So, do we know anything about RV670 yet?

[3dilettante]

I don't know if they even be able to fit the pads if it were a straight shrink.
The pins seem awfully crowded as-is.

They would be changing something pretty significant to either chop the pin-out or make use of the otherwise wasted die space.

 

[cadaveca]

Well, keeping in mind that 55nm is 65nm optical shrink, then it's quite obvious that this is either an entirely new gpu based off of the earlier 65nm Rx6xx-series gpu's, or jsut a copy of said gpu's. Given the perforamcne gap between HD2600XT and HD2900XT, you can feel pretty confident in betting that this gpu's performance is going to be somewhere in between those two gpus.

Given that gpu pinout would be more dense, a cut-down R600 could be possible, but I really question the feasibility of such a product, both to meet market needs(no need, IMHO), and in financial/labour costs. But some extra silicon to deal with leakage and provide more landings seems ok to me...

Ill-fated "UFO" PCB gets a new gpu? Nah, 12-inch OEM cards are using this pcb.

That means either this gpu is intended to use HD2600XT extended pcb...or there really is some poor, poor planning going on.

 

[Apocros]

My idea for the refresh part would be R600 on the 65nm(hopefully with drastically reduced power leakage) ...

on 55G (or even 65G, really), reduced static current comes for free, relative to 80HS.

 

[Arun]

R600 is a 420mm2 chip. Let us be pessimistic and presume that scaling is only 60%, that 10mm2 is analogue, and that UVD(/AVP?) adds an extra 10mm2 on 55nm.

From this, we calculate: ((420-10)*0.6)+10+10 = 266mm2. This is a relatively typical and realistic die size for a 256-bit GPU, and note that this is IMO a worst-case scenario. If you presume that the scaling is better and that you might save some space from having a smaller memory controller, you'd easily get to 210-240mm2.

All I'm saying is that I think R600 on 55nm would be a perfectly viable mid-end GPU. The die size is perfectly reasonable, and comparable to RV570's. Regarding the toolset, it would obviously be more based on the RV610/RV630 work than the R600's. So there is no real waste there considering what they already had to do.

N.B.: I'm not saying RV670 is basically R600 on 55nm. The ratios could be different. It might be slower or faster per clock. It could also be on 65nm. A lot of possibilities exist. But this one is far from impossible, and should not be neglected.

 

[Geo]

RV670 CANNOT be a R600 shrink. That would be some of the most stupid planning I have ever heard of.

Completely new toolset for a refresh? I thought ATI commited to no refresh by going 80nm in the first place?(90nm refresh should be 80nm, and 65nm refresh should be 55nm.)

The name should tell you all you need to know about what RV670 is, I'd think.

 

[cadaveca]

NO kidding. I get it...but obviously some don't. Guessing is more fun, anyway.



"RV"

 

[nicolasb]

The fact that it's called RV670 doesn't tell us whether it's 65nm or 55nm. There's plenty to speculate about.

 

[vertex_shader]

The fact that it's called RV670 doesn't tell us whether it's 65nm or 55nm. There's plenty to speculate about.

When its 55nm than won't coming out this year.
My quess is mid Q4 at 65nm.

 

[nicolasb]

A little more from FUDzilla:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1924&Itemid=34

Our sources confirmed that the original design of the chip was suppose to end up at 65 nanometre but the final version of the chip will be out as a 55 nanometre. The chip will draw less power and will be physically smaller.

It is much easier to put two 55 nanometre chip than two 65 as the card will need less power and that is the plan.

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1923&Itemid=1

Despites all kinds of rumors around, we can confirm that ATI won't be releasing any Rx6xx chips this year. The company will try to finish RV670 this year, but as we've said several times before, this chip is going to show its face in early 2008.

ATI's only hope is to put two RV670 cards together and such a dual chip card should be able to take the performance crown away from Nvidia's G80 Ultra. The biggest problem is that at the time RV670 comes to reality, Nvidia is supposed to have the next generation card ready, it is called G92 by some.

Well, ATI don't have any other choice as its real next generation, the R700 won't be out until Q2 2008 and this is the best case scenario, somehting that ATI usually fails to achieve.

 

[3dilettante]

What is the history with half-nodes when it comes to power draw (besides R600's rather leaky half-node)?

Every statement I've seen about half nodes is that transistor performance is pretty much the same, and that the only real difference was the density.

I am unaware of any design directly ported to a half-node, so I don't know of any comparison.

 

[vertex_shader]

A little more from FUDzilla:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=1924&Itemid=34



http://www.fudzilla.com/index.php?option=com_content&task=view&id=1923&Itemid=1

65nm rv670 ready and not releasing in Q4 because they want to shrink its the stupid thing i ever heard.
If AMD coming out Q4 with the 65nm version, still can shrink the GPU later, and can earn some money in the holiday season, i think Fuad walking in the dark with this rv670 story :smile:

Hopefully rv670 based cards not following the rv630 line with the motto: "not enough fast to compete, but cheaper than nv cards".

 

[nicolasb]

65nm rv670 ready and not releasing in Q4 because they want to shrink its the stupid thing i ever heard.

I think he simply meant that ATI originally intended RV670 to be a 65nm product but have since changed their minds; is that really such an outrageous claim?

 

[vertex_shader]

I think he simply meant that ATI originally intended RV670 to be a 65nm product but have since changed their minds; is that really such an outrageous claim?

The time is not on AMD side now to jump forward, they have no card between 150$<->400$ so releasing a whole quarter later make no sense if everything ok with rv670.

 

[Tim]

R600 is a 420mm2 chip. Let us be pessimistic and presume that scaling is only 60%, that 10mm2 is analogue, and that UVD(/AVP?) adds an extra 10mm2 on 55nm.

UVD is actually 4.7mm2 on 65nm. (according to AMD at the analyst day), that would make it about 3.5mm2 at 55nm.

 

[JoshMST]

From my understanding, the half nodes have typically improved upon speed and power consumption, but not in any huge or generational way. The reason I hear is that the optical shrink does nothing, but the underlying process tech that was based on the older process incorporates all of the advances and fixes of that older process. So in the end there do appear to be improvements, but it is far shy of the full node changes.

 

[Vincent]

RV670 VS G92


ATI should rule out this chip quickly .otherwise, ATI will face another problem

G92GX2

 

[Sound_Card]

That's the thing. Some say G92 is a perfomance part, then other people say it's Nvidia's new flagship. So honestly I don't even know if we could compare RV670 to G92.

Anyways, some tiny little tid bits were released about R700 by AMD. From the wording of the PR slide, seems that they have performance per watt in mind for the chip. Seems AMD might have a major influence on R700.

 

[Die_Hard_AMD]

Hi B3D forum users!
Some add-on letter to latest 2008 "Leo" AMD platform:

 

[Sound_Card]

Hi B3D forum users!
Some add-on letter to latest 2008 "Leo" AMD platform:

I'm confused now. Sigh...

If R680 is going to launch with the Leo platform in 2008, then we won't see R700 untill at best summer 2008.

Its scary to think that G92 is comming in november as well as new nforce socket AM2+ chipsets and AMD wont relase R680/R670 and RD790/SB700 untill 2008. It's also scary to think that 2008 is such a general term.

 

[AnarchX]

HD 2900Pro based on R600?

"ATI Radeon HD 2900 PRO" = ati2mtag_R600, PCI\VEN_1002&DEV_9403

http://www.chilehardware.com/foro/amd-hd-2900-t106198.html