View Full Version : G72 is coming...
Dual core cards are kinda a waste IMO.
Each GPU needs to write to the framebuffer, so, you could have 512mb of ram on a card, but only 256 is actually usable. Ram is very expensive, so it drives the price of the dual core card up... It works the same with SLI, but i'm just talking single vs. dual core.
You would be better off with a similar performing single card :)
(Lets go with "2 GPUs" instead of "Dual Core" to aviod any confusion here hehe)
I wonder if there couldnt be a way to have 2 GPUs share the RAM on the board, I mean you couldnt do that over 2 cards, but when the GPUs are on the same board..
You could have a designated amount of RAM for a shared framebuffer, where both wrote to depending on what mode your running.. well either that or their own framebuffers, either way the compositing could be done in framebuffer right away..
And the rest of the RAM that keeps all textures and data to be shared..
So with for example a 512mb card you have a certain amount dedicated for either 2 separate, or 1 shared Framebuffer, and the rest for shared storage of textures and data, and it would work pretty much like a full 512mb card.
Would be nice if it worked, but then its not really "SLi" anymore.. its just a big expensive and really fast card.
(Lets go with "2 GPUs" instead of "Dual Core" to aviod any confusion here hehe)
I wonder if there couldnt be a way to have 2 GPUs share the RAM on the board, I mean you couldnt do that over 2 cards, but when the GPUs are on the same board..you know what...i think nvidia is gonna try to do something with this in there next iteration of sli.... i think they know that having 1 gig of ram between two board but only using 512 is a waste ..im also thinking maybe this is why ATI has gone with a ring bus on the r520... but who knows...
Anandtech (http://www.anandtech.com/video/showdoc.aspx?i=2570) has some info on G72 and G73.
(Lets go with "2 GPUs" instead of "Dual Core" to aviod any confusion here hehe)
I wonder if there couldnt be a way to have 2 GPUs share the RAM on the board, I mean you couldnt do that over 2 cards, but when the GPUs are on the same board..
you know what...i think nvidia is gonna try to do something with this in there next iteration of sli.... i think they know that having 1 gig of ram between two board but only using 512 is a waste ..im also thinking maybe this is why ATI has gone with a ring bus on the r520... but who knows... wow it looks like i might be right....
http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=/netahtml/search-bool.html&r=3&f=G&l=50&co1=AND&d=ptxt&s1=nvidia&OS=nvidia&RS=nvidia
a plurality of memories including a first memory and a second memory, each of the plurality of memories including a plurality of addressable storage locations, wherein a first storage location in the first memory and a second storage location in the second memory are addressable by a common global address in the global address space; and wow maybe 1 gig of memory is comming sooner then we think!!!
If they're pin to pin combaptible, chances are that we will be seeing AGP variants too.
The only thing that would make me upgrade my 6600GT is a 256MB variant.
The fact that GPUs have been "dual-core"... multi-core even... for years now flies over many heads at 30,000ft. You have a 16-/24-"core" GPU in your PC *right now*.
Exaxtly...well almost like...each quad block is an exact copy of the other sharing the same resource, and probably each vertex shader unit is the same sahring the same resources....parrallelsim on a huge scale and almost multi corish! Thats why you have a 8 quad gpu in the GTX and then 7 quad Gpu in the GT, just knock on of the quad block cores ;)
Redeemed
10-19-05, 02:06 PM
That bit about flip chip GPUs, that is cool. So, essentially, the vcard will become a smaller "motherboard" (maybe, "sister" or "daughterboard" fit better? ;) ), where you can then go and buy which ever GPU you preffer, with whicher cooling method you prefer, and which whatever ram in whatever quantity at whatever speed you prefer. Wow. Imagine a 6800GT with 512mb of 512bit 1.2ns quad-channel GDDR4 memory operating at 1.8GHz atleast. Good crap! I gotta' go change my shorts. ;)
deimos47ca
10-19-05, 02:55 PM
Dual-core (in the sense that Intel mean it) for GPUs makes absolutely no sense whatsoever, dual-core in the sense that AMD mean dual-core (multiple cores on a single die) makes even less sense than that! It doesn't really solve any of the problem facing GPUs, and it adds a whole pile of new ones.
These rumours have been floating around since DC CPUs came out, for the sole reason that people don't understand what DC does for a CPU, and why the CPU guys have gone multi-core at this point in time. It's the latest tech buzz-word, so suddenly everyone wants it for their graphics, because it must be good, right?
The fact that GPUs have been "dual-core"... multi-core even... for years now flies over many heads at 30,000ft. You have a 16-/24-"core" GPU in your PC *right now*.
One multi-die solution which *does* make sense is embedded DRAM, a la Xenos. But that's *not* dual-core, it's an MCM.
bingo
althought the eDRAM is only there to cut costs, since no dedicated 256/512MB of video memory.
That bit about flip chip GPUs, that is cool. So, essentially, the vcard will become a smaller "motherboard" (maybe, "sister" or "daughterboard" fit better? ), where you can then go and buy which ever GPU you preffer, with whicher cooling method you prefer, and which whatever ram in whatever quantity at whatever speed you prefer. Wow. Imagine a 6800GT with 512mb of 512bit 1.2ns quad-channel GDDR4 memory operating at 1.8GHz atleast. Good crap! I gotta' go change my shorts.
Keep Dreamin..
Light Years Ahead!!
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.