Will Multi-VPU require an ATI motherboard chipset?
- Thread starter trinibwoy
- Start date
eh, still doesn't answer my fundamental ATI MVP question: what about different feature sets?
caboosemoose
Regular
This is the original text:
HEXUS has heard from a number of sources about ATI's launch plans for their multi video processing unit touted as 'MVP'. It is no secret to HEXUS or other folk that these cards were always on the, er… cards as the success of NVIDIA with their SLi solution made it logical for ATI to want to get in on the ‘card on card’action, after all, what red blooded bloke wouldn’t like a bit of twin action now and again, eh?
Sadly, it seems that just like NVIDIA with SLi, ATI’s MVP solution will come with a few rigid pre-set conditions before you can enjoy the benefits… so not all that different from a bit of twin action after all… In order to run the ATI AMR solution you’ll first off need to have a compatible mainboard and then you’ll need two cards, but not just any card, oh no… read on.
Now, here’s the clever bit, the system will run with a master card and slave card. The slave card can be ANY ATI card on the market, which is just about as cool as absolute zero, but (and it’s a big BUT).. the master card, which will control the frame buffer, has to support MVP. The ATI solution runs on a tiled based approach - which has a minimum checkerboard size of 32 pixels square.
So now you’re thinking how cool this and you want one…. Well, erm.. the thing is right now there aren’t actually any of these cards available, so although ATI have done us all a massive favour by keeping some use in our old cards for us, we can’t actually go MVP right now. But despair not, as ATI will be the first to market with these cards and they hope that the AIB partners will follow suite.
In a superb retro throw back to the Voodoo 2 days of yore, the 2 cards will have to be connected via an external dongle which will enable the cards to work together in accelerated harmony. This external link, essentially another bit of wire hanging out the back of your machine, has raised more than one or two eyebrows here at HEXUS towers as ATI have been heard to regard NVIDIA's connector between their 2 cards as being a 'poor solution'. One has to ponder on the external dongle issue, though it obvious that this will allow backwards compatibility with older cards, you have to wonder if ATI were having a pop at NVIDIA's neater internal connector just to give us hacks something to write about? Those guys, eh?
Anyway, with the initial launch of the mainboard supporting AMR/MVP, ATI has been talking about bundling the master card with them - in order to stop consumers from getting confused about what is needed to run the multi-card solution. This could be a good thing, especially if they can keep the price down.
Whatever happens, we are more than curious about how ATI will deliver this very much so 'proprietary' standard, which appears to be different from the 'open standard' talked about previously.
HEXUS has heard from a number of sources about ATI's launch plans for their multi video processing unit touted as 'MVP'. It is no secret to HEXUS or other folk that these cards were always on the, er… cards as the success of NVIDIA with their SLi solution made it logical for ATI to want to get in on the ‘card on card’action, after all, what red blooded bloke wouldn’t like a bit of twin action now and again, eh?
Sadly, it seems that just like NVIDIA with SLi, ATI’s MVP solution will come with a few rigid pre-set conditions before you can enjoy the benefits… so not all that different from a bit of twin action after all… In order to run the ATI AMR solution you’ll first off need to have a compatible mainboard and then you’ll need two cards, but not just any card, oh no… read on.
Now, here’s the clever bit, the system will run with a master card and slave card. The slave card can be ANY ATI card on the market, which is just about as cool as absolute zero, but (and it’s a big BUT).. the master card, which will control the frame buffer, has to support MVP. The ATI solution runs on a tiled based approach - which has a minimum checkerboard size of 32 pixels square.
So now you’re thinking how cool this and you want one…. Well, erm.. the thing is right now there aren’t actually any of these cards available, so although ATI have done us all a massive favour by keeping some use in our old cards for us, we can’t actually go MVP right now. But despair not, as ATI will be the first to market with these cards and they hope that the AIB partners will follow suite.
In a superb retro throw back to the Voodoo 2 days of yore, the 2 cards will have to be connected via an external dongle which will enable the cards to work together in accelerated harmony. This external link, essentially another bit of wire hanging out the back of your machine, has raised more than one or two eyebrows here at HEXUS towers as ATI have been heard to regard NVIDIA's connector between their 2 cards as being a 'poor solution'. One has to ponder on the external dongle issue, though it obvious that this will allow backwards compatibility with older cards, you have to wonder if ATI were having a pop at NVIDIA's neater internal connector just to give us hacks something to write about? Those guys, eh?
Anyway, with the initial launch of the mainboard supporting AMR/MVP, ATI has been talking about bundling the master card with them - in order to stop consumers from getting confused about what is needed to run the multi-card solution. This could be a good thing, especially if they can keep the price down.
Whatever happens, we are more than curious about how ATI will deliver this very much so 'proprietary' standard, which appears to be different from the 'open standard' talked about previously.
caboosemoose said:
One cool thing about the V2 SLI solution was that no mobo circuitry external to the V2's was required. You could run SLI on any mobo with the required number of open PCI slots, IIRC. The rear cable was not a "dongle" but rather the communications link required to synchronize the V2's to operate as if they were but a single card.
How on earth might Hexus imagine that coordination of the type necessary for SLI to work might be achieved *without* directly linking the two cards together in some fashion? SLI, whether then or now, is not a single card but rather is two separate cards coordinated and synced to operate together as if they were a single card. IE, without direct communication between the two cards the transparent operation of the pair of cards acting as a single card would become impossible. The idea of using standard system and bus and cpu resources to accomplish the synchronization in software is, I think, exceedingly poor, since you'd be more or less cutting off your nose to spite your face in terms of performance--even if you could get it to work, imo.
Also, I have never "heard" or read an ATi comment that in any way indicated that connecting two cards directly together for SLI operation might be a "poor solution"...
Again, without such linkeage transparent operation of the pair of cards as a single card would become impossible. I think that what Hexus may actually have heard and misconstrued is ATi's reflection on the propietary nature of nV's SLI being a "poor solution" in that it requires proprietary mobo circuitry external to the pair of 3d cards operating in SLI. At least, this would be my guess...I think that people are missing the obvious--that SLI is not and never will be a "single card" but is rather *two cards* linked and synced to operate as a single card.
caboosemoose
Regular
Well, I posted this on the Hexus forum:
Interesting story, but I'll go out on a limb and suggest that large parts of it don't seem to make much sense. The implications of AMR 1. being compatible with existing cards via an external dongle and 2. using tile-based rendering are, umm, a bit challenging. For starters it means the cards will be connected by either VGA, or even more bizarrely, DVI. On the old 3DFX SLI, VGA made sense because it was a *relatively* straight forward task to interleave the outputted scan lines. But if you think about trying to interleave a bunch of pre-rendered tiles outpputed by VGA or DVI, well it sounds like a recipy for latency and image artefacts to me. As to how you would intertleave the DVI output from the slave card into the DVI output of the master, I have no idea. It seems like a very clunky way of going about things, but then what the hell do I know....etc etc etc
Overall, if ATI's multi card is tech is like Hexus say (and I have at least some reason to believe it is) then I suspect it is a stop-gap solution designed to allow ATI to tick that multi-card box in the short run while they work on a more sophisticated solution in the longer term.
Interesting story, but I'll go out on a limb and suggest that large parts of it don't seem to make much sense. The implications of AMR 1. being compatible with existing cards via an external dongle and 2. using tile-based rendering are, umm, a bit challenging. For starters it means the cards will be connected by either VGA, or even more bizarrely, DVI. On the old 3DFX SLI, VGA made sense because it was a *relatively* straight forward task to interleave the outputted scan lines. But if you think about trying to interleave a bunch of pre-rendered tiles outpputed by VGA or DVI, well it sounds like a recipy for latency and image artefacts to me. As to how you would intertleave the DVI output from the slave card into the DVI output of the master, I have no idea. It seems like a very clunky way of going about things, but then what the hell do I know....etc etc etc
Overall, if ATI's multi card is tech is like Hexus say (and I have at least some reason to believe it is) then I suspect it is a stop-gap solution designed to allow ATI to tick that multi-card box in the short run while they work on a more sophisticated solution in the longer term.
caboosemoose said:
DVI gives a digital output. If the hardware is capable then it could be read in from the slave and composited, digitally, with the image from the master board.
Well, this is "new information", so presumably the thinking behind it is that this wouldn't loose any performance from passing the image over the bus and you'd still have the benefits of using the existing board.
caboosemoose
Regular
DaveBaumann said:
Sure, but it still strikes me as a rather low-tech way of going about things. It's a bit of a messy hack if you ask me!
StellaArtois
Newcomer
So - if you're trying to run two monitors, you're gonna be screwed* (if the passthrough connector is taking up a DVI slot on each card)?
*Well, you'll at least end up having to swap cables between running a dual monitor desktop, and single screen games?
Kludgey IMHO, if this info is true.
*Well, you'll at least end up having to swap cables between running a dual monitor desktop, and single screen games?
Kludgey IMHO, if this info is true.
trinibwoy said:
Not very, according to this: http://www.cooltechzone.com/index.php?option=content&task=view&id=1255
I found this part interesting:
Be interesting to see how many profiles they ship with.
This article also denies the dongle thing from the HEXUS piece.
I found this part interesting -
To me, this makes absolutely no sense. First off, why would ATI provide such flexibility only to hobble the newer card. Also, equalizing the clock speeds doesn't say squat since there are other factors at play. What if the slower card has a higher clock speed
If ATI does support different cards working in tandem I'm positive they aren't going to handicap the faster hardware especially if it's true that you will be able to match up a very fast discrete card with integrated mobo graphics. Some sort of load-balancing would be the only reasonable solution.
To me, this makes absolutely no sense. First off, why would ATI provide such flexibility only to hobble the newer card. Also, equalizing the clock speeds doesn't say squat since there are other factors at play. What if the slower card has a higher clock speed
If ATI does support different cards working in tandem I'm positive they aren't going to handicap the faster hardware especially if it's true that you will be able to match up a very fast discrete card with integrated mobo graphics. Some sort of load-balancing would be the only reasonable solution.
Jawed
Legend
http://www.beyond3d.com/forum/viewtopic.php?p=507201#507201
But that doesn't answer the issue of memory bandwidth.
I honestly think this idea that the driver would throttle the faster card's clock to match the slower card's clock is nonsense.
Jawed
But that doesn't answer the issue of memory bandwidth.
I honestly think this idea that the driver would throttle the faster card's clock to match the slower card's clock is nonsense.
Jawed
Similar threads
