Not open for further replies.


cant edit
those mockup pictures showing the slave chips are probably plain wrong, it was just a guess anyways. I just wish they would stop showing them over and over again. :/

the dualchip MCM aproach sounds a lot more plausible at this point


Nov 19, 2007
dont know how any one can say its not true, just have to look at the 4870, no one thought it would have 800SPs and look how that turned out, if any one can pull this off it would be ATI


yeah, the rumor with slave chips and stuff is fanboi speculation, but the post has other specs that rather contradict the second part of the same post and nordichardware sort of agrees with that part of the post.... number of shaders and mcm design... but yeah, lots of salt needed :p
Although, using clamshell mode at 16 bit, with either sideport and or hydra? Its speculation, but, if the ram is there for usage, if sideport and or hydra can converge multi core output... who knows


Sep 29, 2007
remember how the 4870 X2 was supposed to work as 1 gpu... yet it didn't

Honestly lets stop trying to fill our holes with multi GPUs, Last I remember multi GPU was something to extend the life of your product not to make it top of the line:)

I'll be impressed if we can see the 5870 or the 300 GTX do better than 1 295 GTX...

I remember seeing someting like this in sept...honestly I don't beleive 1 bit, neither do I beleive that the 300 GTX will be 3 times stronger than the 280 GTX lol.

:D but nice find. Puts faith Back in AMD:) We'll see what card AMD will come out with next:D
Actually nVidia should have been in a better position to build a modular architecture like that. They bought 3DFX some time ago which was already laying the ground work for that kind of modular card building with its sage and ravage chips. Then again the last time nVidia incorporated some of the 3DFX technology they ended up with the 5800 :D.


The photos I don't believe at all. I also agree that the specs aren't really any better then the 4870 series. 960 to 1000 shaders? The 4870 already has 800, why such a low increase? Bumping up the GDDR5 frequencies is nice, but without a large bump in shaders/shader frequency most of it will be wasted. Core clocks aren't really any higher either. If the 5870 is simply a few more shaders running a hair faster, I won't be impressed.
OK, heres a rumor list
The 740 wont be the only card coming soon at 40nm
When I said GDDR5 can run in Clamshell mode, it typically has run in 32 bit, but at 16 bit you can use 2 devices on it, and if the speeds there, which it is, then you can either do 1 or two things. 1. Use only half the memory in a x2 solution (most likely) or use 2 cores sreamed together with a hydra type controller, as itll be seen as 1 core for the driver.

Thats a few rumors about. Theres alot more as well, but its currently too far away yet to cement the most likely possibilities.

Alot of people assumed 2 cores as 1 with GDDR5, but it may be that to have a 1gig x2 card, all youll need is 1gig of GDDR5, run in clamshell mode. All those costs of GDDR5 may actually be a savings down the road.

Looking at the hydra/sideport scenario, its a possibility, so dont scratch it off the rumors list just yet
Ever since before the RV770 came out there have been rumours that ATI were gunning for a modular design. It makes perfect sence from a business point of view and on that alone it shouldnt be disregarded.
I honestly do think its comming, dont know when and i wouldnt think its this time around as i would have expected better from the 4 series X2 if they were anywhere near getting the cores to play nice together.
Personally, and to be clear this is all personal speculation, But i would have thought that a X2 that runs as a single GPU would be first. Or are we saying that these cards would need to run in tri or quad mode as far as the drivers are concerned ?
Your thoughts please
According to rumor, using hydra and or sideprot, it gives the data stream the appearence of 1 core, whether its 2 or 4, then sends to driver. Rumor has it that it could be on die, or pcb, working with hydra, and using sideport for BW. Its not totally understood at this point, but thats close to what Ive seen. Look up hydra, then go by ATIs comments on the side port, then clamshell mode w/GDDR5, keeping the bus under control, with the rams speeds, and its multi device usage.

Remember, these new cards are all 40nm, so certain things need to be done for sizing, due to pad sizes etc, plus, it also makes sense in that without having to sacrifice speed going smaller on overall die size, no hot spots etc, and also, think about 4 cores, and then wanting 1Gig usage for it. Today wed need 4 gigs to meet that, and itd be a costly scenario as well
Thanks for the places to link to JDJ.
Yes the first thing that struck me was oh good we can actually have use of the ram we have and not half it.
I will look at the articles you sugested, seems your saying the hydra chip will be on the GPU and not linking from the Motherboard which was the way i understood it to work.

Good point, so that kinda rules Hydra out then dosent it ? What with Intel coming into the market place why allow help to others? Hydra on a Intel motherboard or on the PCB of Larrabee ?
Or is that against the competition laws
Not open for further replies.