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

Curacao is R9 270/X, whereas AMD list R7 200 series and R9 290/X only for OpenCL2.0

OpenCL™ 2.0 conformance logs submitted (pending ratification) for: AMD Radeon HD 7790, AMD Radeon HD 8770, AMD Radeon HD 8500M/8600M/8700M/8800M/8900M Series, AMD Radeon R5 M240, AMD Radeon R7 200 Series, AMD Radeon R9 290, AMD Radeon R9 290X, A-Series AMD Radeon R4 Graphics, A-Series AMD Radeon R5 Graphics, A-Series AMD Radeon R6 Graphics, A-Series AMD Radeon R7 Graphics, AMD FirePro W5100, AMD FirePro W9100, AMD FirePro S9150

Though R7 265 is a 7850 launched against 750Ti. Then again I'm not seeing 285 in there.
 
And we still haven't gotten to the bottom of "is Pitcairn and Curacao (and whatever the version was called in 300-series) the same thing - and if so, why is OpenCL 2.0 listed as supported on Curacao but not Pitcairn, while AMD says GCN 1.1 is required
Nvidia had to disable the L1 cache in compute workloads for all the early Kepler parts, but fixed it with a minor change for GK110B. I wonder if Curaçao is a similarly minor fix that allows OpenCL 2.0 to work.
 
It was stated that memory controller of Curacao was redesigned to support higher frequencies. As for Trinidad, maybe power savings are a sign of further refinement. Too bad there are no more prints on the chips themselves, fabs would be easily determined.
 
AMD GCN GPU Disabled Stream Processors Unlockable with Software

Interesting news, it seems that if you own a recent GCN based Pro series GPUs from AMD (Hawaii Pro / Fury Pro / Tonga Pro / etc) is now possible to unlock, at least some of the disabled parts on the GPU.

Over at the OCN forums a user called tx12 released a little utility called CUinfo and this tool can be used to read information about active and disabled CU units in Hawaii, Tonga and Fiji. Basically the tool reads out the GPU core configuration, which can be altered and then flashed back into the BIOS with another tool and some modification. The current status of the tool is that it only reads out info, and you will have to edit and re-flash the BIOS yourself.

Cards like R9 Fury non-X, R9 390, R9 285 could be enabled with more CUs, I say COULD, as there are two methods used by AMD, laser cuts (prevents this from working at all) and through firmware, which would be this methodology.

Be warned though, GPUs are binned, defective CUs could be disabled for a reason, if you enable them ... who knows that will happen. Basically you will be gambling with your graphics card.
http://www.guru3d.com/news-story/am...ream-processors-unlockable-with-software.html
 
I skimmed through the thread and unfortunately, no one has been able to unlock a Tonga GPU to 2048 shaders / 32 CUs yet.
 
They used a 290X as comparison, which doesn't seem very "honest" considering the Hawaii chip is being sold in the 390X line with 5% higher core clocks, 20% higher memory bandwidth and twice the memory amount.

The Nano may very well trade blows with the 390X in most scenarios, though AMD will avoid that comparison because they're planning to price the Nano much higher.
 
The 345mm2 one is Cypress. If we take 12 months for tapeout to release for first HBM GPU - Fiji took this time. This could be Titan planned for 2013.
 
The candidate die and interposer area for the penultimate setup seems most in line with earlier research and estimates for interposer area, being in the ballpark of 800mm2.
Fiji seems to have pushed things marginally higher, possibly enabled by the band of unetched silicon that parts of the component dies extend over.
The arrangement of the DRAM stacks also changed between the two versions.
 
AMD have a "meet the expert" the 27/08 so i tend to believe that the date of launch rumored thoses last week is effectively true ( was the same for the FuryX launch. their was a webinar meeting at the same time )

AMD-Radeon-R9-Nano-Graphics-Card_Fiji-GPU-635x424.jpg


AMD-Radeon-R9-Nano-Graphics-Card_Awesome-635x424.jpg


AMD-Radeon-R9-Nano-Graphics-Card_PCB-635x260.jpg


AMD-Radeon-R9-Nano-Graphics-Card_Back-635x424.jpg
 
Nice looking design indeed. Very compact and tidy. It could have used a protective shield on the rear side though...

Weird though, I would have expected a vapor chamber set under that cooler; AMD has quite a bit of experience with those and seem to like them a lot. I'm curious about the layout of those heatpipes tho, the two big flat fat ones are easy of course, but the thinner one on the left side makes me curious. It seems to be tucked in on top of where the voltage regulators make contact in one end, but where the other ends up I'm not sure... Oh the nerdness! :D
 
A Vapor Chamber is not so good in very small designs like the Nano. You have a direct contact between the VC and the fins for heat transfer. But with a cooler this small, much of your heatspot on the cooler lies directly under the fans' hub. For an axial design like this it is better to move the heat to the sides where the airflow can take it directly and completely away from the card. That's my guess.
 
Doesn't look like it - just that. Wouldn't be the first time I am wrong.
 
Last edited:
You have a direct contact between the VC and the fins for heat transfer. But with a cooler this small, much of your heatspot on the cooler lies directly under the fans' hub.
Uh, I'm not sure I follow you. Vapor chambers are like wide, flat heatpipes. They're meant to transfer heat away from the central contact point; whether the fan hub is under the contact point or not doesn't really make a difference I would think (other than lowering the cooler's efficiency, since there's little/no air flow under the hub itself).

Btw, AMD used a small - possibly even smaller than the Nano's cooler - vapor chamber in the radeon 4970 X2 way way long ago now for the "downstream" GPU.
 
hmm something isn't right there shouldn't Fiji have 8 ACE's? Oldly familiar with the review guide pic but different.

Nano might be a different GPU then the full fury?
 
Back
Top