AMD confirms R680 is two chips on one board

[swaaye]

So, there is no R680 GPU? R680 means RV670XT x 2?

BTW, here's a couple of reviews of a dual HD2600 board that uses a similar PLX chip to communicate.
http://www.tweaktown.com/articles/1167/gecube_s_dual_gpu_gemini_3_sneak_peek/index.html
http://www.hexus.net/content/item.php?item=10349&page=1

 

[PsychoZA]

What system/chip did the 7950GX2 use to communicate?

 

[Arnold Beckenbauer]

What system/chip did the 7950GX2 use to communicate?

http://www.clubic.com/afficher-en-plein-ecran-298700.html
Nvidia's own PCIe switch.

 

[PsychoZA]

http://www.clubic.com/afficher-en-plein-ecran-298700.html
Nvidia's own PCIe switch.

Thanks.

 

[turtle]

http://www.clubic.com/afficher-en-plein-ecran-298700.html
Nvidia's own PCIe switch.

Yep, the same one we'll prolly see on the 8800gx2(s).


As for the topic of this thread...I hung on to the hope of a MCM package dual core RV680 for a long time...but the first pic that showed the back with two distinctive sets of pins snapped me into reality. That was quite a while ago though.

At least we still have R700 to look forward to, so it will happen.

Question is, could there be a R700 derivative with two packages of 4 cores each controlled by only one controller chip (that might include UVD etc), if such thing exists?

That would be the pentultimation (Dammit firefox, don't tell me that's not a word) of crossfire, or multi gpu for that matter. Two packages that share resources the same as a single package would...Could that transfer over to a multiple card design if the former is possible? One card's hub could surrender to the other, or would physical traces be needed from the memory to each package and would the packages need to be linked via a trace to the central hub controller chip?

 

[satein]

http://www.clubic.com/afficher-en-plein-ecran-298700.html
Nvidia's own PCIe switch.

It looks like a magic chip that make the Tri-Sli work with N-Force 680 chipset too However, I cannot remember where I saw the picture of nake 3 chips on the Tri-sli 680 motherboard

 

[Arun]

No, that picture shows a BR03 unless I'm misreading it. The chip on the 780i (and 750i, sigh) is a BR04, which will likely also be used on the 2xG92 config. It's basically the same thing as BR03 except that it's PCI Express 2.0, though. Based on an old picture, I estimated die size to be 43mm², presumably on 65nm.

OT, but my expectation is that we'll see BR04 being sold by NVIDIA to motherboard manufacturers wanting to use SLI on Intel southbridges for Nehalem's Socket 1160. My expectation is that we will also see a new bridge on 45nm that includes a G98-class GPU for Hybrid SLI; that chip could aslo include a QuickPath interconnect and be used as NVIDIA's Socket 1366 northbridge.

As for R680, I very much doubt that using a PLX chip is a significant cost disadvantage, especially given the tape-out costs.

 

[satein]

No, that picture shows a BR03 unless I'm misreading it. The chip on the 780i (and 750i, sigh) is a BR04, which will likely also be used on the 2xG92 config. It's basically the same thing as BR03 except that it's PCI Express 2.0, though. Based on an old picture, I estimated die size to be 43mm², presumably on 65nm.

OT, but my expectation is that we'll see BR04 being sold by NVIDIA to motherboard manufacturers wanting to use SLI on Intel southbridges for Nehalem's Socket 1160. My expectation is that we will also see a new bridge on 45nm that includes a G98-class GPU for Hybrid SLI; that chip could aslo include a QuickPath interconnect and be used as NVIDIA's Socket 1366 northbridge.

As for R680, I very much doubt that using a PLX chip is a significant cost disadvantage, especially given the tape-out costs.

Thank you for clearify me :smile:

So, if BR04 chip is supposed to provide the chance for Intel chipset to do SLi, there would be enough reason on building this chip on its own by NV

 

[Arun]

There's also the possibility that BR04 handles AGP, fwiw, but it isn't perfectly clear to me that it does.

 

[AnarchX]

AGP-bridge was AFAIK BR02 or?

 

[Arun]

Yes, but BR02 doesn't work with G8x/G9x iirc, so there were rumours NV was working on a new chip to replace it. Given the fact that an AGP interface ought to be pretty damn cheap, I wouldn't be surprised if they just included it in BR04 rather than tape-out another chip.

 

[AnarchX]

The rumors were that Revision A05 of BR02 will support G8x/G9x, but it seems it is not ready or partner are not interessted in AGP-part anymore.

But better stop this offtopic here... :smile:

 

[Jawed]

This says the PLX chip is a 6347

http://www.digitimes.com/mobos/a20071219PD206.html

though there's no such part. There is an 8547, which is 3-port (3x 16 lane):

http://www.plxtech.com/products/expresslane/pex8547.asp

which I guess is cheaper than the 9 port 8548 - I didn't notice there was a 3 port variant earlier

Jawed

 

[Jawed]

Winning Combination: Intel's P45 and ATI CrossFireX

This is kinda hilarious:

http://www.hkepc.com/?id=533

implying that 2xR680 CrossFireX will be marketed by Intel.

Jawed

 

[satein]

This says the PLX chip is a 6347

http://www.digitimes.com/mobos/a20071219PD206.html

though there's no such part. There is an 8547, which is 3-port (3x 16 lane):

http://www.plxtech.com/products/expresslane/pex8547.asp

which I guess is cheaper than the 9 port 8548 - I didn't notice there was a 3 port variant earlier

Jawed

Coincidently, at the end of digitimes article stated that ...

AMD is currently planning to integrate the PCI Express bridge chip into its future GPUs so that it does not need to adopt third-party's chips. This design is expected to appear in AMD's next generation R700 series, the sources added

Thus, the PCI-e switch chip will be only for the HD 3870X2 card .

 

[satein]

This is kinda hilarious:

http://www.hkepc.com/?id=533

implying that 2xR680 CrossFireX will be marketed by Intel.

Jawed

I think it is 'OK' for Intel to use the term 'ATi Crossfire' on their product at least to tell the market that their chipset can do multiple cards setup

 

[AnarchX]

The first product will be the Radeon HD 3870 X2 which will feature two RV670XT GPUs and will launch in January 2008 with a price set between US$299-349, noted the sources.

http://www.digitimes.com/mobos/a20071219PD206.html

First $799, then $399-499 and now this?

What the hell AMD tries to ruin themselves and the market?

 

[AlexV]

http://www.digitimes.com/mobos/a20071219PD206.html

First $799, then $399-499 and now this?

What the hell AMD tries to ruin themselves and the market?

At that price, if it's realistic, they'll move piles of these...and maybe that's what they need. The RV670 may just be cheap enough and have good enough yields to still bring money even at such a reduced price. And if they move piles of these, and get a significant market presence from enthusiast(3870X2) to part-time gamer(the 3850 and whatever'll show up under it), and come back into developers' eyes(instead of the current situation with nV working with everyone and ATi eventually patching their drivers, at some point far detached from the game's release), they'll likely turn things around on the graphics front.

 

[BRiT]

At that price, they will indeed move mountains of these things provided the drivers don't prove to be a let down. For their shareholders, I hope they're still maintaining reasonable margins.

 

[vertex_shader]

http://www.digitimes.com/mobos/a20071219PD206.html

First $799, then $399-499 and now this?

What the hell AMD tries to ruin themselves and the market?

Maybe they are finally realized this card not need to compete against the 8800ultra, so before everyone say DOA for this card with 500$ price tag, they drop the price to lower segment where it can be a better buy in some way.
This situation not new because we see this in the whole year, sadly looks like its not end in 2007