Two 7800GTX 512s...on ONE card?

[superklye]

http://www.pcper.com/article.php?aid=195

This HAS to be a joke. Right?

[superklye]

Nevermind. I see there's already a thread about this.

[aAv7]

change your avatar :p

[Tho Jo Smale]

I call BS on Nvidia claiming the core logic splits the x16 pci-e bus into 2 full 16x sub-buses, one for each Gpu then recombines it for transfer to the mobo. It's a conspiracy to eliminate 8x AGP when they know darn well they haven't used the 2gb/s max bandwidth up yet. PCI-e would only, and I repeat only be necessary for SLI. In normal SLI setup does each card not run at 8x pci-e?

[FraGTastiK]

Where did you get that from?I dont think nVIDIA claimed that.Quad SLI works on SLI x16 nF4 motherboards with total of 32 PCI-e lanes wich then splits into four x8 lanes for each GPU just like how two Asus Dual 7800 GT worked in SLI mode on A8N32-SLI motherboard.

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2665

[Tho Jo Smale]

"There is a new chip on the card that NVIDIA developed that acts as a physical PCI Express bus splitter. It takes the single physical X16 PCIe slot that is fed to the card and splits it into two full X16 PCIe connections; one to each GPU. This logic is then responsible for the merging of the data back into a single X16 PCIe connection when it goes back across the bus. The two PCBs are connected under the SLI bridge connections with a custom interconnect that allows the two GPUs to function in SLI mode."

right from the article ;) that Super started this thread with.

I agree that it probably does work as 32x split to 8x for 4 cards. But then this would mean that the pci-e cards only need 8x agp to avoid bottlenecking. If this were in fact the case a single 7800 series card would only need an 8x Agp slot. Nvidia has stated that a 7800gtx would be bandwidth limited in an 8x agp slot. What I'm proposing is that Nvidia chooses not the make the 7800 Agp, not simply that it wouldn't work. They are making a choice about which market share they choose to support and are, in fact, the driving force behind the move to pci-e. I believe that Pci-e is not necessary for a single card solution.

[shabby]

Most likely a typo.

[Son Goku]

Well, for anyone who remembers... Back in the Voodoo 2 days, Quantum 3D came out with the Obsidian card which had essentially had one Voodoo 2 board, pluged into another, which then only took one PCI slot in the PC. I remember, as I used to own one...

[slick]

change your avatar :p

Why? Wedding Crashers is a GREAT movie. :D

[retsam]

I agree that it probably does work as 32x split to 8x for 4 cards. But then this would mean that the pci-e cards only need 8x agp to avoid bottlenecking.AGP 8x != pci-express 8x they are two diffrent techs

[Monnie Rock]

Well, for anyone who remembers... Back in the Voodoo 2 days, Quantum 3D came out with the Obsidian card which had essentially had one Voodoo 2 board, pluged into another, which then only took one PCI slot in the PC. I remember, as I used to own one...


Yeah baby. The X-24 I thought I was hot ***** with 24 megs of video ram and 90fps in Quake II.

I still have mine

[Rakeesh]

Well, for anyone who remembers... Back in the Voodoo 2 days, Quantum 3D came out with the Obsidian card which had essentially had one Voodoo 2 board, pluged into another, which then only took one PCI slot in the PC. I remember, as I used to own one...

Well, PCI was parallel. Basically all PCI ports shared the same fixed amount of bandwidth. So naturally, one slot could take up all of the available bandwidth if it was called for. AGP cards automatically took priority over all of the devices on the PCI bus, and thus is partially why AGP cards had such an advantage over PCI cards; under high loads they didn't have to share anything. PCI-e is serial on the other hand. Each lane is basically its own bus.

It is possible that nvidia cards use up more bandwidth than is available on AGP 8x, but not all of the bandwidth available in 16 PCI-e lanes. Here is my personal thought about what is actually going on here. Just because you essentially have two GPUs sharing the same 16 lanes, doesn't mean that they have to use the same combined bandwidth as two entirely separate cards. Since they are essentially the same device, they can communicate with one another without having to send data through the PCI-e bus in order to do so. Thus there should be enough bandwidth available to accomodate both cards. And from a logical standpoint, they appear to both the OS and the drivers as two separate PCI-express cards, which is what nvidia was referring to.

[retsam]

It is possible that nvidia cards use up more bandwidth than is available on AGP 8x, but not all of the bandwidth available in 16 PCI-e lanes.remember that agp can only communicate unidirectionally as aposed to PCIe's bidirectional communication. so another words, no wait states for read/write...

[Razor1]

"There is a new chip on the card that NVIDIA developed that acts as a physical PCI Express bus splitter. It takes the single physical X16 PCIe slot that is fed to the card and splits it into two full X16 PCIe connections; one to each GPU. This logic is then responsible for the merging of the data back into a single X16 PCIe connection when it goes back across the bus. The two PCBs are connected under the SLI bridge connections with a custom interconnect that allows the two GPUs to function in SLI mode."

right from the article ;) that Super started this thread with.

I agree that it probably does work as 32x split to 8x for 4 cards. But then this would mean that the pci-e cards only need 8x agp to avoid bottlenecking. If this were in fact the case a single 7800 series card would only need an 8x Agp slot. Nvidia has stated that a 7800gtx would be bandwidth limited in an 8x agp slot. What I'm proposing is that Nvidia chooses not the make the 7800 Agp, not simply that it wouldn't work. They are making a choice about which market share they choose to support and are, in fact, the driving force behind the move to pci-e. I believe that Pci-e is not necessary for a single card solution.


I don't know but in the Dell presentation at CES they were talking about a 2 gb vram buffer, regular SLi only has a max of what ever cards you are using. Don't know if this is a play with words, or nV has found a way to share memory between boards across the PCI-e bus.

[mon]

Quad-SLI

MMM wot nexted

[Telomerase]

http://www.3dfx.ch/gallery/albums/album33/2_G_003.sized.jpg

http://phantom.sannata.ru/konkurs/img/quantum3d/mercury.jpg

i think SLI is probably less efficient than SMP (non-dual core) of CPUs... so dual core GPUs would see a huge performance leap from the parallelism. not to mention Quad-SLI isn't exactly "new."

http://digilander.libero.it/F1Land/3dfxarchive/download/media/gallery/quantum3D-mercury-16x3dfxvoodoo2pci-96mb-2.jpg

http://digilander.libero.it/F1Land/3dfxarchive/download/media/gallery/quantum3D-mercury-16x3dfxvoodoo2pci-96mb-1.jpg

[seeker010]

http://www.3dfx.ch/gallery/albums/album33/2_G_003.sized.jpg

http://phantom.sannata.ru/konkurs/img/quantum3d/mercury.jpg

i think SLI is probably less efficient than SMP (non-dual core) of CPUs... so dual core GPUs would see a huge performance leap from the parallelism. not to mention Quad-SLI isn't exactly "new."

http://digilander.libero.it/F1Land/3dfxarchive/download/media/gallery/quantum3D-mercury-16x3dfxvoodoo2pci-96mb-2.jpg

http://digilander.libero.it/F1Land/3dfxarchive/download/media/gallery/quantum3D-mercury-16x3dfxvoodoo2pci-96mb-1.jpg
and therein lies the confusion. gpus are very parrallel. so having a dual core GPU would just mean doubling its execution units. which is what seems to be going on in gpu advances anyway.

[Ancient]

Well, for anyone who remembers... Back in the Voodoo 2 days, Quantum 3D came out with the Obsidian card which had essentially had one Voodoo 2 board, pluged into another, which then only took one PCI slot in the PC. I remember, as I used to own one...
Quantum 3D also had the Voodoo 5 6000, the card that so many were waiting to be released commercially that never saw the light of day, though some were able to purchase them later off of eBay.

Quantum 3D is still around too. Except now they are partners with NVIDIA and have multi-channel systems that can be purchased, though they are primarily used for military simulation.

[ViN86]

lets be honest, what processor can put enough info through to a quad SLI setup? sure youll get insane eye candy, AA, AF, the works at pretty incredible resolutions. but 7 series GT/GTX/GTX512 SLi setups could do the same at modest resolutions.

[watever]

remember that agp can only communicate unidirectionally as aposed to PCIe's bidirectional communication. so another words, no wait states for read/write...

Dont u only need one direction to go? from cpu to gpu? eh i dunno...

[OWA]

lets be honest, what processor can put enough info through to a quad SLI setup? sure youll get insane eye candy, AA, AF, the works at pretty incredible resolutions. but 7 series GT/GTX/GTX512 SLi setups could do the same at modest resolutions.
Yeah, seems like it might be useful for the somewhat new 30" LCDs. Gaming at 2560x1600 might be tough on regular SLI setups.

[slaWter]

Yeah, seems like it might be useful for the somewhat new 30" LCDs. Gaming at 2560x1600 might be tough on regular SLI setups.

^^
Yes, I hope this will change a bit with the new G71's
Otherwise I'm going to use my upcoming 30" on the MacBook Pro.