The G92 Architecture Rumours & Speculation Thread
- Thread starter Arun
- Start date
- Status
What's benefit of 112 scalar processors ? make GTS faster ?
~17% more power if there is no bandwidth-limitation.
But I think Old-GTS will die, because it is to near to 8800GT and nobody would buy it.
Any news on the Nvidia lowing the price for 8800 GTX ?
oh and is the new 8800GT going to better than GTX ?
THX !
G92-high-end has heat problems?
Why does the G92 will have heat problem when the VR-Zone told us that the GeForce 8800 GT Overclocking Is Good.
Hope it might be not a serious problem then.
or maybe dual slot cooling?
Hope it might be not a serious problem then.
fellix said:
or maybe dual slot cooling?
This is exactly what I heard earlier this week. Was wondering when it would be leaked into the public.
Or make the fan rotate a bit faster:May be they just should slap more heat-pipes in there.
http://www.theinquirer.net/en/inqui...he-geforce-gf9800-will-be-as-quiet-as-a-mouse
Because they are not talking about the G92 on 8800GT?
It could be launched earlier because they wanna sell as much GF8800GT cards as they can before RV670 launches... And ever wondered why the GF8800GT is clocked at only a measly 600Mhz (according to rumors)? This could also imply that they were having some problems with heat/power and the single slot formfactor they were aiming for.
And ever wondered why the GF8800GT is clocked at only a measly 600Mhz (according to rumors)? This could also imply that they were having some problems with heat/power and the single slot formfactor they were aiming for.I don't see how both can be true. I suspect FUD at work here. The question is: which is FUD and which is reality?
That's an interesting point of view, although if it's bandwidth limited anyway, why bother? Reducing the core clock, thus power and heat, can also result in a lower BOM because the PCB and cooling can be less expensive. Unlike in the enthusiast segment, increasing performance by X% isn't worth it if it also increases costs by 2X%!It could be launched earlier because they wanna sell as much GF8800GT cards as they can before RV670 launches...
Also, may I point out that if we assume the 8800 GT to be the 'cut-down' SKU, the clock disparity between the 7900 GT/GS and the 7900 GTX was similar if the high-end G92 SKU is clocked at 825MHz+? Indeed, it can easily be seen that 600*(650/450) = 866.67MHz, which is much nearer the previously rumoured G92 core clock...
And variation on 65nm is presumably worse than on 90nm, not better.I said and I'll say it again: I think the 8800 GT has 6 clusters activated and runs at 600/1800, while the top G92 SKU will have 8 clusters and run at 800/2800 or even higher. It wouldn't exactly be very hard to hide a SKU like that if they wanted to. And unlike for the 8800 GT, there is no good reason for AIBs to ever make their own PCBs or cooling solutions here, most likely... At least IMO.
This was a typo... right ?
I know there's GDDR4 memory running at these speeds, but the cost must be stratospheric right now.
This was a typo... right ?
Unless they use GDDR4 or 5, of course. Did anyone ever confirm whether or not NV's G8x/G9x MC has support for anything newer than GDDR3?
1.4GHz GDDR4 is nothing out of the extraordinary. We're talking 1.2GHz for the fastest RV670 (aka Gladiator), for example. The cost of memory has always been very significant on high-end parts anyway, since it's one of the key value differentiators!
You could argue that it might have been a better idea to just go for a wider bus in that case, but that'd also have made the PCB more expensive and increased the number of chips required... So it's not so obvious to me it'd be much cheaper. This solution can also be more power-efficient (the voltage on 1.4GHz GDDR4 remains very reasonable).
Now, of course, if you were talking about 1.6GHz GDDR4 for example, the prices might become much more ridiculous and you might have a point... However, even that might make sense if it's for a very low-volume SKU.
You could argue that it might have been a better idea to just go for a wider bus in that case, but that'd also have made the PCB more expensive and increased the number of chips required... So it's not so obvious to me it'd be much cheaper. This solution can also be more power-efficient (the voltage on 1.4GHz GDDR4 remains very reasonable).
Now, of course, if you were talking about 1.6GHz GDDR4 for example, the prices might become much more ridiculous and you might have a point... However, even that might make sense if it's for a very low-volume SKU.
Not so different when you remember that Nvidia's is 65nm and AMD's is 55.
289 * (55 * 55) / (65 * 65) = 207.
207 vs 194 - not so very different. Maybe the half-node gamble will pay off this time. (Surely it has to some day?
)Well, what I'm saying is that they're pretty much using 3/4th of a G92 to compete against a full RV670. If I'm right, and this is a big if, then clearly the fastest single-chip desktop G92 will be a faster than the fastest single-chip desktop RV670 by a fair bit.
Also, checking Samsung and Hynix's sites, both have been in mass production of 1.4GHz GDDR4 basically forever, and have said so publicly for a while. Samsung has also demonstrated 2GHz GDDR4 while Hynix seems to be claiming they're mass producing 1.6GHz GDDR4. So really, 1.4GHz+ GDDR4 for the highest-end SKUs is hardly unimaginable if the memory controller supports it. And unlike 'slow' GDDR4, it might also less of a problem latency-wise.
nicolasb: 55nm is a 90% linear shrink, so the die size should be 19% lower than on 65nm. That would be 234mm2, thus, not 207mm2...
Jawed
Legend
I thought it might give some people pause for thought on the recurring "yields" question, since people love to fawn over tiny dies.Not so different when you remember that Nvidia's is 65nm and AMD's is 55.
289 * (55 * 55) / (65 * 65) = 207.
207 vs 194 - not so very different. Maybe the half-node gamble will pay off this time. (Surely it has to some day?
It would appear NVidia's aiming to make two categories of SKUs from it, just like G80. Which begs the question, what size bus does the full chip have? Is it likely to stay at 256-bits, or will it have more ROPs/MCs and be 320/384-bits?
It seems strange to me that NVidia wouldn't release the fully-enabled chip before Christmas if the part-enabled chip is launching in about 1 month. If the heat rumours mean anything then perhaps they indicate that fully-enabled chips are scarce? Would that be that "fine-grained" redundancy in (in-)action
Doesn't add-up...
Jawed
- Status
Similar threads
