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

[dbz]

In HardWare.fr's latest review, the GTX 980 is only 8.7% faster at 2560×1440:
http://www.hardware.fr/articles/935-19/recapitulatif-performances.html

That seems like a reachable target for a mild refresh. As for 4K, it might be doable with a Crossfire.

Couple of points there: The Sapphire Tri-X in the test is already overclocked. By recent accounts, the 390X has a 2.88% clock speed advantage over the Tri-X.
The second point is that if you average the averages of all the games tested, the 980 isn't 8% (100 / 92 ) faster but 9.92%, so I'm not too sure how a 20% rounding error managed to appear.

 

[pharma]

Couple of points there: The Sapphire Tri-X in the test is already overclocked.

Agreed. In this graph it'd make more sense to compare apples-to-apples, or the Sapphire Tri-X to the Gigabyte 980 G1 Gaming.

 

[Nemo]

Couple of points there: The Sapphire Tri-X in the test is already overclocked. By recent accounts, the 390X has a 2.88% clock speed advantage over the Tri-X.
The second point is that if you average the averages of all the games tested, the 980 isn't 8% (100 / 92 ) faster but 9.92%, so I'm not too sure how a 20% rounding error managed to appear.

2.88% -- GPU Clock speed over the Tri-X
15.4% -- Memory Clock speed over the Tri-X

 

[Silent_Buddha]

Agreed. In this graph it'd make more sense to compare apples-to-apples, or the Sapphire Tri-X to the Gigabyte 980 G1 Gaming.

Not really as the Sapphire Tri-X is going to be reportedly slower than the reference 390X. And presumably there will be OEM overclocked versions of the 390X with better performance. So based on pre-release information/rumors it's a valid comparison to try to figure out where the reference 390X will slot in (comparing potential reference 390X to reference 980).

Regads,
SB

 

[pharma]

Not really as the Sapphire Tri-X is going to be reportedly slower than the reference 390X. And presumably there will be OEM overclocked versions of the 390X with better performance. So based on pre-release information/rumors it's a valid comparison to try to figure out where the reference 390X will slot in (comparing potential reference 390X to reference 980).

Oh, I thought Sapphire Tri-X was higher clocked than the upcoming 390X reference since most rumors/info I've seen are 390X non-reference specs. Just a few more days ....

 

[wiak]

i wonder how the effect of the Full Tonga with 6Gbps GDDR5, Color Compression and higher memory amount will have on bandwidth/performance

 

[wiak]

Hallo, guys!

First of all, I would like to tell you that I am back after some quite long absence.

Unfortunately, I see some quite strong confusing and negative (in general and against AMD) things circling around, probably coming from the nvidia pr trying to hurt AMD.



AMD Confirms GCN Cards Don’t Feature Full DirectX 12 Support – Feature Level 11_1 on GCN 1.0, Feature Level 12_0 on GCN 1.1/1.2

Read more: http://wccftech.com/amd-confirms-gc...-10-feature-level-120-gcn-1112/#ixzz3cDXM9ujS

Nonsense to which the response is this:

Roy@AMD @amd_roy
@Ramzinho @Thracks zero, absolutely zero. AMD supports DX12. Period.

Anyways - how many times in the past did we see how nvidia didn't support some versions of DX (DX 10.1, DX 11.2, DX 11.3, etc)?

In this case the feature level doesn't even mean that those support DX12dot1 but DX12 feature level 12_1...

AMD DirectX 12 GCN Support:

Model Graphics Core Next Architecture DirectX
Radeon HD 7000 series GCN 1.0 DX12, feature level 11_1
Radeon HD 7790 GCN 1.1 DX12, feature level 12_0
Radeon R7 260 (X) GCN 1.1 DX12, feature level 12_0
Radeon R9 270 (X) GCN 1.0 DX12, feature level 11_1
Radeon R9 280 (X) GCN 1.0 DX12, feature level 11_1
Radeon R9 285 GCN 1.2 DX12, feature level 12_0
Radeon R9 290 (X) GCN 1.1 DX12, feature level 12_0

NVIDIA Direct X 12 GPU Support:

Model DirectX
GeForce 900 Series (Maxwell 2.0) DX12, feature level 12_1
GeForce 700 Series (Maxwell 1.0) DX12, feature level 11_0
Partial feature level 11_1 support
GeForce 700 Series (Kepler) DX12, feature level 11_0
Partial feature level 11_1 support
GeForce 600 Series (Kepler) DX12, feature level 11_0
Partial feature level 11_1 support
GeForce 500 Series (Fermi) DX12, feature level 11_0
Partial feature level 11_1 support
GeForce 400 Series (Fermi) DX12, feature level 11_0
Partial feature level 11_1 support

Read more: http://wccftech.com/amd-confirms-gc...-10-feature-level-120-gcn-1112/#ixzz3cDaSCiXA

this means nothing, until windows 10 is out and all the *working* drivers and cards are out...

 

[wiak]

Old 7850 could perform at that level, but that would mean Radeon 370 is even slower than 270. At the same time it is too high for Bonair. Maybe there is still hope for new chip.

thank you for flying BonAIR

 

[wiak]

This time there might be a full Tonga thrown in, and an overclocked Hawaii with extra RAM (assuming this is what "enhanced" means). I don't know if that will be enough, though.

In particular, it's going to be hard for AMD to argue that Fiji doesn't need any more than 4GB of RAM but that Hawaii really benefits from 8GB.

i do think AMD will release 8GB HBM2 Fury cards later on, just that the first batch is 4GB HBM versions
memory amount dont always matter that much, case in point, do they (both nvidia/amd) still sell underpowered cheap cards with stupid amount of memory?
and amd has stated that their memory usage optimization is pretty bad on pre-fury cards (they didnt trow developers/engineers at it), mainly due to having alot of GDDR5 memory to play with

this year is heating up to being pretty interesting in the tech world, with Fury, DX12, Vulkan, Win10, DDR4, NVMe/PCIe, USB-C 3.1

 

[hoom]

"The first thing I noticed is the absence of Fury."
Are we sure Fury isn't AMD internal card reference rather than a brand?

 

[wiak]

"The first thing I noticed is the absence of Fury."
Are we sure Fury isn't AMD internal card reference rather than a brand?

i still got a batmobile in the closet

 

[SimBy]

i do think AMD will release 8GB HBM2 Fury cards later on, just that the first batch is 4GB HBM versions
memory amount dont always matter that much, case in point, do they (both nvidia/amd) still sell underpowered cheap cards with stupid amount of memory?
and amd has stated that their memory usage optimization is pretty bad on pre-fury cards (they didnt trow developers/engineers at it), mainly due to having alot of GDDR5 memory to play with

this year is heating up to being pretty interesting in the tech world, with Fury, DX12, Vulkan, Win10, DDR4, NVMe/PCIe, USB-C 3.1

How easy it is to actually replace HBM with HBM2? Any changes to Fiji memory interface required?

 

[silent_guy]

I don't think it requires major changes. But I doubt it's very useful in terms of performance, so you'll only get the memory increase.
HBM2 availability is only expected sometime next year, I believe? At that point, you wonder if there's still a point: if AMD and Nvidia are going to have their big next-gen process die ready at a similar time, it will wipe the floor with Fiji. Maybe they'll let Fiji live for a long as 28nm is cheaper. But from a competitive point of view, the 8GB will probably be too late.

 

[Esrever]

Its weird how suddenly people feel 4GB wasn't enough any more even without any real proof of games needing more. 980 was fine as a top end part with 4GB for a while, suddenly when titan x launches with an absurdly high amount of vram, 4GB suddenly isn't enough?

 

[Grall]

@Esrever It's also the thing where even current games can max out a 4GB card at highest settings, and who knows what'll happen with future games.

 

[RecessionCone]

Its weird how suddenly people feel 4GB wasn't enough any more even without any real proof of games needing more. 980 was fine as a top end part with 4GB for a while, suddenly when titan x launches with an absurdly high amount of vram, 4GB suddenly isn't enough?

Pretty hard to argue 3.5 GB is far too little, while 4 GB is comfortably enough.

 

[Entropy]

@Esrever It's also the thing where even current games can max out a 4GB card at highest settings, and who knows what'll happen with future games.

It depends on your priorities in gaming. Do you want a high performance card to play with all settings maxed out at high resolutions, or do you want to achieve consistently high framerates with good image quality?
If it's the latter, then it's difficult to see that 4GB is any problem at all. If you fall into the first category, you might want to sell a 4GB card within a couple of years once FinFET + HBM has dropped from novelty pricing. Personally, my main interest in substantially increased graphics performance would be to drive VR-equipment, since the current console generation will largely determine general asset quality levels for the forseeable future.
Thus, memory footprint is likely to stay below 4GB, as framerates will target staying solidly above 90fps.
Again, YMMV depending on priorities.

 

[CarstenS]

It also mentions "new functionality", though (which could refer just as well to new hardware features as to new software features, but it leaves the window open)

New functionality is under the energy efficient bullet point. Even an additional power-state would be new functionality in this regard.

I would never even worry about products on 28 nm anymore. They are just a stop gap, while the wait for next-gen process is not so much anymore.

The year for AMD is meant to be 2016 with Arctic Islands and Zen.

I am not sure, if it is finacially feasible do worry about business only every other year.

--
Rather on-topic:
Any estimates on how much space savings could be possible due to lower clocking for the memory interface per 64/128-bit partitions? I mean, it's eight times as wide after all.

 

[Rurouni]

What I want is at least a 3GB Tonga. I really wanted to buy a 285, but what stopping me is the 2GB memory. Is Tonga currently the most complete AMD GPU in terms of feature set?
Btw, how is the possibility for AMD to call Hawaii (and others) purely based on the core itself (memory controller, max number of CU) and add features that exist in Tonga (compression, improved tessellation, better power consumption, etc) while still call the chip Hawaii?
Probably a better question would be will AMD use the improvement on Tonga for all 3xx series or it would still tied to Tonga? I imagine some improvement will be applied across the 3xx series (like better power consumption), but feature wise it would be the same as their 2xx equivalent?

 

[gamervivek]

It also mentions "new functionality", though (which could refer just as well to new hardware features as to new software features, but it leaves the window open)

It's GCN1.1, unless AMD are improving GCN by rebranding it. "2nd generation" was also bandied about during Hawaii launch.

And the R7 370 "rebrand" and for that matter the 380 and 390 "rebrands" apparently supports VSR alongside 290 and 285.

Unless AMD are using it to differentiate between the 3xx and 2xx cards, at least the rebranded GCN1.0 cards that don't get a mention, it very strongly hints to the new gpus at least having the hardware scaler like the other GCN1.1 and above gpus.