polygons, but really any primitive. Polygons are popular because they are easy, both in math and modeling. But the same holds true for voxel rendering, or even a system that splats spheres or complex CSG.

And the polygon renderers became popular to accelerate in hardware because the method scales so nicely. One can throw an infinite amount of geometry at it (streaming over the bus) and it just keeps going at the same speed, building up an image. However, as the size of the scene increases for a raytracer, the working set continues to grow because the entire scene must be accessible for every ray. One can stream a polygon for every brick in the Great Wall of China to an 8 meg video card, and it'll work fine. But try raytracing that Great Wall. Does your machine have a terabyte of volatile memory?