More SLI
- Thread starter PatrickL
- Start date
I thought that the V2 SLI used the VGA passthrough to also do the SLI? Hrm, maybe I'm mistaken. I had thought that digital recombination of the image was not done until the Voodoo5.Mordenkainen said:
It doesn't really matter, because if this were true, then they'd be bundled with any SLI-capable motherboard.
I really doubt it, considering the connection is probably just a PCI Express 1x connection.
Voodoo 1/2 SLI passed an analog single between the 2 cards using the SLI cable. It DID degrade the image vs what you would get with just a single card for some people.
Jawed
Legend
Chalnoth said:
ATI's IGP is rumoured to be based upon X600. If it also has mimimum local memory with an HyperMemory architecture (e.g. 64MB local), then it may be necessary to have more than PEG x1 to make it work.
I don't know what a sensible minimum number of lanes would be for X600 IGP. But going beyond two or four would imply to me that ATI would be seriously considering de-integrating IGP from NB and making a "PEG Router" the foundation of its forward looking chipsets. If ATI's doing this right, then it has to consider a future where higher performing IGP becomes necessary (e.g. X700).
If the chipset is required to provide HyperMemory support for IGP, it may turn out to be a neat solution to the problem of sharing textures across both GPUs in a Multi Rendering configuration.
If ATI's MR is based on quad "load-balancing" rather than alternate frame rendering or some other large-grained parallelism (such as SLI style "split screen"), then both cards are always going to need the majority of textures per frame.
The HyperMemory architecture with integrated PEGx16->x32 routing (so that all textures can be broadcast to both cards when demanded by either card - hence requiring only one texture fetch from the system RAM) may be the only way for ATI to deliver efficient texture data bandwidth for MR.
There's also the question of MR's quad-based impact on geometry calculation. It might be sensible to integrate geometry swapping.
Ultimately all I'm doing is thinking of alternatives to a link-based SLI-like architecture (using PEG itself the link) and wondering if HyperMemory support becomes an intrinsic concept in a shared IGP/16-32x PEG Router architecture.
Jawed
Jawed
Legend
Chalnoth said:
If IGP requires more than 4 lanes to support the performance of an X600 part, then it may be more fruitful to break the IGP out to a separate chip that acts as a PEGx16 node, which also provides onward routing to a single add-in card. Not for MR but for 3-monitor support.
I'm talking about an architecture not a single multi-function chip.
I'm not suggesting that IGP becomes part of MR. I'm suggesting that the broader architectural components that may be required for MR (HyperMemory and PEG Routing) would also be applicable in a distinct chip IGP - so that there's a large functional overlap between the two guises of Kaleidoscope: IGP and MR.
In all cases mobo NB sees a single 16-lane PEG solution hanging off its side, regardless of whether the chipset is for a single add-in card, IGP or MR.
Jawed
Chalnoth said:
Nah, the V2 is still only a 3D accelerator and needed a 2D card to work so it had the passthrough VGA cable in the back. I still have my remaining V2 hooked to a GF2 GTS on my older computer in this way. I always thought that was the reason for the degradation and not the internal SLI cable. I certainly never noticed any additional degradation from single V2 to when I had two V2 boards in SLI configuration.
Yes, but only SLI capable mobos with a nVidia chipset and probably AMD only. Btw, is nVidia jumping on the Intel mobo bandwagon or what?
I see.
Chalnoth said:
Exactly. That's why I was surprised when Dave said only nForce 4 mobos would come with the SLI connector and not the card themselves. If that means all SLI-ready mobos carry the connector as Fodder pointed out then that's not going to be a problem.
