AMD: R8xx Speculation

[LordEC911]

That link claims 1.5x the ALUs, 2.4x the TMUs, and 2x the ROPs (all with ~1.1x the bandwidth). I haven't been following things too closely, but are we really expecting a lower ALU:TMU ratio?

Ummm... 1.2x TMUs...
48TMUs compared to RV770's 40TMUs.
With a 1.2x clock increase that is only 1.44x more texturing power.

But yes, that chart is old and they are trying to say that the source was right all along.

 

[Ailuros]

I'll burst if I don't say it: so much for RV7x0 being supposedly so "close" to X11 compliance and ATI supposedly having a relatively short path to it. If I look at the amount of units and the rumored die area under 40nm I don't see a "short path" as a layman but would of course like to stand corrected.

 

[Scrat]

That's missing a 64-bit chip, though.

Jawed

Is a new chip needed in that market segment? RV710 already has DX10.1, and it's all you need for the upcoming Windows 7. nVidia followed with their DX10.1 low end parts, Intel is still without it. I think those DX10.1 are enough to make Intel run for its money, no need for more R&D.

There's space for a shrink to 32bit + GDDR5 in the future, tho...

 

[ShaidarHaran]

I'll burst if I don't say it: so much for RV7x0 being supposedly so "close" to X11 compliance and ATI supposedly having a relatively short path to it. If I look at the amount of units and the rumored die area under 40nm I don't see a "short path" as a layman but would of course like to stand corrected.

Wait. What does unit count have to do with functionality?

 

[Davros]

anyone read this :
http://techcombo.com/the-future-of-graphics-cards

i dont understand this quote
"In comparison to nVidia, who hasn’t got any 40nm desktop cards or any with GDDR5 and a major ice breaker is the fact that their DX10 is a very stripped down version, more like DX9.d."

 

[neliz]

anyone read this :
http://techcombo.com/the-future-of-graphics-cards

i dont understand this quote
"In comparison to nVidia, who hasn’t got any 40nm desktop cards or any with GDDR5 and a major ice breaker is the fact that their DX10 is a very stripped down version, more like DX9.d."

One big piece of fanboy poo-poo.

 

[Ailuros]

Wait. What does unit count have to do with functionality?

I wasn't thinking of unit count at all since I haven't the vaguest idea yet what each unit is capable of. Rather rumored die area and from that a very roughly speculated transistor count.

 

[Kaotik]

anyone read this :
http://techcombo.com/the-future-of-graphics-cards

i dont understand this quote
"In comparison to nVidia, who hasn’t got any 40nm desktop cards or any with GDDR5 and a major ice breaker is the fact that their DX10 is a very stripped down version, more like DX9.d."

I assume you mean the DX10 part, and tbh I don't get it either, only thing coming to mind would be the fact that their geometry shaders at least first seemed more like checkbox feature, extremely slow and so on, on the other hand, the chips are almost DX10.1 anyway, they miss some features, don't know exactly which, but they support quite few of them too, you just need to use NVAPI to use them (I think FarCry2 uses NVAPI on NV cards to get some of the DX10.1 features up and running)

 

[entity279]

One big piece of fanboy poo-poo.

= the author lacks fundamental knowledge about the subject he's covering. This is particullary amusing:

ATI on the other hand are slightly better as their RV770 uses ~30 of it’s core for GPU tasks, O.K granted it’s still not much but with ATI it’s understandable, the whole fusion project and all

 

[neliz]

= the author lacks fundamental knowledge about the subject he's covering. This is particullary amusing:

I couldn't even make that out in his diarrhea of nonsense.

 

[Ailuros]

Uhmmm why are we even paying attention to that kind of rubbish again?

 

[ShaidarHaran]

I wasn't thinking of unit count at all since I haven't the vaguest idea yet what each unit is capable of. Rather rumored die area and from that a very roughly speculated transistor count.

I see. Still tough to say how much difference there is between generations without knowing what those extra trannies count for though, wouldn't you say?

 

[digitalwanderer]

Uhmmm why are we even paying attention to that kind of rubbish again?

Because it's silly season, and that is what we do during silly season.

 

[LordEC911]

Just a little bit more on the codenames matching with die-sizes, plus this whole MCM thing is still going strong.

----- AMD RV8XX gossip on some of the actual situation

September 17 in San Francisco issued a formal version of a full range of evergreen.
AMD DX11 graphics chips into high-end, performance, mainstream, entry-四档, high-end, code-named Cypress, the performance of code
Codenamed Redwood, are the mainstream of the two code-named Juniper and Cedar, entry, code-named Hemlock.

And flagship product Cypress uses the MCM design, is composed of 2 Redwood, Redwood chip size 300 square millimeters,
Juniper chip size of 181 mm2, Cedar die size 120 mm2. Currently, AMD DX11 performance and mainstream Products has proceeded smoothly, but the flagship product in the chip package Cypress encountered a little problem, Cypress
May be deferred to the AMD DX11 performance, after the mainstream market

http://www.forum-3dcenter.org/vbulletin/showpost.php?p=7455220&postcount=1884

 

[hoom]

I can't believe the idea that there are 3 different sized chips.

Totally breaks the sweet spot strategy. Unless, somehow its suddenly become much easier to make smaller derivative chips?

 

[dkanter]

Do you guys really think they are doing an MCM, instead of packaging two dice in the same PCB?

It seems like from a power and heat perspective, the latter is more attractive, since you could put the two GPUs a few inches away from each other to avoid having 400W in the same approximate area. You also get to spread out your VRMs, caps, etc. over more board area.

This is a marginally believable rumor, since Magny Cours will be using a MCP as well, so clearly the expertise exists inside AMD.

DK

 

[MfA]

Heatpipes are extremely efficient, once you get the heat into them a couple cm more or less doesn't really matter ... the distance between the dies is pretty much irrelevant for the temperature of the dies themselves, and I doubt the package substrate or the PCB is impacted by the hotspot.

 

[hkultala]

I can't believe the idea that there are 3 different sized chips.

Totally breaks the sweet spot strategy. Unless, somehow its suddenly become much easier to make smaller derivative chips?

1 chip for performance/high-end, 1 chip for mainstream, 1 chip for low-end.

They have been doing this for about 5 years.

 

[Pete]

Ummm... 1.2x TMUs...
48TMUs compared to RV770's 40TMUs.

The donanimhaber chart (a screengrab of Hexus' April piece, as AnarchX's link shows) shows 96 vs. 40 INT8, or 48 vs. 20 FP16. (I'm guessing you just glanced at the "INT8/48ppc" line.)

 

[LordEC911]

The donanimhaber chart (a screengrab of Hexus' April piece, as AnarchX's link shows) shows 96 vs. 40 INT8, or 48 vs. 20 FP16. (I'm guessing you just glanced at the "INT8/48ppc" line.)

No I was glancing at the Gtexels....
Edit- At the wrong columns... Awesome. My bad.