AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

… and then there's gameworks.

Actually, this is especially rich. Gameworks only needs to provide implementations for 11_1 or 12_1 and nothing inbetween, you don't want to make a specialization for hardware you don't produce obviously. Same as: ICC doesn't support XOP.
 
Actually, this is especially rich. Gameworks only needs to provide implementations for 11_0 or 12_1 and nothing inbetween, you don't want to make a specialization for hardware you don't produce obviously. Same as: ICC doesn't support XOP.
fixed.
This mean also that gameworks should work on all D3D12 GPUs unless it force any NVIDIA proprietary extension.
 
fixed.
This mean also that gameworks should work on all D3D12 GPUs unless it force any NVIDIA proprietary extension.

Thanks.
I think they will always have a down-compatible [and slow] implementation in addition to the state-of-the-art one, instead of none.
 
Well yeah wccf finally seems to have got the memo, at least they credit their source which is of course two pages before in this thread. I should've linked to the post itself.
Actually, this is especially rich. Gameworks only needs to provide implementations for 11_1 or 12_1 and nothing inbetween, you don't want to make a specialization for hardware you don't produce obviously. Same as: ICC doesn't support XOP.

Nvidiai dont support 11_1 .... only 11_0 .. But they support 12_1, only on Maxwell .. As we can think will be the case of Fidji .. or we get a problem there if it dont support 12_1... ( this mean the specification of it have been included afterward ) .

As for Gameowrk .. whatever is the level features of it or not, tthis will allways been a problem on non Nvidia gpu's . wait the next batman, we will get some funny things to discuss about it.
 
So if its OK for NV to claim partial 11_1 support for Maxwell1, can't AMD claim partial 12_0 for GCN 1.0? (partial 12_1 for GCN 1.1/1.2?)

Edit: anyways, thats a damn big die on Fiji O_O

I've been fearing an R600 type big underperforming mis-match.
But maybe its gonna be more RV770 style balanced big compute density improvement?

Edit2: is there actually some evidence that 4GB is likely to be a significant limit?
I'd like to see some VRAM usage graphs proving it to be so.
 
Last edited:
Hardware Battle has a report in Korean on AMD's 300 series and the Fury lineup. Strangely enough, the important information is in a picture file and I get an error when putting the site URL through Google and Bing translators. With the help of my parents, I was able to translate it (we are Korean). Our translation of the Fury portion of the report is below.

The most anticipated new product from AMD is Fury, and following what has been previously mentioned here, the Fury lineup will consist of three models. These models are Fury Nano, Fury XT, and Fury Pro.
  • The Fury series will first come in reference models and both water-cooled and air-cooled products will be released. Non-reference models will come to market as early as mid-August or as late as early September. For the next month or more, only reference models will be released.
  • As expected, the supply of reference models will be limited.
  • We believe that the AMD Fury series is sufficiently competitive with the TITAN X and the GTX 980 Ti.
  • AMD Fury will have a GPU and memory on top of a small interposer, so the concentrated heat from the interposer region looks unsatisfactory concerning the TDP, but in the reference models this problem is limited.
  • The heat from the cooler is of a high level (like the TITAN X), but the noise level is of no concern.
 
iMacmathician, thanks for that fresh info!
Two of the five points are trying to say "The interposer gets really hot, but it still works."
 
They could make a card with dual-Tongas and 3+3GB later when DX12 games and VR have a substantial presence in the market.
It might be cheaper to do than Fiji cards, and power consumption could be decent enough.
 
384-bit wouldn't make sense without 48 ROPs to go along with it. Clock it to 1.1Ghz, up the memory clocks and capacity, and of course nvidia's new-gen maxwell optimizations are here to help and 780Ti is within sights.
 
384-bit wouldn't make sense without 48 ROPs to go along with it. Clock it to 1.1Ghz, up the memory clocks and capacity, and of course nvidia's new-gen maxwell optimizations are here to help and 780Ti is within sights.

Probably. The current Tonga SKU looks like to be ROP limited in the fill-rate tests. But a ~200MHz bump for the core could warrant all the 6 memory controllers going wired.
 
XFX uploaded some 390X pics on their 290X DD page:

http://xfxforce.com/en-us/products/...-290x-double-dissipation-edition-r9-290x-edfd

R9-390X-8DFR_4.JPG


http://videocardz.com/56145/xfx-radeon-r9-390x-pictured

The images are at the bottom of the page. Rest of the page still seems to have 290X specs etc.
 
Back
Top