A RAID controller that stripes a dozen drives?Quote:
Originally Posted by aidanjt
I have an SLI motherboard despite no intention of ever using SLI. The other features were better than the cheaper models. There are now far cheaper boards out there with more PCIe channels though.
Not sure if this is the original one I heard about, but:Quote:
Originally Posted by DavidM
http://www.bionicfx.com/info.html
People do physics on GPUs already. A modern GPU can do anything that you can express as a highly parallel set of calculations.
Many thanks - always interested in projects like this.Quote:
Originally Posted by DanceswithUnix
Very interesting reading.
Another question, what is the advantage of the DFI Ultra-D with 2 x PCIe x16 and it is not SLI, surely they had something in mind at the design stage ?
I think there are 2 sides to the not-quite-SLI motherboards.Quote:
Originally Posted by behappier
1/ NVIDIA currently ask for money per board for the SLI sticker. That might relax as crossfire becomes more competative so in future you _might_ end up with SLI. DFI did a board in the early SLI days that could be modded to SLI with a bit of soldering.
2/ Some people want multi monitors, which just require multi graphics cards so why pay the extra? I gather a flight sim played surounded by monitors is pretty impressive!
Shame it doesn't seem to have matured as a product yet. The ability to sing a song in the Albert Hall thanks to your graphics card seems pretty neat :)Quote:
Originally Posted by DavidM
I believe it was http://www.theinquirer.net/ where I read about Nvidia being able to do physics on their GPU, but can never find anything with their search facility.
This makes me wonder if, some time in the future, chipsets like the 6150 integrated graphics will become more popular as if gives you a couple of extra pipelines for physics/sound/whatever sitting around!
Physics Processing Units (PPUs) are an interesting issue. I can imagine that a special purpose chip could do many types of physics processing much faster than a general-purpose CPU. And that such a device could conceivably use a lot of bus bandwidth (such as a PCI-express 16-lane). So I could perhaps envision a PPU and GPU taking the two PCI-express slots on an SLI or Crossfire mobo. One question though is whether the SLI/Crossfire mobo (or driver) will even allow the second slot to be used for something other than a graphics card. We don't have a clear answer on that yet. Anyone out there with an answer?
You make a very good point. Why buy a special-purpose PPU, when general-purpose multi-core CPUs might well serve the purpose sufficiently well? On the other hand, I've contacted the PPU-maker about other uses for their PPU, uses other than gaming -- such as science-simulations for research, educational tools for demos, charting-and-graphing calculations, and so forth. These non-gaming applications would likely help the general acceptance of these PPU boards. The PPU-maker indicates that they are planning on it, but that they intend to hit the gaming market first.Quote:
Originally Posted by shaithis
Anyway, there seems to be a race between the PPU applications on the one hand, and the general-purpose multi-core CPUs on the other hand. So it's difficult to predict. Will the game developers write for the PPUs, or the multi-core CPUs? Who will win? Hard to say.
But it's all affecting my thinking about SLI/Crossfire mobos. I'm unlikely to ever use the second slot for graphics, so is that slot good for anything else? Even the serious possibility that it might be good for something (a PPU, a high-speed ethernet card, a high-tech RAID controller) can give hope to the matter. On the other hand, if the second slot is devoted (due to hardware or software limitations), now and evermore, to be only for graphics, then that changes the picture. In that case, it is perhaps better for me to go with a non-SLI/Crossfire mobo. Any thoughts?
Given how much faster GPUs are compared to CPUs (you'd need about a 100GHz CPU to keep up with a GPU for rendering) in their given domain, I suspect a PPU would be a LOT faster than doing physics on the second core.
The same was sead about GPUsQuote:
Originally Posted by shaithis
There is/was an Ultra-to-SLI mod for the chipset (the Ultra & SLI are the same physical part just with different bridge connections). nVidia charge more for the SLI part than the Ultra so this was a way to provide an SLI board at a cheaper price.Quote:
Originally Posted by behappier
You can also use 2 gfx cards to provide display for 4 monitors.
Nvidia and ATI have each recently announced that their graphics cards can function as Physics Processing Units (PPUs). On a SLI/Crossfire mobo, one graphics care does the traditional graphics processing, while the other graphics card does the physics processing. That announcement changes the picture dramatically.
First, it strongly suggests an answer to my original question: Do these SLI/Crossfire mobo/drivers lock-in that second graphics port so it can be used only for graphics? Answer: No. The second graphics slot can function for something else other than graphics. It can function as a PPU, for example.
Second, the announcement means there will be a four-way split for the market-share on physics processing:
- Special-purpose physics PPU cards
- Nvidia graphics cards used as PPU
- ATI graphics cards used as PPU
- Perform the physics processing within other cores of a multi-core CPU chip
Each of those methods is likely to have it's own software interface, and game developers will have to write their software for a specific one of those. At least at first. So there will be a four-way battle for who dominates the market. (Perhaps there will eventually emerge a standard for the PPU interface? Or perhaps that will be implausible, considering these radically different approaches?)
At any rate, the four-way split of the marketplace, might also mean a slower emergence of PPU processing. Any thoughts?
A well written API could take any of these 4 cases into account, but the code may not be optimised to its fullest this way :(
I still think SLI PPU is pretty silly.. the overhead to get it upgraded from time to time is just toom high...
Its cool when you think its a free feature if u have SLI, but when ur upgrading later.. you'd have to chuck 2 cards and get another 2 in order to get the PPU feature working again..
Contrast to SLI.. a dedicated PPU unit solution will free ur GPU from liabilities later when upgrading as you're upgrading each separately..
Unless Nvidia and ATi can bind both PPU unit and a GPU unit into a single card solution with SLI only as a performance complement of some sort.. This is actually a reminescence of how 2d and 3d cards were during the Voodoo days.. 3D cards are totally separate entities.. If the big guys could ideally swallow Ageia and get them to work with NV or ATI guys to integrate PPU into the card themselves... But thats just wishful thinking..
Nvidia and ATI are saying that for physics processing, the two graphics cards can be different (and not even in the same family!). That is, one graphics card does the physics processing, and the other card (even from a different family) does the graphics processing. If they can dis-connect the two cards that way, then the two cards don't have to be upgraded at the same time. To upgrade you would de-commission your worst graphics card; turn your better graphics card into a physics processor (PPU); and buy one of the latest graphics cards to be your monster graphics processor. That actually doesn't sound too bad ... if it works. The doubt is whether a graphics card can do the physics processing well enough to stay in the game (pun).Quote:
Originally Posted by sawyen
I wouldn't be surprised if Nvidia or ATI eventually buys Ageia (the maker of PhysX cards).
Doesn't physics processing rely on GFlops of processing data?
Therefore ATI/Nvidia won't need to buy Ageia as they already have cards that can do the same thing far faster :)Quote:
Originally Posted by ATI's Will Willis