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

Yeah, just the "X" implies either "10" or "what?" to me. Another letter wouldn't have hurt, having an XL and LE or something easy to distinguish performance by.



I think Unwinder might go after them if they tried that, and I'd be rooting for him! :devilish:

a little ati (pre-amd) history lession (how i remember it)
LE = means slow
XL =means a little faster than slow
PRO = means faster
XT = even faster
XTX unholy fast
 
Well. it's „your“ hardware and „your“ drivers. So you should know what's going on there, shouldn't you? Why would the drivers report this feature if it isn't there?
I don't know why you think Dave should be omniscient. It's quite easy for someone to not know every detail at a company the size of AMD and it's easy for bugs to creep into software, etc.
 
I just took our review sample HD 5770, so it is basically a 1st or 2nd-batch Juniper part (first batch went to devs seeding DX11) and installed the last WHQL (Catalyst Omega 14.12).

Turns out it does indeed report OpenCL 2.0 compliancy. So even with a moustache, I call the AMD-spec-page argument invalid.
View attachment 734

Maybe they hired some awesome software guys behind your back or maybe it's just akin to having a dx9 card on vista.
OpenCL 2.0 platform does not mean all devices are OpenCL 2.0. End of discussion.
 
With the amount of rebranding....how was Nvidia able to get a new arch. in Maxwell from top to bottom at 28nm, while AMD struggles to push out just the Fury...??

I see this as definitive win for Nvidia this round, never-mind how Fury will benched....i have a feeling AMD needs to sell Fury cheaper because Nvidia has an open secret: overclocking!
 
With the amount of rebranding....how was Nvidia able to get a new arch. in Maxwell from top to bottom at 28nm, while AMD struggles to push out just the Fury...??
It's simple. Nvidia's previous product line (700) was based on rebrands, while AMD's was half new (Bonaire /fixed/, Hawaii), half rebranded (Pitcairn/Curacao, Tahiti). This time Nvidia has four new GPUs (GM107, GM204, GM206, GM200) and AMD's product line seems to be based (again) on two new GPUs (Fiji, Tonga) and two rebrands (Hawaii/Grenada, Curacao/Trinidad).
 
Last edited:
It's simple. Nvidia's previous product line (700) was based on rebrands, while AMD's was half new (Pitcairn ...
Guess you mean Tonga. Anyway it is not that simple, GCN 1.0 to 1.2 is by far not as much of architectural change as Kepler to Maxwell is. We don't even know if Fury has something new, beside the memory.
 
For the past 4 years Nvidia already made two major architectural revisions, with Kepler and Maxwell in all possible directions -- power, performance and features. I wonder how long AMD will keep the guts of GCN untouched? Tonga got some minor ISA tweaks (beside the frame-buffer compression), but that hardly counts.
 
Tonga, not Hawaii. Tonga is less than one year old now. I consider all Bonaire, GM107 and Tonga as a part of the newer product lines. Anyway, it doesn't change anything on the fact, that AMD released 4 new GPUs since the first GCN generation and Nvidia also released 4 new GPUs since Kepler generation (not counting iGPUs and ultra-low-end products).
 
Not sure what is going on there. OpenCL 2.0 compliance occurs with Bonaire and later.
Could you comment on the Cape Verde / Oland / Pitcairn(Curacao) thing, too?
Since AMD does list them as supporting OpenCL 2.0 in HD8000M series and later, but not in HD7000 or HD8000 (OEM) series
 
Better heat dissipation for GDDR5?

ULk6AXM.jpg
 
Guess you mean Tonga. Anyway it is not that simple, GCN 1.0 to 1.2 is by far not as much of architectural change as Kepler to Maxwell is. We don't even know if Fury has something new, beside the memory.

I think he meant Bonaire.

Looking back at 285's benchmarks against 280, it could be a 15% improvement in some games for a grand total of like 6% improvement overall. It might not be worthwhile to put it in Hawaii when it's 1.1 already.
 
Not really. GDDR5 dissipates more heat trough the PCB
Better heat dissipation for GDDR5?
Not really. Packaging type of the GDDR devices dissipates more heat through the soldering pads on the PCB. I would put heatsinks on the backside of the board, right over the memory chips.
 
Compared PCB pic Jawed posted to my own launch edition of R9 290X and it's identical to the smallest SMD element on PCB. Power delivery chokes are from different manufacturer as well as solid capacitors, but placement is identical.

So this is not even a PCB revision ...
Source for R9 290X detailed pics from my modding thread on another forum.
 
Seriously? "Since when" I'd say, but if this is really going on, then it's gotta be since a very long time.
Why haven't I heard about this before?
It goes back to the dirty old days of Geforce FX with fp16, shader substitution etc.
You don't remember when the NV recommended software list was leaked and a quick scan of the 'independent balanced' reviews turned up a very strong correlation between the list & the tested software/resolutions?
Software which turned out to be the ones NV had put optimisations into/had shader substitutions for and which in a number of cases included very obvious visual fidelity issues that were not noted by the 'independent balanced' reviews?

GCN2 and above.
ZOMG major GCN revision confirmed! GCN 1.0 -> 1.1 -> 2.0 :runaway:

probably just a typo or marketing naming slip or something?
 
Back
Top