So, do we know anything about RV670 yet?

[ShaidarHaran]

?

If someone has posted I worthy news link, I try to share it.

I was giving you a hard time.

Anyways, I think I know where he got the prank from. Fudo however makes the joke sound so corny. He kind of kills it for me...

It's always disappointing to see knowledgeable people giving that hack any credit. I wish people would stop reading his "news" altogether. It's not like he has independent sources or anything anyway. All he does is repost stuff from here and VR-Zone and HKEPC and the like.

 

[w0mbat]

"RV620" = ati2mtag_RV610, PCI\VEN_1002&DEV_95C0
"RV620 " = ati2mtag_RV610, PCI\VEN_1002&DEV_95C5
"RV620 " = ati2mtag_RV610, PCI\VEN_1002&DEV_95C7
"RV635" = ati2mtag_RV630, PCI\VEN_1002&DEV_9590
"RV635 " = ati2mtag_RV630, PCI\VEN_1002&DEV_9596
"RV635 " = ati2mtag_RV630, PCI\VEN_1002&DEV_9597
"RV635 " = ati2mtag_RV630, PCI\VEN_1002&DEV_9598
"RV635 " = ati2mtag_RV630, PCI\VEN_1002&DEV_9599
"RV670" = ati2mtag_RV630, PCI\VEN_1002&DEV_9501
"RV670 " = ati2mtag_RV630, PCI\VEN_1002&DEV_9505
"RV670 " = ati2mtag_RV630, PCI\VEN_1002&DEV_9507
"RV670 " = ati2mtag_RV630, PCI\VEN_1002&DEV_9500
"R680" = ati2mtag_RV630, PCI\VEN_1002&DEV_950F

R680? Just a new name for the dual-RV670 like R670 or a new high-end chip?

 

[Kowan]

Oh... hope it's a high end. :smile:

 

[nicolasb]

FUDzilla:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=3481&Itemid=1

We got our hands on what looks like the final specificationd of the RV670XT card. This one will run at an impressive 825MHz for the engine and 2,400MHz for the memory. It has 320 Stream processors just as many as the R600, but it's limited by the 256-bit memory controller.

It is a DirectX 10.1, PCIe 2.0 card with Shader model 4.1 support and it's a 55 nanometre design. This will be the first graphics chip ever to be produced at 55 nanometre.

The Radeon HD 2950XT card will only use GDDR4 memory, while the 2950PRO will end up with GDDR3 at some modest speed. This GPU is planned for introduction in November and ATI's partners are really excited about it.

 

[hoom]

Hmm, I just worked out what has been subconsciously freaking me out about RV670
We haven't even heard rumor of 45nm AMD CPUs & even Intel is only just starting to deliver 45nm yet ATI is about to launch a big 55nm GPU

 

[Kaotik]

Hmm, I just worked out what has been subconsciously freaking me out about RV670
We haven't even heard rumor of 45nm AMD CPUs & even Intel is only just starting to deliver 45nm yet ATI is about to launch a big 55nm GPU

Remember that 55nm is just optical shrink of 65nm, and 65nm CPU's have been around for ages.
IIRC there was some talk too, that CPU manufacturers just don't tell about the half-nodes, but still are using them, as in - some of the 65nm CPUs might be 55nm if they were in the GPU-world where people tell about half-nodes

 

[hoom]

Nonetheless, we are in the situation where officially AMDs smallest process chip is not a CPU but a GPU.

 

[trinibwoy]

Nonetheless, we are in the situation where officially AMDs smallest process chip is not a CPU but a GPU.

How do you know that they don't have 55nm CPU's? Did you just completely ignore the post above yours? :smile:

Just because they don't market it as such it doesn't mean that 55nm CPU's are "unofficial".

 

[INKster]

How do you know that they don't have 55nm CPU's? Did you just completely ignore the post above yours? :smile:

Just because they don't market it as such it doesn't mean that 55nm CPU's are "unofficial".

Even the designations ("65nm", "45nm", etc) are merely an aproximate average of the size of all transistors.
In a 65nm chip there are some 90nm features, some 45nm ones, etc.

 

[hoom]

Show me anywhere that states officially that AMD has 55nm CPUs & I'll concede the point

 

[Farhan]

Even the designations ("65nm", "45nm", etc) are merely an aproximate average of the size of all transistors.
In a 65nm chip there are some 90nm features, some 45nm ones, etc.

I think it's supposed to be half of a minimum width polysilicon (metal now in intel's 45nm process) line, which determines minimum gate length to a large extent.

edit: i doubt anyone strictly follows this rule... lots of these process tech numbers you see are fudged.

 

[lik]

How do you know that they don't have 55nm CPU's? Did you just completely ignore the post above yours? :smile:

Just because they don't market it as such it doesn't mean that 55nm CPU's are "unofficial".

55nm is not a standard node. It is a half node adopted by TSMC. AMD and Intel's own fabs do not have such concept of half node, and do not fab chips at such half node. Actually if you talk to CPU guys, lots of them do not have any idea of 55nm technology, because they don't pay attention to TSMC.

Unless AMD let TSMC fab its CPUs, which is not possible due to a number of issues. But that is beyond the topic of this thread.

 

[dizietsma]

55nm is not a standard node. It is a half node adopted by TSMC. AMD and Intel's own fabs do not have such concept of half node, and do not fab chips at such half node. Actually if you talk to CPU guys, lots of them do not have any idea of 55nm technology, because they don't pay attention to TSMC.

Good summation. That's not a bad point on TSMC though, more the other cpu guys.

 

[3dilettante]

The half R600 memory bus makes me wonder how close a dual RV670 package would be pin-out wise.

The MCM as a whole would have the same number of memory pins.
It could have the same number of PCI-E pins, as I don't see what good it would do to give each chip an interface.

Putting power and ground aside, I wonder if there are that many pins that would require duplication in an MCM.

As for power and ground, RV670's rumored being half the TDP of R600.
That doesn't necessarily mean that one RV670 has half the power and ground pins, and it might not work to share.

Smaller processes tend to like cleaner current and better amperage when it comes to getting higher clocks, which favors more pins.

If this is not a problem, it might be that a 2xRV670 MCM will be mostly pin-compatible with current R600 PCBs.

It might be mechanically larger, but perhaps some PCB design R&D costs can be saved.

 

[trinibwoy]

55nm is not a standard node. It is a half node adopted by TSMC. AMD and Intel's own fabs do not have such concept of half node, and do not fab chips at such half node. Actually if you talk to CPU guys, lots of them do not have any idea of 55nm technology, because they don't pay attention to TSMC.

I think you should tell that to Kaotik, not me My post was in context of his earlier statement.

 

[AnarchX]

If this is not a problem, it might be that a 2xRV670 MCM will be mostly pin-compatible with current R600 PCBs.

It might be mechanically larger, but perhaps some PCB design R&D costs can be saved.

That I thought too, also the name of Dual-RV670 (R670 or R680) indicates that it is a bit more than a RV670-Gemini.

The HD2900XT-cooler with a little modification for the 2 Dies would be enough for two RV670 with 2x62W TDP, since R600XT-GPU had a TDP of 140W.

Maybe of that Dual-RV670 would not need the expensive PLX-PCIe-bridge, like the previous Geminis.

 

[hoom]

Regarding the 256bit memory, any guesses as to whether its 4 ring stops or just two R600 style ones?

 

[Jawed]

Regarding the 256bit memory, any guesses as to whether its 4 ring stops or just two R600 style ones?

Prolly 4 ring stops, since this is would be the "same configuration" as R580. Each ring stop would have an associated ALU/TEX/RBE (80,4,4) unit associated with it, I reckon.

I wouldn't be surprised if we see the return of 32-bit memory channels, too.

Jawed

 

[turtle]

That I thought too, also the name of Dual-RV670 (R670 or R680) indicates that it is a bit more than a RV670-Gemini.

The HD2900XT-cooler with a little modification for the 2 Dies would be enough for two RV670 with 2x62W TDP, since R600XT-GPU had a TDP of 140W.

Maybe of that Dual-RV670 would not need the expensive PLX-PCIe-bridge, like the previous Geminis.

A kentsfield-esque product? That would be quite interesting...

 

[ShaidarHaran]

I wonder how many people will think the above is real, and how quickly it will spread around the intarweb.