AMD: R8xx Speculation

[fellix]

Perhaps the problem is the lack of a "Future DX11 hardware speculation" thread?

Hm, not so distant future, as it seems.

 

[Blazkowicz]

Is there actually a market for a GT21x part?
Maybe if GT200 had bought DX10.1 but they already have low end DX 10 models out there.
I'd say 55nm G92b + their existing low end models (65nm must be pretty cheap & high yielding by now regardless of the bigger dies?) are adequate.

NV thus gets to save engineering resource for a bigger push on the next gen.

low end GT21x are still mildly useful to have on laptops (a huge market really).
I wouldn't spit on a AMD regor + GT216 or dual core VIA nano + GT218 if that's cheap and with a big battery life.

ditching all of the work on GT21x seems a bit cruel for the engineering teams isn' it?

 

[Ailuros]

Perhaps the problem is the lack of a "Future DX11 hardware speculation" thread?

Jokes aside that would be the typical B3D thread someone would expect; the problem then would be that it would lack in popularity compared to this and the G300 thread. I'm afraid football stadiums f.e. would have far more empty seats if it wouldn't be for fanatical followers....

 

[keritto]

It's doubtful that Nvidia will keep the same 4:1 MADD/SFU ratio. And 96+32 for Ati? You're proposing 6:2 MADD/SFU? That really doesn't make sense from a utilization perspective.

Sorry i don't get it how it's 4:1 ratio and not 1:1 ratio when half of SPs in the 1/8 (1/10) core are MADD and other half SFUs, all pumped up at least to double frequency of the i/o part of the chip??

And 3:1 seem quite fine to me for ATi in the light of growing tessellation needs. And to stream little bit more (dividable by DP-dual precision needs) than DX10.1 architecture could do itself.

 

[trinibwoy]

Sorry i don't get it how it's 4:1 ratio and not 1:1 ratio when half of SPs in the 1/8 (1/10) core are MADD and other half SFUs, all pumped up at least to double frequency of the i/o part of the chip??

SFU instructions take 4 clocks to complete. So while the number of units is the same the throughput is actually 4:1.

And 3:1 seem quite fine to me for ATi in the light of growing tessellation needs. And to stream little bit more (dividable by DP-dual precision needs) than DX10.1 architecture could do itself.

Why would tessellation increase the ratio of SF:MADD instructions? What you're proposing would put even more pressure on the compiler to find independent instructions to issue each clock.

 

[ShaidarHaran]

Has anyone else noticed that all the NVidia discussion is in the ATI thread and vice versa ?

The grass is always greener on the other side, right?

 

[liolio]

The R800 could consist of two chips according to this new:
http://www.fudzilla.com/content/view/13832/1/

 

[Kaotik]

The R800 could consist of two chips according to this new:
http://www.fudzilla.com/content/view/13832/1/

did anyone even ever think about any other possibility than X2-card, aka R800, at the high end?

 

[homerdog]

What are the odds we'll get something other than AFR this time around

 

[rpg.314]

Low.

 

[Ailuros]

did anyone even ever think about any other possibility than X2-card, aka R800, at the high end?

That's what the newsblurb actually describes if someone reads it careful enough; cliff note: nothing we wouldn't expect.

 

[hoom]

Yeah, I read that & was thinking "Sweet, if RV870 is two chips, R800 must be... Hey, wait no, R700 was 4870 X2 -> RV870 is still single chip"

 

[neliz]

The reaseon nVidia became more important for TSMC:
R8X0's produced in Dresden?

 

[trinibwoy]

Yeah I noticed that too. It would certainly spice things up. But if things remain the same GF might have more capacity than AMD's CPUs and GPUs require so they'll still be shopping for more clients. Aren't they building a new fab in NY as well?

 

[Lukfi]

I don't believe that. In articles that I've read about Globalfoundries, 40nm manufacturing was never mentioned.
http://www.brightsideofnews.com/news/2009/5/26/the-future-of-globalfoundries-revealed.aspx
Here it says that GF is currently isntalling 32nm bulk equipment into "Fab 1 module 2" (formerly AMD Dresden Fab 38).

 

[neliz]

I don't believe that. In articles that I've read about Globalfoundries, 40nm manufacturing was never mentioned.
http://www.brightsideofnews.com/news/2009/5/26/the-future-of-globalfoundries-revealed.aspx
Here it says that GF is currently isntalling 32nm bulk equipment into "Fab 1 module 2" (formerly AMD Dresden Fab 38).

that article also says that TSMC only has 1% of the 40/45nm capacity of F1M2. they are ramping up 32nm for production at the end of the year.

This xbit article is a bit more elaborate: http://www.xbitlabs.com/articles/other/display/globalfoundries-interview-2009_4.html

And for your lack of 40nm mentioning:

X-bit labs: Do you plan to offer so-called “half-node” process technologies, which are optical shrinks of available fabrication processes?

Tom Sonderman: Yes we do.

X-bit labs: When does AMD plan to start outsourcing its chipset and graphics chip production to Globalfoundries?

Tom Sonderman: We intend on competing for AMD’s graphics business in the 32nm/28nm technology node.

So 32/28 production should be on it's way in the end of the year. I guess that leaves plenty of space in the fabs to do 40 or.. even 45 at Dresden for R8X0.

 

[Lukfi]

I still don't see 40nm mentioned anywhere. Of course they will do 28nm, which is a half-node from 32nm. But they aren't doing 45nm bulk, neither its respective half-node, 40nm.

 

[Sound_Card]

Yeah I noticed that too. It would certainly spice things up. But if things remain the same GF might have more capacity than AMD's CPUs and GPUs require so they'll still be shopping for more clients. Aren't they building a new fab in NY as well?

They broke ground, but not a single brick is up. Sort of just sitting there right now as they talk to NY about tax specials.

 

[Jawed]

This:

http://www.pcper.com/article.php?aid=713&type=expert&pid=3

shows 40/45nm bulk (i.e. not SOI) but seemingly not aimed at GPUs. It also shows 32/45nm SOI aimed at GPUs. Scattergun slide?

Jawed

 

[neliz]

This is the stuff from financial analysts day 2008
http://pradeepchakraborty.blogspot.com/2008/11/amds-roadmap-2009-provides-lots-of.html

* For the bulk manufacturing processes AMD uses to manufacture its chipsets and GPUs, AMD plans to have access to 55nm, 40nm and 32nm manufacturing capabilities at:
- TSMC/UMC (Taiwan)
- Fab 38 (Dresden)
- Fab 4x (Saratoga County, NY)

GF's own website says that they have bulk available of just about anything they have on SOI vaguely naming 45nm and (it's?) half-node. but nowhere near as we can do XX at XX.