ATI Multi-VPU up and running. . .

[psurge]

There's a paper on a graphics architecture called Pomegranate that scales everything (geometry, rasterization, pixel processing, etc...). If I recall correctly, this was accomplished by feeding the output of each stage of the graphics pipeline (transform, rasterize, shade, etc...) into a butterfly network, allowing work at all stages in the pipeline to be load balanced.

Edit: Here's the paper:
http://graphics.stanford.edu/papers/pomegranate/pomegranate.pdf

 

[MuFu]

And a link to some Inq psuedo-sense on multi-core GPUs:

http://www.theinquirer.net/?article=21822

When we talked with ATI's CEO, Dave Orton confirmed to us that the industry will have to move that way once it starts to make financial sense.

 

[Rys]

Should that be the method employed by a solution from ATI, should they be seeking a multi graphics solution, Rys hasn't quite got how the AA may work right, IMO.

Yes. Bleh, my bad. Not sure where I plucked multipassing the tiles from. My arse, it seems

 

[Hyp-X]

In order to increase geometry processoring you would need something like the 3DLabs solution with an additional geometry chip OR you need to bin the polygons in the driver. Don't know the performance trade-off with binning in the driver.

You cannot bin for supertiling - the chance that a geometry doesn't overlap both chips is really small (well, thats the point in tiling that way).

You can bin for split screen - but that has other issues.

 

[wireframe]

Is there some particular reason why I have not seen the word "kaleidoscope" mentioned once in this thread? Just wondering...

 

[Dave Baumann]

Because it bears no relation.

 

[Geo]

Because it bears no relation.

That's actually the most definitive thing I've heard about Kaleidoscope yet!

 

[Joe DeFuria]

Because it bears no relation.

That's actually the most definitive thing I've heard about Kaleidoscope yet!

I would speculate (read: based on nothing whatsoever) that Kaleidoscope could refer to one of two different things:

1) http://www.beyond3d.com/forum/viewtopic.php?t=20447

Just in case there any doubt, GPUs will ship with multiple 'units' as illustrated here just as Cell ships with multiple SPUs...

or

2) A total revamp of the multi-display technology. (So long "hydravision").

 

[kemosabe]

We have a claim that the "kaleidoscope display controller" is related to AMR, which according to Dave doesn't exist. So we're making remarkably steady progress.

 

[CyFactor]

We have a claim that the "kaleidoscope display controller" is related to AMR.

And if that claim is innacurate, it brings into question the accuracy of the rest of their data, doesn't it.

"It bears no relation."

 

[Geo]

We have a claim that the "kaleidoscope display controller" is related to AMR, which according to Dave doesn't exist. So we're making remarkably steady progress.

What Dave said didn't exist was "AMR", and offered "MVP" instead on the same thread. Pics of the new ASUS mobo bear out that the tech is indeed called "MVP" (barring a last minute change before general public availability).

This thread subject here where he posted "bears no relation" does not refer to the "AMR" tag.

So, Wavey doesn't have a weasel (gee, must be getting close to release season!) --Kaleidoscope bears no relation to ATI's multi-vpu tech.

 

[Dave Baumann]

And if that claim is innacurate, it brings into question the accuracy of the rest of their data, doesn't it.

You'd think, wouldn't you.

 

[Geo]

I would speculate . . .

Thanks, Joe. I'm actually familiar with the speculation on this one --I was just pleased to hear a simple declarative sentence with no caveats by someone who wouldn't have offered it without a high degree of certitude. For such a catchy name, there's not been much red-meat dug up on it yet. This probably means its either kinda-nice-no-big-deal-for-most or big-deal-commit-suicide-after-reading. (probably the former tho, darn it).

 

[Jawed]

Is there some particular reason why I have not seen the word "kaleidoscope" mentioned once in this thread? Just wondering...

http://www.beyond3d.com/forum/viewtopic.php?p=478638#478638

Jawed

 

[wireframe]

Is there some particular reason why I have not seen the word "kaleidoscope" mentioned once in this thread? Just wondering...

http://www.beyond3d.com/forum/viewtopic.php?p=478638#478638

Jawed

I stand corrected Sorry about that. I'll blame it on either ctrl or 'f' having come undone.

 

[Geo]

Back to the topic at hand. . .

If ATI's MVP allows you to MVP two gpu's of the same generation even tho they aren't the same card. . .

Would this mean that the poor bastards who start life with a chipset-integrated ATI gpu could MVP it with an add-in discrete? Seems to me that might be kinda useful for some. . .Mumsy buys you the cheapo-special "for your homework", you save your paper-route dollars and buy a real bad-boy add-in, and then spank your homeys who have the same discrete that you do. . .

 

[wireframe]

One big advantage of an alternate tile rendering approach is the inherent load balancing, so the software doesn't have to do that.

Doesn't AFR accomplish that as well tho? And ATI already has experience with that.

ATI does not reveal the Multi-VPU concepts to the general public, but for its add-in card partners it has posted a brief overview of its technology in its roadmap update. At least for now there is one important update from what was previously reported – ATI wants two identical graphics cards to be installed, not two different boards as said before based on information from unofficial sources. Here is what ATI says to its partners:

* Multiple ATI RADEON X800 XT boards cooperatively rendering a single frame;
* Requires two physical x16 connectors on the mainboard;
* Load balancing and synchronization implemented entirely in software;
* No physical connector requires between devices;
* Currently assumes two identical graphics devices installed in both connectors;
* Offers several user selectable modes of multi-processing;
* Works with any PCIe north bridge.

My bold in the last quote, one that many seem to be basing thoughts on. "Single frame" would seem to indicate AFR is not the method ATI is going with.

 

[Geo]

Ah, so. . .I'd forgotten about that one. Which apparently shoots down not one but two of my recent musings. Alas, alas. . .

 

[Kaotik]

One big advantage of an alternate tile rendering approach is the inherent load balancing, so the software doesn't have to do that.

Doesn't AFR accomplish that as well tho? And ATI already has experience with that.

ATI does not reveal the Multi-VPU concepts to the general public, but for its add-in card partners it has posted a brief overview of its technology in its roadmap update. At least for now there is one important update from what was previously reported – ATI wants two identical graphics cards to be installed, not two different boards as said before based on information from unofficial sources. Here is what ATI says to its partners:

* Multiple ATI RADEON X800 XT boards cooperatively rendering a single frame;
* Requires two physical x16 connectors on the mainboard;
* Load balancing and synchronization implemented entirely in software;
* No physical connector requires between devices;
* Currently assumes two identical graphics devices installed in both connectors;
* Offers several user selectable modes of multi-processing;
* Works with any PCIe north bridge.

My bold in the last quote, one that many seem to be basing thoughts on. "Single frame" would seem to indicate AFR is not the method ATI is going with.

Added another bolding, which imo points to supporting both AFR and SFR

 

[Unknown Soldier]

* Offers several user selectable modes of multi-processing;

Added another bolding, which imo points to supporting both AFR and SFR

Several might mean that ATI has other modes available too. So what other modes could ATI add to the potion?

US