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

[Deleted member 13524]

So, it comes down to, how likely is it a 17 year old will have access to this and be posting about it? And if someone thinks perhaps his parents are the ones with access... How likely is it that someone who has access to this wouldn't be able to afford computer stuff (even hand me downs) for their kid?

1 - If he's from Southeast Asia, there's a good chance that the parent who works at an OEM factory has a high level of responsibility together with a shitty salary.

2 - He could just be working on his sense of responsibility and self-sustainability (good for him!).

3 - Having wealthy parents doesn't automatically mean getting a big allowance, and hand-me-downs may not ever be possible if both parents use laptops at home (which is what most home computers are nowadays).


Yes, the 17 year-old post is a trigger-warning for the authenticity of that screenshot and everything else he claims in that thread, but the parent-works-for-OEM/AMD->he's-rich->kid's-rich->shouldn't-have-to-work-for-hardware logic isn't really valid outside some US states and EU countries.

So R7 360/360X should be Pitcairn/Curacao (GCN 1.0):
R9 270 == R7 360 == AMD6810 Pitcairn PRO
R9 270X == R7 360X == AMD6811 Pitcairn XT

If AMD occupies the whole $100-$200 range with a 3 year-old chip, they're bound to lose that range to nVidia. The GM206 is quite a bit faster, consumes a lot less and nVidia can place it in the same price range.
Here's hoping they don't do that.

 

[DmitryKo]

Iceland is definitely a 6 CU VI chip - basically updated Mars/Oland

If it's an assumption base on Iceland being R5 M25x/260x in CodeXL, AMD basically uses the same Oland chip for the whole R5 M230/M240, R5 240 and R7 M240/M25x/M26x, R7 250 lineup, and still uses VLIW4/Terascale2 for the lowest end R5 210/220/2300. Nobody cared until now, but since Nvidia has had fully Direct3D12 capable parts for the last 6 generations since at least GeForce 400 series (Spring 2010), that will become a disaster for marketing very soon.

They should instead bin Iceland with 128 bit GDDR5 for updated M230-260 parts and their 300-series counterparts, like they did with Cape Verde initially, and move Oland to the lowest end, so they finally have GCN parts with "full DirectX 12.0 support" throughout the whole line-up.

 

[DmitryKo]

If AMD occupies the whole $100-$200 range with a 3 year-old chip, they're bound to lose that range to nVidia.

Not if they move Pitcairn down to the R7 360 tier and price it accordingly ($130-150), so that R9 370 would be a $200-$250 Tonga card, R9 380 would be a $300-$350 Hawaii card, and R9 390 would be a $400-500 Fiji card.

 

[mczak]

They should instead bin Iceland with 128 bit GDDR5 for updated M230-260 parts and their 300-series counterparts, like they did with Cape Verde initially, and move Oland to the lowest end, so they finally have GCN parts with "full DirectX 12.0 support" throughout the whole line-up.

But Iceland isn't higher end than Oland (though I'd agree it's about time to move something like that to the low end). I don't know how large it is (it learned a few tricks) but presumably not much larger than Oland (if anything I'd expect them to have cut gddr5 capability as that was VERY rarely used on Oland, plus with fbc it may not be all that necessary).
The 6 CU VI part is confirmed among others by the open source radeon driver which has all that information there, so it should be quite reliable (there were also rumors about this way before that, it's even possible this thing is already shipping as the alleged part numbers are indeed all the same as Oland).

 

[DmitryKo]

The 6 CU VI part is confirmed among others by the open source radeon driver which has all that information there, so it should be quite reliable

Uhm... OK, I changed the table to make Iceland a clone of Oland.

I'd still assume Iceland will probably move down to R5 320-340 - so Bonnaire should also move down to R7 350, since R7 360 slot is occupied by Pitcairn now, and Cape Verde will be either retired or dedicated to mobile parts, just like it is now.

 

[lanek]

The season is now open: I have not get time to uplod them separetely , so sorry for cpu slide with it.


I like so much this 3D representation of HBM..

 

[silent_guy]

So... They have nothing new to report?

 

[lanek]

So... They have nothing new to report?

its only a part of official slides sended right now, but well effectively, nothing really new ( maybe its why they have send them to sites today )

 

[Kaotik]

The slides are available here: http://edge.media-server.com/m/s/xzfr4avf/lan/en (also the Q&A session is still live on webcast)

 

[3dilettante]

I'm not sure GCN's 2x energy efficiency after a node shrink and a shift to FinFET leaves much room for a significant change to GCN.
I'm not anticipating the competition going through the same transition doing measurably less than this, so while the improvement is welcome the relative positioning and competitiveness does not seem to change.

 

[DmitryKo]

300-series OEM cards are an almost direct rebrand of the 200 series, although a bit underperforming at lower end:

http://wccftech.com/amd-launches-ra...eon-r9-360-oems-based-tonga-pitcairn-bonaire/

http://www.amd.com/en-us/products/graphics/desktop/oem/r9
http://www.amd.com/en-us/products/graphics/desktop/oem/r7
http://www.amd.com/en-us/products/graphics/desktop/oem/r5


Could it be that the 300-series will be OEM-only, just like Radeon HD8000 series?

 

[Silent_Buddha]

Not holding my breath on this, but they are claiming that they'll be competitive with Intel again on the CPU front with the "Zen" cores coming next year (if they don't get delayed).

Regards,
SB

 

[Jawed]

The specs match what's on AMD's page which is linked from the article (although AMD's page doesn't list chip codenames and memory clocks).

Is there an OEM GTX960 with 4GB of memory? That might be the only way to sell R9 380. Gawd, what a horribly deceptive name.

Oh, and there's R7 300 series OEM GPUs, too:

http://www.amd.com/en-us/products/graphics/desktop/oem/r7#

 

[Grall]

I'm not sure GCN's 2x energy efficiency after a node shrink and a shift to FinFET leaves much room for a significant change to GCN.

Interesting observation! Not that it really matters to someone like myself though; I buy high-end GPUs for their performance primarily; power consumption is a relatively distant concern...

 

[3dilettante]

These are power-limited devices. So whatever amount of performance can be had per Watt is multiplied by the power budget, and the vendors have examples of cards at each power band.

 

[Jawed]

 

[lanek]

Well, good CG render skill from this AMD guy. I will bet for really light volume air density scattering ( ) + a simple DOF.. It is an excellent job on compositing and texturing.. ( funny, i will bet it is made on Blender, as the result is really similar of what could be done with it really simply, but cant say for sure )

 

[Deleted member 13524]

So it will be a short card after all.

Maybe that other card with the watercooler was the dual-Fiji?

 

[Deleted member 2197]

Uses 2.5D, not 3D stacked memory ....

HBA is a technology that uses multiple stacked DRAM dies with dedicated memory paths into a single package, cutting down on power-draw, thermals, and PCB materials. There will be three HBM memories available, 2Hi, 4Hi and 8Hi. If you look at a slide that AMD shared you will notice this is not '3D' stacked memory, but '2.5D'.

It is 2.5D because the the DRAM layers are stacked next to the graphics processor. The DRAM layers and interposer are connected with a vertical through-silicon via tsv-channels that rest on a base layer. Tsv's form paths through the silicium and offers shorter paths. According to AMD HBM opposed to GDDR5 will offer a threefold in performance per watt and will result into 50% less power consumption. Details about the cards and solutions using HBM AMD has not shared yet.

http://www.guru3d.com/news-story/amd-confirms-hba-high-memory-bandwith-for-graphics-cards.html

 

[Grall]

Uses 2.5D, not 3D stacked memory ....

Yes of course, as otherwise the GPU would nuke itself into oblivion from the heatload.