The Official G84/G86 Rumours & Speculation Thread

Does that mean that the B3D G84/6 article will actually appear on launch day? :oops: :p


This new site didn't write itself, y'know. :smile: And since we're following NDA's, I have to ask "what's a G84/G86?" :cool: Tho Rys does seem to be playing with something or other of late. Might be a Parhelia tho, ya never know with him! :p
 
This new site didn't write itself, y'know. :smile: And since we're following NDA's, I have to ask "what's a G84/G86?" :cool: Tho Rys does seem to be playing with something or other of late. Might be a Parhelia tho, ya never know with him! :p

Heh fair enough. It's all appreciated :)

BTW, by pointing out that Asus breached NDA aren't you indirectly acknowledging the existence of the product and thereby breaching your NDA? :oops: :p
 
IIRC, there was a similar subject a while back on the NV43, but it turned out to have insufficient transistors to back the claims.
Maybe it was Unwinder who first talked about it ? Can't remember exactly
Thats in relation to G73 / 7600. It has 3 shader quads while RivaTuner reported allocation for 4.
 
Guy over at [H] ran a 3dmark01 multi-texturing bench and came up with ~ 8200MT/s which points to 12TMU's @ 700Mhz (the guy's core clock).

So 3 clusters, 48 shaders? G84 is 4 clusters, one disabled for the GTS?
How does that tally with the 3196MT/s single-texturing score?

3dmark.jpg


Jawed
 
How does that tally with the 3196MT/s single-texturing score?

Not sure what you mean...the single-texturing numbers have never meant anything to me because they are never aligned with any theoretical fillrate number due to ROP/bandwidth limitations. Is there any information in the single-texturing score?
 
BTW, by pointing out that Asus breached NDA aren't you indirectly acknowledging the existence of the product and thereby breaching your NDA? :oops: :p

Well, it's inferred reasoning of course, due to the lack of reviews at major sites like Anand, [H], etc. <he slithered gracefully>
 
Not sure what you mean...the single-texturing numbers have never meant anything to me because they are never aligned with any theoretical fillrate number due to ROP/bandwidth limitations. Is there any information in the single-texturing score?
I don't understand 3DMk01's tests, before my time basically, so I was hoping for some interpretation.

http://www.techreport.com/reviews/2006q4/geforce-8800/index.x?pg=5

It turns out the ratio of G80's single- and multi-texturing rates in 06 is pretty much identical to G84's in 01. So in these tests the multi-texturing rate aligns very closely with the theoretical single-texturing rate.

Jawed
 
http://www.fudzilla.com/index.php?option=com_content&task=view&id=501&Itemid=1

It's definitely faster than 7600GT. Though, amusingly, 3DMk03 is definitely the most honest about the performance improvement, judging against the games tested.

I guess that just means Q4 and FEAR are past their sell-by date.

Jawed

Wow, that's a pretty strong showing from that overclocked 7600GT - I hope that magic driver really comes through for the 8600 GTS's sake.

And those numbers pretty much confirm 12 TMU's.
 
Wow, that's a pretty strong showing from that overclocked 7600GT - I hope that magic driver really comes through for the 8600 GTS's sake.

And those numbers pretty much confirm 12 TMU's.

More people should be using Archmark... The truth was out there all along.

edit:

Or would you have guessed about 32/64 TMUs in G80 if I'd given you this:
Code:
[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              GeForce 8800 GTX/PCI/SSE2 v2.1.1
Resolution          1024x768 @ unknown refresh rate
Comment             8800 GTX, FWxxx.xxHQ, XPx86
Method              Flush
[...]
[b]Texturing[/b]
Mode                R5G6B5A0 Z0 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               10.504 GPix/s

----[b]Trilinear filter[/b]----------------------------
    1               10.548 GPix/s

Rather not, me'd think. But if you'd had the complete Results you'd come pretty close by doing the math:

Code:
[b][url=http://www.zeckensack.de/archmark/]ArchMark 0.50[/url][/b]
Driver              GeForce 8800 GTX/PCI/SSE2 v2.1.1
Resolution          1024x768 @ unknown refresh rate
Comment             8800 GTX, FWxxx.xxHQ, XPx86
Method              Flush
[...]
[b]Texturing[/b]
Mode                R5G6B5A0 Z0 S0
--[b]Textured fillrate[/b]-----------------------------
----[b]Bilinear filter[/b]-----------------------------
    1               10.504 GPix/s
    2               7.348 GPix/s
    3               5.678 GPix/s
[b]    4               4.549 GPix/s[/b]

----[b]Trilinear filter[/b]----------------------------
    1               10.548 GPix/s
    2               7.377 GPix/s
    3               5.699 GPix/s
[b]    4               4.567 GPix/s[/b]
 
Last edited by a moderator:
More people should be using Archmark... The truth was out there all along.

Must say I'm confused as to what point you're trying to make :oops: I'm assuming Archmark's bilinear multi-texturing test is similiar to 3dmark's multi-texturing test and they both convey the same info - that G80 has 32 TMU's at 575Mhz.

The trilinear test betrayed the double filtering power but I don't think that's what we're discussing here re-G84. I'm taking the 3dmark numbers as evidence of the number of texture addressing units in the part.
 
What's this "Forceware 158.16" i've been hearing about ?
Since the current known driver is the 101.70, there is a huge numbering jump coming from there...
 
Just summarising those Fudzilla numbers:

8600GTS v 7600GT
  • bandwidth 143%
  • fillrate 121%
  • bilinear rate 121%
  • MAD 129%
  • games 132%
8600GTS v 7600GTOC
  • bandwidth 111%
  • fillrate 104%
  • bilinear rate 104%
  • MAD 112%
  • games 113%
Jawed
 
The game benchmarks imply that the 8600GT will have a very hard time outperforming a 7600GT. Ouch. This could have been an opening for AMD in the mobile space, but leaked data imply otherwise - DX10 compliance seems to take a heavy performance toll for both players.
 
I do think you will find in the future there will be less leaks regarding this type of thing.

A shared secret isn't a secret.... The only way to prevent "this type of thing" is to not ship hardware, or cripple the software that comes with it.

[showing huge restraint by passing over *both* jokes]

I'll be interested if/when any info comes to light as to why this sort of statement has been made, not once, but twice in this thread...
 
Back
Top