PCI Express SLI by Alienware!

[Dave Baumann]

The only proprietary components required will be Alienware’s exclusive software layer and an Alienware-designed video merger hub.

Mmmmm, that doesn't sound like its alternate frame rendering, but more like slitting the frame up between two boards and having them render half the frame each. I wonder how they sync the two boards.

 

[Joe DeFuria]

The only proprietary components required will be Alienware’s exclusive software layer and an Alienware-designed video merger hub.

Mmmmm, that doesn't sound like its alternate frame rendering, but more like slitting the frame up between two boards and having them render half the frame each. I wonder how they sync the two boards.

Actually, I disagree.

I mean, if you wanted to do AFR with two separate video cards, I would assume you would in fact need some kind of hardware to "interleave" (merge) the two video signals to the single monitor.

 

[Dave Baumann]

Possibly. I'm just wondering if they've got the guy that used to have a method for this (I forget the name name - but it cropped up around the time 3dfx went with the V3); its thought he went to NVIDIA but nothing has emerged from there, and this sounds like it has similarities with his method.

 

[Colourless]

The idea so will not work properly with Direct3D at all. Locking resources, rendering to textures and a number of other things will be totally broken or so slow it will be slower than with a single board.

 

[Joe DeFuria]

It is a bit of a mystery...B3D definitely needs to get one of these to test.

Seriously...this would be a nice canditate perhaps for a brief interview at least.

 

[Dave Baumann]

I mailed the press contact yesterday, but I assume they are all at E3 at the moment.

 

[flf]

Everyone has seen the post on Slashdot that mentions the blurb at Ars Technica, yes?

The relevant piece:

Update: Readers on the floor of E3, with whom I've been in communication, have seen this running in an alpha state. This is not vaporware. This is not a multi-monitor technology, either. The current implementation splits the screen in half, assigning rendering for each half to one of two cards using a software load balancer to try and ensure proper synchronization. (You may recall that 3dfx's SLI technology split the workload by alternating lines. This is a slightly different approach.) The motherboard is being produced for Alienware by iWill, and Alienware is currently saying that they expect users to see a ~50% performance boost over single card implementations.

 

[Tahir2]

Yes I saw it and currently I dont know what to make of it except that if the screen is split into two then I hope it is split on the horizontal axis otherwise one card would be doing more of the work....
Sounds intriguing but I aint holding my breath for this one!

 

[AlphaWolf]

Yes I saw it and currently I dont know what to make of it except that if the screen is split into two then I hope it is split on the horizontal axis otherwise one card would be doing more of the work....
Sounds intriguing but I aint holding my breath for this one!

Well perhaps the ideal setup is meant to be a fx5200 for rendering sky, and a 6800nu for the rest of the screen. It's a cost saving over a 6800U.

 

[CyanBlues]

if this is possible i wonder if ati can come up with even better temporal aa settings from using 2 cards

 

[Joe DeFuria]

Everyone has seen the post on Slashdot that mentions the blurb at Ars Technica, yes?

If I had seen it, then I would have known the answer. Thanks for the link.

 

[Dave Baumann]

if this is possible i wonder if ati can come up with even better temporal aa settings from using 2 cards

The full multi-processor R3x0 boards already use a method where one quad on one processor will render pixels with one sample pattern (up to 6X AA) while another quad on another processor will render the same quad of pixels but with different sample patterns - the results are blended together before the final display. Its not quite TAA since its all working on the same frame, but it is distributed AA processing across multiple processors.

Anyway, can anyone remember the name of the SLI alternative that cropped up a few years back? Its bugging me now!

 

[Tim Murray]

Alternate Frame Rendering? Rage Fury Maxx.

Apparently Metabyte had a Parallel Graphics Configuration one...

 

[AlphaWolf]

Alternate Frame Rendering? Rage Fury Maxx.

Apparently Metabyte had a Parallel Graphics Configuration one...

Apparently alienware had licensed parallel graphics processing (PGP) in 1999.

 

[Dave Baumann]

Ahhhh, yes, metabyte, that was it

http://www.sharkyextreme.com/hardware/metabyte_onsite/

 

[3dcgi]

Some of you may be interested in reading about the Chromium project at Sourceforge. Here's a description.

Chromium is a system for interactive rendering on clusters of graphics workstations. Various parallel rendering techniques such as sort-first and sort-last may be implemented with Chromium. Furthermore, Chromium allows filtering and manipulation of OpenGL command streams for non-invasive rendering algorithms.

http://chromium.sourceforge.net/

 

[BRiT]

Damnit, did no one read my post back on page 2 that had linked to all this information already? [sharky/metabyte/split-screen]

 

[Pete]

Yes I saw it and currently I dont know what to make of it except that if the screen is split into two then I hope it is split on the horizontal axis otherwise one card would be doing more of the work....

Yes, I believe that point was brought up when Metabyte's tech was intro'ed a few years ago, that the card rendering the sky would be held up by the card rendering, well, everything important. But I'm not sure it's even possible to split the screen vertically (left and right, rather than top and bottom). Don't video cards draw from top to bottom? Can they draw another way? Can software actually force a card to only render half-width? Or was it just that joining the two images of a left-right split screen was more difficult than top-bottom?

 

[no_way]

Can software actually force a card to only render half-width?

Hmm, a single clip plane insert into command stream would do the trick, no ?

 

[Colourless]

Scissor rect is better