AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

fellix said:
Yes, compute features do matter here much more. And don't forget that DP performance increase is much more that the plain difference in ALU count.
AMD may have a problem in the consumer space if a GK104 with similar die size doesn't have ECC and 1:2 DP overhead. Nothing dramatic, but Nvidia already paid the price of a switch to a high-geometry throughput and compute oriented architecture earlier. It's not even clear if AMD has taken care of the former this time around.
 
unified as in support for DT+mobile?

Well there's that of which you speak as well as the drivers themselves. Since the arch is a bit different with GCN I am wondering how this will play out with win7 wddm 1.1 as well as win8 wddm 1.2.
 
Nvidia already paid the price of a switch to a high-geometry throughput and compute oriented architecture earlier. It's not even clear if AMD has taken care of the former this time around.
They have taken care. Maybe not to the same amount as nVidia did with Fermi (still only 2 setups and slower for medium and high tess factors), but it is definitely improved with about 3.5 times the tesselation performance of Cayman (probably done by holding the data in the L2, somewhat similar to what Fermi does). With this tesselation performance they will still lose in directed tests, but it probably removes quite a bit of the bottleneck for the amount of tessellation one can expect in games for the near and medium term (as seen in Unigine heaven where the Tahiti performance is supposed to be 2.4 times that of Cayman and about 50% in front of a GTX580, if I compare the right numbers :rolleyes:).
 
If they announce that but I can't buy the cards then yes, it is.

If you can't but I can, do you consider it a paper launch? Maybe your country is not a priority to AMD, but it doesn't mean it's a paper launch, it's allocation to the most important markets.
 
If they announce that but I can't buy the cards then yes, it is.

and much crying was had??????

ie who really gives a shit, logistics at this time of year is figgen nutts. a week after new years makes sense from that point of view.


edit: every silly season degustator is hear campaigning hard for the green team, maybe you can link these performance numbers for us?
 
They have taken care. Maybe not to the same amount as nVidia did with Fermi (still only 2 setups and slower for medium and high tess factors), but it is definitely improved with about 3.5 times the tesselation performance of Cayman (probably done by holding the data in the L2, somewhat similar to what Fermi does). With this tesselation performance they will still lose in directed tests, but it probably removes quite a bit of the bottleneck for the amount of tessellation one can expect in games for the near and medium term (as seen in Unigine heaven where the Tahiti performance is supposed to be 2.4 times that of Cayman and about 50% in front of a GTX580, if I compare the right numbers :rolleyes:).
They tweaked their geometry engine once again but it is basically the same as in Cypress and nowhere even close to what NV have since GF100. It does look like their tesselation/geometry performance is "enough for current moment" though thus it's not a problem. From everything that's been leaked now the real problem looks to be the cost of the cards relative to their performance - it does look like Tahiti costs too much for what it brings to the table. I'm still hoping that we'll see more than what was in these leaked slides because this was underwhelming for a 28nm $500+ videocard.

If you can't but I can, do you consider it a paper launch? Maybe your country is not a priority to AMD, but it doesn't mean it's a paper launch, it's allocation to the most important markets.
So having, let's say, about 500 cards for the whole world is not a paper launch? OK then.
If I can't buy it a week after launch then it's a paper launch to me. And yes, I'm talking about myself. Always.
And last time I checked Russia is one of the most important market for PC gaming.
 
If I can't buy it a week after launch then it's a paper launch to me. And yes, I'm talking about myself. Always.
Pretty sure most of Nvidia's launches have been paper according to me then ..

Not even sure why I'm wasting my time on this. :???:
 
So having, let's say, about 500 cards for the whole world is not a paper launch? OK then.
If I can't buy it a week after launch then it's a paper launch to me. And yes, I'm talking about myself. Always.
And last time I checked Russia is one of the most important market for PC gaming.

So, what you're saying is that GTX 480 was a paper launch. Good to know.

Regards,
SB
 
They tweaked their geometry engine once again but it is basically the same as in Cypress and nowhere even close to what NV have since GF100.
I agree that Fermis geometry handling is more advanced (I didn't say otherwise, Tahiti has no distributed Tess for instance). ;)
 
...

So having, let's say, about 500 cards for the whole world is not a paper launch? OK then.
If I can't buy it a week after launch then it's a paper launch to me. And yes, I'm talking about myself. Always.
And last time I checked Russia is one of the most important market for PC gaming.

And everyone in Russia buys a highend videocard as a New Year present. And everyone in Russia owns a highend videocard.
 
So having, let's say, about 500 cards for the whole world is not a paper launch? OK then.

Hmm. and my numbers say it's a lot (and I mean, a lot) better than Cayman, but we'll see next year.


(btw, 500 pieces sounds like the availability of the 590 at launch.)
 
Back
Top