The G92 Architecture Rumours & Speculation Thread

Status
Not open for further replies.
I'm guessing that G92 might be something like G71 was to G70 although not a simple die shrink but more like 192 SPs & 256 bit MB @ 65nm -- much faster than G80 at single chip level (but slower or the same at doing AA) but not top end in NV40/G70/G80/R580/R600 kind of sense -- top end spot would be filled with GX2 board using two G92s in SLI.

(Though I'd personally prefer it to be a new cutting edge top end chip with 256+ SPs and 512 bit MB... Yeah, well...)
 
Well what you describe sounds rather like G70 to NV40.
Not really. Going from 130nm to 110nm was relatively small shrink and they've opted for more complex chip and not for cost-effectiveness.
What i'm saying is that _maybe_ they'll try to do both this time because 90nm->65nm allows this -- make a chip that is smaller than G80 and costs less to produce (like G71 was comparing to G70) but at the same time make a chip that is faster than G80 (like G70 was compared to NV40 but margins were more or less the same).
But it's a wild guess really :)

G92 to G80 will be a bit like RV530 to R350... ;)
Whaat?.. 8-\
OK i guess that means that G92 isn't high-end chip after all... Or not?
 
Last edited by a moderator:
Yyyy what does it exactly mean????

G92 is not yet guaranteed to be the codename for a high-end GPU.
In essence, he is hinting that maybe it will have similar (as in, modest) performance gains when compared to the 8800 GTX/Ultra as the RV530 (Radeon X1600) did against the R350 (Radeon 9800) because it's supposed to be the performance midrange part of the Geforce 9 series (a role more akin to a 8800 GTS 320MB than a 8600 GTS currently).

Of course, I'm just guessing here... ;)
 
Well a good midrange card would be news indeed.

What I'd like to see:
16rops
16tmus
128scalars
256bit/512MB

All for $200.
 
G92 to G80 will be a bit like RV530 to R350... ;)
  • significantly increased ALU:TEX ratio
  • reduced TMU count
  • reduced ROP count
  • significantly higher clock(s) - almost compensating for the reduced TMUs/ROPs
  • all for lots less money
Jawed
 
  • significantly increased ALU:TEX ratio
  • reduced TMU count
  • reduced ROP count
  • significantly higher clock(s) - almost compensating for the reduced TMUs/ROPs
  • all for lots less money
Jawed

And that is surprising because... ?
With the exception of the first generation of midrange DX10 cards, all midrange cards tend to outperform the previous top of the line.
And the RV530 is not one, but two generations ahead of R350 (don't forget the forgettable X700/RV410 in between), for not a lot more performance.

The RV530 equivalent from Nvidia (G73/7600 GT) was still significantly faster, even with slightly lower image quality.
 
Not really. Going from 130nm to 110nm was relatively small shrink and they've opted for more complex chip and not for cost-effectiveness.
What i'm saying is that _maybe_ they'll try to do both this time because 90nm->65nm allows this -- make a chip that is smaller than G80 and costs less to produce (like G71 was comparing to G70) but at the same time make a chip that is faster than G80 (like G70 was compared to NV40 but margins were more or less the same).
But it's a wild guess really :)


Whaat?.. 8-\
OK i guess that means that G92 isn't high-end chip after all... Or not?

maybe the new highend flagship refresh of G80 / GF 8800 is going to be G90 (as was thought a few years ago) and the G92 is an upper mid-range part.

regardless of what Nvidia calls it now, be it G90 or G92 or G9000 or GeForce 9800, the NV55 is what we're talking about -- Inq mentioned NV55 today.

G80 / GeForce 8800 = NV50 btw, even though Nvidia hasn't been using the NVxx nomenclature for quite some time now.
 
Last edited by a moderator:
Well, i doubt that they'll do a midrange DX10.1 part before the high end DX10.1 part. So if G92 isn't high end (and G98 is certainly not) then it looks like we're missing something :)
 
No, G92 is high-end (& D3D10.1).

However, in its single-chip configuration, it is more high-end in the sense of the 8800 GTS than in the sense of the 8800 GTX or Ultra. I'm thinking in terms of current pricing, fwiw. And some SKUs will most likely sell for less than the current 8800 GTS 320MiB...

What I just said is an indirect conclusion based on reliable information, but this should really be so obvious that I'm not sure that even matters! :)
 
No, G92 is high-end (& D3D10.1).

However, in its single-chip configuration, it is more high-end in the sense of the 8800 GTS than in the sense of the 8800 GTX or Ultra. I'm thinking in terms of current pricing, fwiw. And some SKUs will most likely sell for less than the current 8800 GTS 320MiB...

What I just said is an indirect conclusion based on reliable information, but this should really be so obvious that I'm not sure that even matters! :)

So we are potentially talking about a GPU thats very close in overall performance to the 8800GTX, has full DX10.1 support and can cost less than an 8800GTS 320MB?

Let the good times roll :D
 
So we are potentially talking about a GPU thats very close in overall performance to the 8800GTX, has full DX10.1 support and can cost less than an 8800GTS 320MB?
I said that one SKU would likely have >= 8800 GTX performance, and one SKU would likely cost less than the 8800 GTS 320MB. That's all. I think you are reading too much in what I said! :)
 
No, G92 is high-end (& D3D10.1).

However, in its single-chip configuration, it is more high-end in the sense of the 8800 GTS than in the sense of the 8800 GTX or Ultra. I'm thinking in terms of current pricing, fwiw. And some SKUs will most likely sell for less than the current 8800 GTS 320MiB...

What I just said is an indirect conclusion based on reliable information, but this should really be so obvious that I'm not sure that even matters! :)


If it is a single chip, how are they stating 2x glfops performance over g80(GTX)? Are they also moving to a multichip program?
 
Status
Not open for further replies.
Back
Top