The G92 Architecture Rumours & Speculation Thread
- Thread starter Arun
- Start date
- Status
1.2GHz is a joke, for sure!
Just swap the numbers around the point and all will come to place.
Or a real typo.
The article also mentions something called "8800GTX-320MX"...
Anyway, if the shader clock is in fact that low at 1.2GHz, then the 8800 GT, in order to trump the "classic" 8800 GTS (96 sp's also at 1.2GHz) either has to have substantially different scalar processors when compared to the current G8x crop, or a lot more than 64 in order to compensate.
I vote for 128 sp's
.
Sorry, I just saw this post and had to stop and LMAO.
SH, you def called that shit.
Sorry for the unintentional spreadage of FUD.
As for 8800GT, I wonder if CJ is hinting 96/128, IE it is a die-shrink of G80, and this is the GTS-like part?
Suddenly it sounds like things could be a little bit more interesting.
Last edited by a moderator:
I agree that the SPs could very well be tweaked. Better co-issue ops or something. Even if it has 345.6 GFLOPs like the GTS doesn't mean the utilization is the same. Hell even the G84 is better in that regard.
At such a low stock clockspeed though, it should be a record setting overclocker.
At such a low stock clockspeed though, it should be a record setting overclocker.
Now where was I hinting at that? There are more options.
And the shaderspeed seems a tad low...
NV 8800GT scores 11760 in 3Dmark06
That would be a lot for a 4 Cluster part, but maybe G9x clusters have a lot improved vertex-power, what likes the 3DMark with his synthetic requirements. (see G84 and R600)
That would be a lot for a 4 Cluster part, but maybe G9x clusters have a lot improved vertex-power, what likes the 3DMark with his synthetic requirements. (see G84 and R600)
I like how our (or at least my) understanding of 8800 GT's performance is slowly progressing, heh.
1) Slower than 8800 GTS.
2) On par with the 8800 GTS.
3) Between the old 8800 GTS and the new 8800 GTS.
4) On par with the new 8800 GTS.
5) Between the new 8800 GTS and the 8800 GTX, if not faster.
I'm sorry, but I could believe 4C at 1 and maybe 2, but I had to go to 6C to get to 3 and 4... And to get to 5, I need to go above 6C. That does give credence to CJ's hints though!
Looks like the 8800 GT will have a performance advantage over the RV670, but AMD's positioning is still wayyy better this round than with G80 vs R600, since they have a nice cost advantage (196mm2 vs 289mm2) now! Even after taking the 55nm advantage into account, this becomes 196mm2 vs 234mm2, nice!
I guess the only thing that might still give G92 both a cost and a performance advantage is if there is significant redundancy and there's a faster SKU in the wings. There isn't much evidence of that though, but it'd be interesting; not very hard to see how the initial rumours of 800MHz+ clocks would combine nicely with GDDR4. Anyway I already explained some of this both publicly and privately numerous times, so I'll stop here.
1) Slower than 8800 GTS.
2) On par with the 8800 GTS.
3) Between the old 8800 GTS and the new 8800 GTS.
4) On par with the new 8800 GTS.
5) Between the new 8800 GTS and the 8800 GTX, if not faster.
I'm sorry, but I could believe 4C at 1 and maybe 2, but I had to go to 6C to get to 3 and 4... And to get to 5, I need to go above 6C. That does give credence to CJ's hints though!
Looks like the 8800 GT will have a performance advantage over the RV670, but AMD's positioning is still wayyy better this round than with G80 vs R600, since they have a nice cost advantage (196mm2 vs 289mm2) now! Even after taking the 55nm advantage into account, this becomes 196mm2 vs 234mm2, nice!
I guess the only thing that might still give G92 both a cost and a performance advantage is if there is significant redundancy and there's a faster SKU in the wings. There isn't much evidence of that though, but it'd be interesting; not very hard to see how the initial rumours of 800MHz+ clocks would combine nicely with GDDR4. Anyway I already explained some of this both publicly and privately numerous times, so I'll stop here.
No, NV goes back to decated architecture: 4 PS-clusters and 2 VS/GS-Clusters.
I just wonder what's test system of the inquirer.
http://www.theinquirer.net/gb/inquirer/news/2007/10/03/nvidia-admits-defeat-g92-vs
They never mentioned it.
So who is really lead the flag now ?
http://www.theinquirer.net/gb/inquirer/news/2007/10/03/nvidia-admits-defeat-g92-vs
They never mentioned it.
So who is really lead the flag now ?
Last edited by a moderator:
8800 GT has 112SPs?
http://translate.google.com/transla...&hl=de&ie=UTF-8&oe=UTF-8&prev=/language_tools
Found @ PCinlife:
7 Cluster @ 600/1500MHz for $249/229?
http://translate.google.com/transla...&hl=de&ie=UTF-8&oe=UTF-8&prev=/language_tools
Found @ PCinlife:
7 Cluster @ 600/1500MHz for $249/229?
The "new" 8800GTS would have a higher memory bus (320 vs 256), but not so sure about the no of ROPs/TMUs.
Basing on G80 architecture, is G92 a 7 cluster part (16SP per cluster)?! or 4?
Think this is why G92 is *big* in terms of die size compared to what you would expect for a mid/high end solution.
The plot thickens..
Basing on G80 architecture, is G92 a 7 cluster part (16SP per cluster)?! or 4?
Think this is why G92 is *big* in terms of die size compared to what you would expect for a mid/high end solution.
The plot thickens..
The "new" 8800GTS would have a higher memory bus (320 vs 256), but not so sure about the no of ROPs/TMUs.
Basing on G80 architecture, is G92 a 7 cluster part (16SP per cluster)?! or 4?
Think this is why G92 is *big* in terms of die size compared to what you would expect for a mid/high end solution.
The plot thickens..
I wager that G92 is a 8 cluster GPU.
8800 GT has 112SPs?
http://translate.google.com/transla...&hl=de&ie=UTF-8&oe=UTF-8&prev=/language_tools
Found @ PCinlife:
7 Cluster @ 600/1500MHz for $249/229?
I hold piles of stock on Nvidia for decade.
Is that truth Nvidia give us a good cost/performance product ?
Becauase of everything what I acquired by or came across the AIB companies, and they all said that Nvidia is eager to utilise G92 from mainstream to high-end.
since Crysis, the evolution game is only playable by 2 times of 8800GTX performance, or more than 3 cards in one case.
Blackiice2007
Newcomer
So if 8800GT have 112SP , it will have 24 Texture units and 20 Rops ?
Who can explaine me whats the cluster number thing...
Who can explaine me whats the cluster number thing...
Last time I read the news, the matter at hand looked like this:
8800 GTS 320MB - end of life
8800 GTS 640MB - upped shader unit count from 96 to 112. nVidia won't even bother to advertise, this will be left up to the board partners, optional.
8800 GT - price-wise falls in slightly under or at 8800 GTS 320MB. Performance-wise it's good enough to merit an upgrade of the 8800 GTS 640MB to keep the performance envelopes sufficiently apart.
Now, explain to me what exactly implies that this card will even remotely threaten 8800 GTX/Ultra (as Arun seems to be thinking in view of the alleged 3DMark scores)?
This is the real world, where you do not make your mid-range product so fast that it outperforms your top three bestselling SKUs.
8800 GTS 320MB - end of life
8800 GTS 640MB - upped shader unit count from 96 to 112. nVidia won't even bother to advertise, this will be left up to the board partners, optional.
8800 GT - price-wise falls in slightly under or at 8800 GTS 320MB. Performance-wise it's good enough to merit an upgrade of the 8800 GTS 640MB to keep the performance envelopes sufficiently apart.
Now, explain to me what exactly implies that this card will even remotely threaten 8800 GTX/Ultra (as Arun seems to be thinking in view of the alleged 3DMark scores)?
This is the real world, where you do not make your mid-range product so fast that it outperforms your top three bestselling SKUs.
Last edited by a moderator:
- Status
Similar threads
