'SLI On A Stick' Reviewed 188
Bender writes "What would happen if you took NVIDIA's multi-GPU teaming capability, SLI, and stuck it onto a single graphics card? Probably something like the GeForce 7950 GX2, a 'single' video card with dual printed circuit boards, dual graphics processors, dual 512MB memory banks, and nearly twice the performance of any other 'single' video card. Add two of these to a system, and you've got the truly extreme possibility of Quad SLI. We've seen early versions of these things benchmarked before, but the latest revision of this card is smaller, draws less power than a single-GPU Radeon X1900 XTX, and is now selling to the public."
Re:Wow (Score:3, Informative)
While they may be overkill for your average user, for (game) developers these things will be goldmines..
-K
Re:Wow (Score:3, Informative)
Less than one Radeon X1900XT (Score:1, Informative)
In case you're like me (Score:5, Informative)
Re:Completely off topic (Score:1, Informative)
The more you know.
Re:Completely off topic (Score:1, Informative)
Re:Wow (Score:2, Informative)
This is not Quad SLI, its Dual SLI at best. (Score:1, Informative)
so SLI = 2 cards or more
normally you have 1 GPU per card, so that means that with the standard or "old" SLI you would have a Dual GPU setup.
now, with 2 GPU's per card you can link those 2 cards together and get SLI, but with Quad GPU power.
So, it's not "Quad SLI", but "Quad GPU" or "Dual SLI".
Re:Wow (Score:4, Informative)
Oblivion at 1920x1200? Good thing I don't have an Apple Cinema Display. Personally I think Oblivion's game engine is a bit overrated. Ok it's pretty but not *that* much prettier than the other freeform 3D games that don't kill my GFX card. Right now I'm working on a HOMM5 addiction instead...
Pro and cons (Score:3, Informative)
Also recently announced on slashdot, the developpement of a standart hypertransport connector [slashdot.org] (as part of the HT 3.0 revision).
So maybe in a near future you'll see motherboards featuring HyperTransport connectors, in which you could directly plug CPU/DDR board, GPU/GDDR board, or specialist copreocessor boards (Phys-X, FPGAs, ultra-high speed raid & networks board for severs, etc...)
On the other hand : supporting different size of memory for a GPU is something that is going to bring more complexity and costs. (this was something discussed in forums about building open-source GFX card [duskglow.com]) so diffrent GPU/GDDR board speaking among them thru HT bus is more likely than a GPU board with upgradeable GDDR.