http://fudzilla.com/graphics/item/21171-cayman-is-a-264-billion-transistor-chip
2.64 bn transistors and 389mm2, sayeth fudo.
2.64 bn transistors and 389mm2, sayeth fudo.
Last edited by a moderator:
AMD: R9xx Speculation
- Thread starter Lukfi
- Start date
DavidGraham
Veteran
http://www.tcmagazine.com/tcm/news/misc/32319/tsmc-spending-10-billion-new-28nm-fab
This confirm 28nm delay ?
Completed by: http://fudzilla.com/graphics/item/21172-radeon-hd-6970-has-250w-powertune-maximum-power
I share the feeling that Cayman has some stuff disabled, not sure if its solely because of yields though. For me it smells more like strategy.
Conspiracy Theory:
When Charlie said that allocation for HD6950 was changed to HD6970, what if there wasnt a change in the chips? Imagine HD6970 was going to be 1920 SPs, and instead of a real allocation from 1536 SPs to 1920 SPs, what was changed was just the name?
The chip might have 1920 SPs, but launched with only 1536 enabled. This would give space for AMD to launch a HD6980 in the future, or even a renamed product for the 7xxx series, while waiting for the 28nm process to mature (quite possibly it will only be available on Q4 2011 or Q1 2012).
I cant imagine them developing another 40nm chip meanwhile. Maybe they are just playing their cards for the near future?
highly doubtful imho. AMD have no reason to hobble the 6970 if it can't beat a 580GTX. They can claim the 5970 is their offer against 580, but it is in dwindling supply and less attractive owing to being a dual-chip and having more variation in gaming framerates (higher max but lower mins than 580).
DavidGraham
Veteran
About "hoping" for production in 2011 , the new facility is still under construction , that means production at the end of 2011/beginning of 2012 at best .
If anything, it seems to do the opposite with the part about production in 2011.
What?
fab16 comes in 2014!
And the text says:
It gives no quarter and I have not read anything about "hope"
NathansFortune
Regular
Pretty much. Very similar to the whispers I've heard and I'm sure others here have heard them too. That TSMC have silently changed 28nm delivery from Q1 2011 to 2011 is confirmation enough that we're in for delays. If Fab 16 is really just breaking ground now we're going to be in for a long wait until 28nm becomes commonplace (cheap!).
I know people say there are too many companies dependant on 28nm being delivered early in 2011, but it will take as long as it takes.
All of this just goes to show how amazing Intel really are!
Its because [H] are ATI no nVidia no AMD no nVidia fanboys, obviously.
Me get disappoint if true only 1536sp for 6970....and price is close to $450.
Any one think 6990 is the 1920sp part? Was it ever confirm Antilles is dual gpu? A single core 1920sp @ 850mhz with more ROPs and TMUs with 4GB(eyefinity), will surely beat 5970. It will not beat GTX570x2 but does it need to?
Any one think 6990 is the 1920sp part? Was it ever confirm Antilles is dual gpu? A single core 1920sp @ 850mhz with more ROPs and TMUs with 4GB(eyefinity), will surely beat 5970. It will not beat GTX570x2 but does it need to?
At the very least, the texture rate and clock is consistent with 80 units, or 20 SIMDs.
The pixel fillrate at those clocks is consistent with 32 ROPs.
The pixel fillrate at those clocks is consistent with 32 ROPs.
It could also be Cayman Pro ending with 8 (1408SPs) and you are seeing the corner of that 8.
Also, you do realize that GPU-Z does not read anything from the ASIC right? The values are manually input into the program and tied to the device ID which it then reads and fills in the correct info.
I believe the 10% smaller simd size includes everything, which means tmus too. And don't forget simds only take up a fraction (less than half for sure) of the area, so the die size increase must be mostly elsewhere. But I'm not sure where - doubling the geometry stuff, and the improved rops imho shouldn't increase area much. Though people speculated that internal bandwidth (L2->L1) isn't quite high enough for Cypress (afaik it's the same per clock for Juniper, Barts, Cypress) so maybe that got a boost too, I would have no idea though how much real estate this would take up.
Well it's not much faster though hence the comparison seems valid.the 6870 is a more powerful card and thus more prone to system bottlenecks.
sorry I had to.
And really, it should be very similar. It is pretty much the same, after all, though of course drivers will make a difference (and one game not supported by only CF or SLI could easily skew result in one direction).
Not sure gpu-z can read this out correctly yet?
Wait what - 20 SIMDs? So 1280 SP aka 6950?
edit:
Or one of those rumored 6970's with crippled SP numbers?
Which either means a fake or that the database GPU-Z is using is simply out of date
Nope. I just played with photoshop and the 8 below overlaying it and it's not an 8. It's a zero or a fake.
NathansFortune
Regular
Looks like a 0, which would indicate 1920 or 1280 barring any fakes.
Also, GPU-Z is just a database and lookup table right? So it would require inside knowledge and someone to input the Cayman data.
Also, GPU-Z is just a database and lookup table right? So it would require inside knowledge and someone to input the Cayman data.
Well, at least it can read the shader count of previous generations of AMD chips:
http://www.techpowerup.com/articles/other/155 (560 SP 4830)
http://www.techpowerup.com/reviews/HIS/Radeon_HD_6850/ (1120 SP 6850)
Those two were not "AMD-sanctioned" models, still they carry the same Device IDs as their "correct-SP" brethren.
Similar threads
