• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

wccftechAMD Pirate Islands : R9 300 Series Alleged Specifications Detailed

Page 9 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
3DVagabond said:
How did we decide it was the mem controller holding clocks back and not simply lower spec RAM? Why pay for 7G ram when you have a 512 mem bus that doesn't require RAM to be that fast?
Click to expand...

Not sure actually since 7950/70 could push ram clocks to 1.8ghz, well above the 1.2ghz that came stock on some of those cards and the R290 series. Most R290s have elpida so they just suck by default. Mine with elpida crashed with 1.35ghz ram clocks, which is horrid, since my 7950 with hynix ran fine at 1.7ghz.
 
Silverforce11 said:
Not sure actually since 7950/70 could push ram clocks to 1.8ghz, well above the 1.2ghz that came stock on some of those cards and the R290 series. Most R290s have elpida so they just suck by default. Mine with elpida crashed with 1.35ghz ram clocks, which is horrid, since my 7950 with hynix ran fine at 1.7ghz.
Click to expand...

what would you say a asus 270x direct cu 2 should be able to reach

i am able to get 1175 and 1500 stable on mine

1200 and 1550 both crash

this is all on stock voltage

should i increase the voltage?
 
3DVagabond said:
Why pay for 7G ram when you have a 512 mem bus that doesn't require RAM to be that fast?
Click to expand...


They don't need to go back with vRAM speeds once faster GPUs will need even more BW to achieve its potential. Staying with 384 bit buses on 3xx mm² class chip will even give them advantage in higher resolutions.... But higher buses means more vRAM modules on the PCB, then its more cost to AMD anyway.


3DVagabond said:
How did we decide it was the mem controller holding clocks back and not simply lower spec RAM?
Click to expand...


Is not about RAM clocks or even RAM bandwidth, but rather is about how well the GPU uses the available vRAM Bandwidth. Mem controller have nothing about with vRAM clockspeeds.
 
@buletaja

- The Xbox One's has two Shader Engines, each with 6 CUs.
----
I keep noticing that we have yet to see;

Maui
Iceland/Topaz
Tonga/Amethyst

I think Maui, Iceland, and Tonga are the hybrids of Volcanic and Pirate Islands.

Maui
Iceland
Tonga

Can do DX11.2(and lower feature levels) Pixel Shading and DX12 Compute Shading, while;

Bermuda
Fiji
Treasure

Can only do DX12 Compute Shading, for lower feature levels there probably would be a JIT/Runtime Compiler for Pixel Shader to Compute Shader.

---
Maui, Iceland, Tonga have the ACE F32 units but also have the GCP/Geometry Engine/Rasterizer/ROPs.

Bermuda, Fiji, Iceland have only have the ACE F32 units but do not have any fixed function pixel render pipeline. It is all compute based which explains why they want to move to HBM. Where you are BW constrained rather than ROP constrained.

---
Mantle(current)CI/VI/VPI;
Compute as frontend for graphics pipeline, Compute runs asynchronously ahead and prepares & optimizes geometry for graphics pipeline.

^-- Maui, Iceland, Tonga: Geometry Engine and Rasterizer removed for Compute Geometry Engine and Compute Rasterizer.

Mantle(future)PI/etc;
Compute is the graphics pipeline.

^-- Bermuda, Fiji, Treasure: ROPs removed and the graphics pipeline is fully compute.

---
The evolution:
(SI/CI/VI) GCP -> Rasterizer -> Geometry Engine -> CUs -> ROPs.
(VPI)ACE -> CUs -> ROPs.
(PI) ACE -> CUs.
 
Last edited:
norseamd said:
what would you say a asus 270x direct cu 2 should be able to reach

i am able to get 1175 and 1500 stable on mine

1200 and 1550 both crash

this is all on stock voltage

should i increase the voltage?
Click to expand...

your limits will be similar to 7870 users
Once you start going over 1100 core you usually hit the point of diminishing returns where you need increasingly more voltage.
My 7870 is running at 1100 core with 1.156v I need 1.2+ to hit 1150 likely 1.3v to hit 1200
1100/1500 was stable here, but I lowered the mem to 1250 to play it safe
 
NostaSeronx said:
@buletaja

- The Xbox One's has two Shader Engines, each with 6 CUs.
----

You are wrong
Look at the Compute CP, it is not ACE

it is derivative of F32

Do you think who is the one that help xxx build or research the VI/PI

look at the clue of 6 ops, are you forgot i posted HSA
Ops related to FMAC 256 operation (it is for PI/carrizo)
and one nextgen console

i posted this again
==============
check the CPU part 1 Core have 6 ALU
correspondent with 3 CU per CU is 16 Ex unit

APU-CPU6-GPU-768-660x660.jpg


1 CU = 16 x FMAC 256
1 CU = 16 x 8 ALU
1 CU = 128 ALU
1 CU = 2 times ALU

Future APU
=========
6 FMAC 256 of 1 module <====> 6 CU

or Compute cores =
1 Core 6 ops <===> 6 CU
8 core 48 ops <==> 48 CU

look at Mike Mantor slide how many F32 block it is 2 block
it is 64 thread

but the slide from Bryan Black show 32CU but 2 F32 block
because the CU is 128 ALU

The CU is composeable cores, 16 tiles, Current GCN is only 4 tiles Vec16 remember but note composeable
it is why Carrizo is said as Volcanic tiles
Click to expand...
 
Last edited:
Soulkeeper said:
your limits will be similar to 7870 users Once you start going over 1100 core you usually hit the point of diminishing returns where you need increasingly more voltage. My 7870 is running at 1100 core with 1.156v I need 1.2+ to hit 1150 likely 1.3v to hit 1200 1100/1500 was stable here, but I lowered the mem to 1250 to play it safe
Click to expand...

well shit

i could probably go to 1525 on stock voltage

would need full testing

my stock voltage is 1206 mv
 
1.6 Sounds ridiculously high for the VDDC or VDDCI of any modern GPU. For instance, ~1.1 is the norm at load for VDDCI on my 290x. Around 1.2 V for VDDC. I would not go over 1.3 V.
 
Pirate Islands GPUs - 28nm for Q4 2014
Fantasy Islands GPUs - 20nm or 14nm for 2015/2016

IMHO, 28nm 550mm2 or 700mm2 costs cheaper than 350mm2 20nm.
 
StereoPixel said:
Pirate Islands GPUs - 28nm for Q4 2014
Fantasy Islands GPUs - 20nm or 14nm for 2015/2016

IMHO, 28nm 550mm2 or 700mm2 costs cheaper than 350mm2 20nm.
Click to expand...

No way. AMD can't possibly do anything more at 28nm without a completely new architecture. We're just not getting anything new from them this year. It's that simple.
 
Techhog said:
No way. AMD can't possibly do anything more at 28nm without a completely new architecture.
Click to expand...

GCN 2.0 it is a new architecture.
150nm -- R200 >> R300
55nm -- RV670 >> RV770 >> RV790
40nm -- Cypress XT >> Cayman XT
28nm -- Tahiti GCN 1.0 >> Hawaii GCN 1.1 >> Fiji GCN 2.0 (?)

Fiji and Bermuda was leaked in HWiNFO32 in 2013, so they have working test samples on 28nm.
 
Last edited:
StereoPixel said:
GCN 2.0 it is a new architecture.
150nm -- R200 >> R300
55nm -- RV670 >> RV770 >> RV790
40nm -- Cypress XT >> Cayman XT
28nm -- Tahiti GCN 1.0 >> Hawaii GCN 1.1 >> Fiji GCN 2.0 (?)

Fiji and Bermuda was leaked in HWiNFO32 in 2013, so they have working test samples on 28nm.
Click to expand...

Don't argue with him. He said there won't be anything new from AMD this year, so there will be none!
 
The stream processors count seems way too high to be true, I mean 4224 on the 390 is crazy, I can't see such high count.

I would love it to be true, I mean we definitely need a major jump, since in the past 3 years all GPU's have been the same, with barely any improvements.

I mean the whole GTX 7 series and AMD R series are basically slightly improved GTX 6 and AMD 6000 series GPU's.

So hoping to be true and hoping to see major price cuts of all graphic cards.
 
Slick, I have to disagree with you, because there have been massive improvements. I've used the 7950 and now R290. Both are 28nm and the performance gain is massive. Likewise for numbers from gtx680 to 780ti.
 
SlickR12345 said:
The stream processors count seems way too high to be true, I mean 4224 on the 390 is crazy, I can't see such high count.

I would love it to be true, I mean we definitely need a major jump, since in the past 3 years all GPU's have been the same, with barely any improvements.

I mean the whole GTX 7 series and AMD R series are basically slightly improved GTX 6 and AMD 6000 series GPU's.

So hoping to be true and hoping to see major price cuts of all graphic cards.
Click to expand...

Well actually, Id say that the GTX 780 Ti is roughly ~50% faster across the board vs a GTX 680 (Just check the AT bench tool). Same story with 7970GHz compared to the R290X. The improvements are there. Its just that its taking alot longer than what the industry was like 5~10 years ago hence the perception that nothings improved much.
 
I think the issue is that we can still buy 680's (rebadged as 770's) and 7970's (again rebadged as 280X's) over 2 years later as current products. That makes it seem like nothing is advancing.
 
AMD will not have 28nm (EDIT: I meant 20nm!) GPUs this year (Jan. 1, 2014 to Dec. 31, 2014). AMD's fiscal year is the same as calendar year.

http://seekingalpha.com/article/214...-2014-results-earnings-call-transcript?page=5

David Wong - Wells Fargo

Great. And when will we expect you to introduce GPU products at a node below 28 nanometers? Will have you have any 20-nanometer GPUs this year or next year?
Lisa Su - SVP and General Manager of Global Business Units

David, I think what I said earlier sort of what we're doing in terms of technology strategy, we are 28 this year, we have 20-nanometer in design, and then FinFET thereafter. So that's the overall product portfolio.
David Wong - Wells Fargo

That includes GPUs, Lisa?
Lisa Su - SVP and General Manager of Global Business Units

That's the overall product portfolio, so I'm not being specific about graphics versus other products.
 
Last edited:
You must log in or register to reply here.

TRENDING THREADS

Back
Top