ATI SLI

[lavaheadache]

Originally posted by: Jeff7181
Isn't it incorrect to call ATI's method "SLI?" Since SLI = Scalable Link Interface... referring to the connection made between the cars with that little connector? If I read things correctly, ATI's will not use such a connector, the communication between the two GPU's will be ALL on the PCI-Express bus.

I thought SLI stood for Scan Line Interleave or something like that, meaning one chip does even lines and the other chip does the odd lines.

 

[BenSkywalker]

I thought SLI stood for Scan Line Interleave or something like that

It did, when 3Dfx was doing it. Now nV is using a new definition.

 

[Gamingphreek]

Scalable Link Interface

 

[lavaheadache]

oic, now i know

 

[Jeff7181]

Originally posted by: lavaheadache

Originally posted by: Jeff7181
Isn't it incorrect to call ATI's method "SLI?" Since SLI = Scalable Link Interface... referring to the connection made between the cars with that little connector? If I read things correctly, ATI's will not use such a connector, the communication between the two GPU's will be ALL on the PCI-Express bus.

I thought SLI stood for Scan Line Interleave or something like that, meaning one chip does even lines and the other chip does the odd lines.

I think Scan Line Interleave had each GPU render every other horizontal line, which was fine back then... but today with such heavy use of shaders (I think mainly pixel shaders), every other line doesn't necessarily split the load evenly between GPU's.

 

[Jeff7181]

Originally posted by: apoppin

Originally posted by: Jeff7181
Isn't it incorrect to call ATI's method "SLI?" Since SLI = Scalable Link Interface... referring to the connection made between the cars with that little connector? If I read things correctly, ATI's will not use such a connector, the communication between the two GPU's will be ALL on the PCI-Express bus.

ATI's SLI to work without connector

ATI'S WAY OF doing SLI called AMR (ATI Multi Rendering) will work differently from Nvidia's now world famous SLI. ATI will make its SLI work without a small printed circuit board (PCB) to interconnect the cards.

SLI uses two PCIe graphic ports and in the case of Nforce 4, works at PCIe 8X for each card and renders your picture with both cards. For this operation Nvidia SLI needs a special interconnection between two graphic cards. It?s a small cable or PCB that transfers up to four GB per second from one card to another. For this operation, coordination is crucial as one card needs to know what the other card is doing.

ATI found a way to make its SLI work without this interconnection. In ATI's case, whenever it makes its SLI ready, you will just plug two graphic cards in motherboard powered with ATI chipset and it will all work without interconnection PCB or cable.

The upcoming chipset will have hardware support for dual graphic card rendering mode. It will be all done through the chipset and we guess that it will be wired in the new RS482 and RX482 chipsets. All the communication between two cards will be done through motherboard or chipset wires.

ATI is working hard to catch up with Nvidia in the SLI game, and it will try to make its chipset and drivers ready for launch as soon as possible. This is expected sometime in Q2 2005.

edit: doesn't ANYONE read?
this is the SECOND time i posted this.
:roll:

Is that really a Scalable Link Interface though? Cause too many people are using "SLI" to describe two video cards working together... that doesn't necessarily mean they're using SLI.

 

[apoppin]

Originally posted by: Jeff7181

Originally posted by: apoppin

Originally posted by: Jeff7181
Isn't it incorrect to call ATI's method "SLI?" Since SLI = Scalable Link Interface... referring to the connection made between the cars with that little connector? If I read things correctly, ATI's will not use such a connector, the communication between the two GPU's will be ALL on the PCI-Express bus.

ATI's SLI to work without connector

ATI'S WAY OF doing SLI called AMR (ATI Multi Rendering) will work differently from Nvidia's now world famous SLI. ATI will make its SLI work without a small printed circuit board (PCB) to interconnect the cards.

SLI uses two PCIe graphic ports and in the case of Nforce 4, works at PCIe 8X for each card and renders your picture with both cards. For this operation Nvidia SLI needs a special interconnection between two graphic cards. It?s a small cable or PCB that transfers up to four GB per second from one card to another. For this operation, coordination is crucial as one card needs to know what the other card is doing.

ATI found a way to make its SLI work without this interconnection. In ATI's case, whenever it makes its SLI ready, you will just plug two graphic cards in motherboard powered with ATI chipset and it will all work without interconnection PCB or cable.

The upcoming chipset will have hardware support for dual graphic card rendering mode. It will be all done through the chipset and we guess that it will be wired in the new RS482 and RX482 chipsets. All the communication between two cards will be done through motherboard or chipset wires.

ATI is working hard to catch up with Nvidia in the SLI game, and it will try to make its chipset and drivers ready for launch as soon as possible. This is expected sometime in Q2 2005.

edit: doesn't ANYONE read?
this is the SECOND time i posted this.
:roll:

Is that really a Scalable Link Interface though? Cause too many people are using "SLI" to describe two video cards working together... that doesn't necessarily mean they're using SLI.

it's "their" 'sli'. i guess "sli" is becoming a catch-all phrase for multi videocard rendering.

"technically" it's NOT "sli" but close enough for a geek forum and for us to know what they are talking about.

 

[humanentity]

i read somewhere maybe at the inquirer that Ati is gonna pass rights to Intel to manufacture Ati Multi Rendering compatible Chipsets. So you dont need to buy an ati mobo to have AMR you will be able to buy an intel powered mobo with amr if u want.

maybe the hard way to go would be looking for amd64 compatible mobo with amr capabilities, then u should be forced to get an ati powered mobo because i suppose intel does not make chipsets for amd64 mobos,... doesnt it?

 

[Drayvn]

Originally posted by: humanentity
i read somewhere maybe at the inquirer that Ati is gonna pass rights to Intel to manufacture Ati Multi Rendering compatible Chipsets. So you dont need to buy an ati mobo to have AMR you will be able to buy an intel powered mobo with amr if u want.

maybe the hard way to go would be looking for amd64 compatible mobo with amr capabilities, then u should be forced to get an ati powered mobo because i suppose intel does not make chipsets for amd64 mobos,... doesnt it?

IT would be hilarious if they did!

 

[Like a Fox Yet Again]

I still don't understand why ATi would make it to where only THEIR mobos could support the new "SLI" cards they will be coming out with. It just doesn't make sense, why would someone buy an SLi mobo for the reason of having "SLI" only to find out they have to buy ATi's board AND cards and sell their mobo. Not many people gonna be happy if that ends up happenning. Jus my $ .02

 

[OnEMoReTrY]

Originally posted by: Like a Fox Yet Again
I still don't understand why ATi would make it to where only THEIR mobos could support the new "SLI" cards they will be coming out with. It just doesn't make sense, why would someone buy an SLi mobo for the reason of having "SLI" only to find out they have to buy ATi's board AND cards and sell their mobo. Not many people gonna be happy if that ends up happenning. Jus my $ .02

The same reason the ATI mobo's won't have NVIDIA SLI... wheres the logic in that? why would an NVIDIA motherboard have ATI AMR?

 

[302efi]

LOL...If ATI cant even produce enough single cards now, how in the hell are they going to be able to handle the bigger demand when each person NEEDS 2 CARDS ??!!!

LOL @ ATI

 

[VirtualLarry]

Originally posted by: Noob
When Alienware showed off it's X2 MB at E3. I think they said that they co-developed it with Intel (can't view the video for some reason) Here is the link:
http://www.alienware.com/alx_pages/main_content.aspx
And here is an article to back it up:
http://dotnet.org.za/craig/archive/2004/10/06/4505.aspx
It says that the X2 MB is based on tech from a future Intel MB. And I think that Intel is still partners with ATI. So there should be an ATI SLI MB eventually.

Except that, Alienware's SLI stuff was based on patented technology originally developed/owned by Metabyte (the Wicked3D guys), and NV and Alienware (apparently) struck some kind of a deal, so it wouldn't make sense to let ATI in on the party. NV's lawyers aren't that stupid to allow that to happen, IMHO.

Still, I would love to see a software-only technology, that would allow running a single game, to drive multiple GPUs and multiple video-displays, for "immersive surround gaming", regardless of the brand of video card(s) used. If MS or Intel somehow got the rights to do that, that might allow it to happen. That would be the ideal scenario, to me at least.

 

[VirtualLarry]

Originally posted by: apoppin
it's "their" 'sli'. i guess "sli" is becoming a catch-all phrase for multi videocard rendering.

Pretty much.

I was thinking about this some time back. Considering how current video cards work, in terms of "overlay surface" support, used for things such as video-capture cards, seperate DVD-decoder cards, and the like, I think that ATI's AMR solution will be similar. Given the raw bandwidth now available with PCI-e, it actually shouldn't be all that difficult at all, theoretically.

The way that I'm guessing that it will work, is that the scene to be rendered, will essentially be sent to both cards (pre-loading the textures, etc.), and one card will have a clip region assigned, causing it to render, say, the top half of the screen, with the other card rendering the bottom half, using a different clip region. Both of these cards are rendering to off-screen buffers at this point. When the "slave" card is done rendering, that rendered buffer data is sent over the PCI-E bus directly (?) from one card to the other, and composited on the back-buffer of the "master" card, pretty-much like how a PCI video card's overlay feature works. Then when the offscreen buffer-compositing is finished, the "master" card does a page-flip to cause it to be displayed. (Optionally waiting for a vsync retrace event.)

That would be the simplest method that I can think of that it could be done with two entirely dissimilar PCI-E video cards. I suppose that it could also be done slightly differently, if there is a way to pull texture data, and then send destination rendering info, directly between cards over the PCI-E bus, but that would be far slower and more complex.

(However, that might be an interesting way to effectively 'SLI' several PCI-E NV 6200TC cards. That might really push the limits of the PCI-E controller and available system DRAM bandwidth though. Perhaps one could develop cards that used the 6200's TC texturing/rendering method, but include more video-RAM on the cards themselves, and use the TC mechanism to access each other card's RAM, instead of both going to/from (shared) system RAM?)

 

[Like a Fox Yet Again]

bump, good thread

 

[Bateluer]

Originally posted by: Schadenfroh

Originally posted by: Noob
I didn't know that Nvidia did that. If Nvidia is smart, they won't sell the rights to ATI.

they wont, ATI will develop their own method without infringement on nvidia's patents. ATI already has a multichip proccess, remember the RAGE FUry MAXX?

Everybody always forgets about the Rage Fury MAXX cards.

I think ATI will have their own multi-GPU technology, likely a refined AFR system like that used in the MAXX.

Multi-GPU tech is an EXTREMELY niche market though. 1) The cost of two high end PCIe vidcard is between 800 and 1000 USD. 2) Next gen card will likely outpeform current SLI configs at half the cost.

Good for bragging rights in the present though