Support NeoGAF

[Mrbob]

This isn't the overpriced Alienware package either!!!

http://www.hardwareanalysis.com/content/article/1728/

On early driver revisions which only offered non-optimized dynamic load-balancing algorithms their SLI configuration performed 77% faster than a single graphics card. However Nvidia has told us that prospective performance numbers should show a performance increase closer to 90% over that of a single graphics card. There are a few things that need to be taken into account however when you're considering buying an SLI configuration. First off you'll need a workstation motherboard featuring two PCI-E-x16 slots which will also use the more expensive Intel Xeon processors. Secondly you'll need two identical, same brand and type, PCI-E GeForce 6800 graphics cards. For workstation users it is also a nice extra that with a SLI configuration a total of four monitors can be driven off of the respective DVI outputs on the graphics cards, a feature we'll undoubtedly see pitched as a major feature for the Quadro version of the GeForce 6800 series SLI configuration.

Once 3rd party board manufacturers get PCI-E boards out this should be a viable option! SLI IS BACK.

 

[golem]

omg...

 

[GigaDrive]

no fucking way

 

[Mrbob]

I wonder how much heat dual 6800Us on heatsink/fan cooling put out.

 

[GigaDrive]

btw, I have no doubt that ATI will respond.


there were cards from Evans & Sutherland with dual R200s / Radeon 8500s
as well as E&S cards with dual and quad R300s / Radeon 9700. while that might not exactly be SLI, it's the same result, if not better.

 

[Norse]

I can see it now....you power up your pc and the lights in your house dim. How big will the power supply be?

Any way, they had benchmarks for a system like this a month or so ago. Alienware did it. bout 47% increase in benchmark scores with second video card installed.

This would be great if you could buy 2 75 dollar cards and get the performance of a single 6800. hehe

 

[Panajev2001a]

Do you think that if I use this

as my PSU that I should be quite prepared for such configurations ?

 

[TekunoRobby]

Ha ha ha ha awesome. I can't wait for this (I'm a dual proc/dual chip whore).

 

[Discharger]

SLI rocks

Of course with the price of todays video cards, it's only going to be a viable option for people with tons of money to blow, but it's nice that the option is there. Give it a year and a half or so after this is on the market, and pretty soon the whole range will support it. At that point, that's when its going to come into its own for the most of us.

 

[GigaDrive]

hey Voodoo2 SLI was like $600... or maybe $500 cuz I think 3Dfx offered a deal if purchasing two Voodoo2 cards.

now you're gonna have to shell out $1000~$1200 for two GeForce 6800s. I dunno if SLI is for only the Ultra cards or what but we a talking 1000 to 1200 bucks.. for a 77 to 90 percent performance increase over one card.

 

[McFly]

Do you think that if I use this as my PSU that I should be quite prepared for such configurations ?

No, you need a third for the CPU.

Fredi

 

[hooo]

No, you need a third for the CPU.

Fredi

And I thought that was just for cooling...

 

[SyNapSe]

SOOOooooO not mainstream. wtf, would some of you actually purchase a machine this expensive just to play games!?

Hrm, well I guess people do buy those Alienwares and they get pretty damn expensive.

 

[Bregor]

Nice, but it doesn't explain how the half images from the two cards are recombined for display on a single screen.

 

[Mrbob]

OOOooooO not mainstream. wtf, would some of you actually purchase a machine this expensive just to play games!?

Hrm, well I guess people do buy those Alienwares and they get pretty damn expensive.

Well that is right now because there is hardly anything out on the market which offers 2 X16PCI-E slots. Wait about six months when PCI-E boards become more readily available.

I found a Tomshardware article that goes into more detail:

http://www20.graphics.tomshardware.com/graphic/20040628/index.html

In the article he makes it sound like Nvidia is going to offer this on all 6800 solutions, which means the 6800, 6800GT, and 6800U! In fact it sounds like you can mix and match card manufacturers together, but Nvidia recommends you stay with the same brand (I.E. You buy an Evga 6800GT you should buy another Evga 6800GT).

Bregor, the article answers your question about how the two cards display the picture.

This is all very interesting news. Once PCI-E becomes more viable and gains more support, prices on everything will drop. It does beg the question though: Are two 6800NU linked together better than one Geforce 6800Ultra? The price difference right now for 2 6800NU are $299 a piece versus $499 for the GF6800Ultra. Well, I guess it isn't a viable solution right now as these PCI-E cards won't be available until September. Maybe later.

But I'm thinking in about a year, year and a half, when the 6800GT price drop around $200 bucks. I wouldn't mind dropping $400 on a 6800GT SLI style setup!

 

[Bregor]

Bregor, the article answers your question about how the two cards display the picture.

I just read it a second time and could not find the explanation. They explain how the cards are linked so that they can divide the screen up for rendering, but not how the resulting frames are re-assembled into a single picture.

Not that I doubt them ... I'm just curious. The Alienware solution fed the analog output from the two cards into a special combiner.

There has been some speculation on other forums that Alienware (who supposedly has patents on some of their tech) and Nvidia cooperated to develop this. It's largely based upon this statement:

Exact performance figures are not yet available, but Nvidia's SLI concept has already been shown behind closed doors by one of the companies working with Nvidia on the SLI implementation.

[H]ardOCP has their article up now, and Anandtech will have one soon as well.

 

[Mrbob]

Hmm, I dunno. Perhaps it is patented technology Nvidia doesn't want to fully reveal?

HardOCP preview up!

http://www.hardocp.com/article.html?art=NjMz

Now in 2004 NVIDIA has introduced their new SLI technology, but note that it does not stand for Scan Line Interleaving. This new technology is called "Scalable Link Interface" or SLI for short. Of course all of us old timers still get excited when we here "SLI" and NVIDIA knows this. The new SLI is possible because of the new PCI-Express bus that was introduced last week. NVIDIA is telling us that when used with the Unreal3 engine, they are seeing increases in frame rate of up to 87%. To qualify this statement, I think it is needed to note here that an engineer stood behind this statement, not just NVIDIA's marketing department.

Tricky Nvidia.

This is how the new Scalable Link Interface technology works between both GPUs. As you can see these cards are NOT rendering even or odd lines, they are in fact rendering whole sections of the screen at a time. In this picture it looks like one card is rendering three fourths of the screen while the other card is doing one fourth of the screen. NVIDIA has a dynamic load balancing logic that will assign each card even workloads rather than simply splitting the screen into equal size portions. The load balanced image is then put together for you on your display.

Well, there is that U Shaped bridge chip. I wonder if that has something to do with it? I'm still reading the Hardocp article I'll see if I can find something.

 

[Fafalada]

From what is explained at Tom's, this only stands to accelerate front buffer rendering, or render targets that are directly tied to front buffer camera.
Actually since applications aren't "aware" of this, the load balancing drivers pretty much can't do anything with offscreen targets except duplicating them across both cards - anything else would stand a great chance of rendering things completely wrong.

I suppose it's better then nothing for people that have money to waste on multiple cards, but I'm not impressed.

 

[Keio]

3dfx will never die! I can still hear that Voodoo Mojo...

 

[Mrbob]

From what is explained at Tom's, this only stands to accelerate front buffer rendering, or render targets that are directly tied to front buffer camera.
Actually since applications aren't "aware" of this, the load balancing drivers pretty much can't do anything with offscreen targets except duplicating them across both cards - anything else would stand a great chance of rendering things completely wrong.

I suppose it's better then nothing for people that have money to waste on multiple cards, but I'm not impressed.

But will Nvidia be able to get a huge boost out of performance from this method? They say so but I don't know how much of it is BS and how much if it is true.

Right now the cost of entry is outrageous for this type of setup. A year, two years down the road it won't be. Especially if Nvidia has plans for all 6800 cards to be able to link up.

I figure they have to be careful with this too. Lets say they release their new videocard in 14 months, and I'll call it the Nvidia 7700 Ultra. It is an awesome card, but it costs 500 bucks. The 6800GTs are now 175 bucks each. Through various benches we figure out that two 6800GTs linked together can come close to the performance of the 7700 Ultra. You could buy two 6800GTs for 350 instead of buying their top of the line next gen card and save some money while getting nearly the same performance! In theory this works but we'll have to wait and see.

 

[marsomega]

btw, I have no doubt that ATI will respond.


there were cards from Evans & Sutherland with dual R200s / Radeon 8500s
as well as E&S cards with dual and quad R300s / Radeon 9700. while that might not exactly be SLI, it's the same result, if not better.

I think they can easily respond.

ATI need only respond with a press release and some team up with some offshoot place that needs that sort of functionality. ATI's R3XX class GPUs are scalable up to 256 gpu's or more I Think. Don't quote me on that though since I don't remember the exact details; moreover ATI had a pre-release some time ago where some company made something I think was called GXCUBE or something like that used a number of ATI's R3XX in parallel or what not.

So technically, they don't need to lose sleep responding to that.


Mr Bob, I agree with you on calling Nvidia's bluff. We all know and seen that twice the CPU or GPU is a far cry from twice the performance; Let alone a 50 percent increase. We also know two wrongs don't make a right. And since we know Nvidia's 6800 architecture isn't perfect, the weaknesses of the architecture it self are emphasized. Least I think.

Another issue is the memory limitation. PCIx still can't match a graphic cards own bandwidth, meaning the data has to be preloaded twice no? If so, the motherboard and the cpu would bottleneck. I don't see how they could get such a performance increase.

Then again I'm not a tech head, can anyone clarify?

 

[Mrbob]

Nvidia is being a bit sneaky on the naming. I found this and placed it in one of my posts above but I'll repost here to make it easier to find:

Now in 2004 NVIDIA has introduced their new SLI technology, but note that it does not stand for Scan Line Interleaving. This new technology is called "Scalable Link Interface" or SLI for short. Of course all of us old timers still get excited when we here "SLI" and NVIDIA knows this. The new SLI is possible because of the new PCI-Express bus that was introduced last week. NVIDIA is telling us that when used with the Unreal3 engine, they are seeing increases in frame rate of up to 87%. To qualify this statement, I think it is needed to note here that an engineer stood behind this statement, not just NVIDIA's marketing department.

Sneaky, Sneaky nvidia.

 

[marsomega]

I know, I edited part of my response out. But I was calling on them on it till I read the quotes more carefully. But believe it or not, it got a few people on here so I can imagine its very effective in deceiving people. Since many don't know or don't remember what SLI really was. (It was a method of dividing up the work between two graphic cards) Very clever...

 

[dream]

Who needs this?

Voodoo2 SLI was useful because it was the only way to get 1024x768 at the time. Who needs two NV40s and why?

 

[marsomega]

Mr. Bob, OH. MY. GOD. I just read some of the other forums on RAGE3D non-ati card forums and NVnews or what not. Talk about sitting ducks....these people are thick. Can't blame Nvidia, they really hit it with this clever scheme.

 

[GigaDrive]

Voodoo2 SLI was useful because it was the only way to get 1024x768 at the time. Who needs two NV40s and why?

those who will want dramatically better framerates in upcoming games that will make one NV40 choke (i.e. Unreal 3)

 

[golem]

Who needs this?

Voodoo2 SLI was useful because it was the only way to get 1024x768 at the time. Who needs two NV40s and why?

need is not part of the equation (unless maybe you want to run full res on an apple widescreen or something)

 

[dream]

So people are actually going to set up a dual-NV40 rig in anticipation for games that are scheduled to ship in 2006?

(along with more powerful single-card solutions that will probably offer even better performance than a SLI'd pair of NV40s)

need is not part of the equation

Touche. But what tangible benefits does it offer? When you can run at 16x12 with full AA and AF on a single card...throwing more silicon in a box is the ultimate example of diminishing returns.