r9 300 The pirate islands

Heh 50% more ROPs? They really want to push that ultra resolution lead they have!
 
meh -- my pair of 290's push my 1440p display nicely

I'd be more interested in power consumption and it's hashing ability :)
 
Looks like possibly an early target plan, but I wouldn't read too much into it.

Based on the W9100, I do like specs for the 390x for work at 4K. Wouldn't be surprised if the final product is a bit better than this outline though ...!
 
Trimlock said:
Heh 50% more ROPs? They really want to push that ultra resolution lead they have!
Click to expand...

As questionable as wccftech is, I think that is going to be possible at 20nm process.

Could AMD squeeze 50% more ROPs on a die size that is roughly the same as a 290/290X yet on 20nm process?

Likely. However, that's a lot of wishful thinking considering there are other parts of the GPU there on the same die. The information mentioned supposedly came at a time before TSMC announced issues with 20nm manufacturing, so it's likely these specs may have changed. 4224 SP on a 512-bit bus with 7GHz memory looks like it'll outperform 290X by a wide margin if these are true and unchanged before it's release.

Also, don't read the comments, it's headache-inducing. One mentioning there will be HBM-- high bandwidth memory-- that's stacked is unlikely to show up in PI this year or next. Nvidia announced Pascal recently with stacked memory, but that product isn't expected to come out until 2015 to 2016 timeframe, and not in Maxwell cards. The best I can think of based on Nvidia's announcement of Pascal and AMD's partnership with Hynix for HBM stacked memory is that products using these from either company won't see fruition until 2015 to 2016 based on how long it takes to design a product from paper to wafer.
 
Last edited:
octoberasian said:
Also, don't read the comments, it's headache-inducing. One mentioning there will be HBM-- high bandwidth memory-- that's stacked is unlikely to show up in PI this year or next. Nvidia announced Pascal recently with stacked memory, but that product isn't expected to come out until 2015 to 2016 timeframe, and not in Maxwell cards. The best I can think of based on Nvidia's announcement of Pascal and AMD's partnership with Hynix for HBM stacked memory is that products using these from either company won't see fruition until 2015 to 2016 based on how long it takes to design a product from paper to wafer.
Click to expand...

AMD is likely to have it sooner than Nvidia though...

AMD showed off stacked memory behind closed doors a long time ago.
HBM late this year or early next year is quite possible.
 
LordEC911 said:
AMD is likely to have it sooner than Nvidia though...

AMD showed off stacked memory behind closed doors a long time ago.
HBM late this year or early next year is quite possible.
Click to expand...

That's good news then. If not the Rx-300 series, but maybe the Rx-400 series and possibly Excavator APUs in 2015, perhaps?
 
octoberasian said:
That's good news then. If not the Rx-300 series, but maybe the Rx-400 series and possibly Excavator APUs in 2015, perhaps?
Click to expand...

The old roadmaps, from 2009, that were made before Ibiza on 32nm got re-worked had the highend GPU, Tiran, and the original Kaveri with stacked memory sometime in 2012, probably 2H.
Obviously a bunch of these products got moved around, delayed and cancelled due to a multitude of reasons but AMD still had working prototypes of stacked memory back before 2011.

If they really have been able to get the cost/complexity issue down to normal levels, I don't see why they wouldn't push forward with at least the highend APUs if not highend GPUs since it is supposed to be 40-60% more performance with 30-40% power savings compared to GDDR5. http://sites.amd.com/us/Documents/TFE2011_006HYN.pdf.
 
The 390X looks beastly if the specs hold up. I hope they can get it out sooner than later; I need an upgrade.
 
So against octoberasian's recommendation to stay out of the comments section I found some good posts against the obvious trolls. One person pointed out the possibility of having two ROPs per shader cluster which would essentially remove the bottle neck completely.

Obviously this will be very hard to do without them going to 20nm, but if they can pull it off we would have a perfect gpu core for 4k, this would also have big benefits for their professional level cards too.
 
LordEC911 said:
AMD is likely to have it sooner than Nvidia though...

AMD showed off stacked memory behind closed doors a long time ago.
HBM late this year or early next year is quite possible.
Click to expand...

This is very good news to me and would have epic results with their APUs.
 
Trimlock said:
So against octoberasian's recommendation to stay out of the comments section I found some good posts against the obvious trolls. One person pointed out the possibility of having two ROPs per shader cluster which would essentially remove the bottle neck completely.

Obviously this will be very hard to do without them going to 20nm, but if they can pull it off we would have a perfect gpu core for 4k, this would also have big benefits for their professional level cards too.
Click to expand...

I think it's very likely that Pirate Islands will be on 20nm, assuming TSMC can get their shit together. AMD has pushed 28 nm as far as they can with Hawaii, and any further development would be money down the drain.
 
Ozymandias said:
I think it's very likely that Pirate Islands will be on 20nm, assuming TSMC can get their shit together. AMD has pushed 28 nm as far as they can with Hawaii, and any further development would be money down the drain.
Click to expand...

Well I'm not holding my breath, I just hope global foundries can pick up the contract.
 
They probably told Global Foundries that if they can get them chips before TSMC then they got the contract.
 
You must log in or register to reply here.
Back
Top