AMD RX 400 series (Polaris) MegaThread! FAQ & Resources

Page 39 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


HBM will significantly up bandwidth, add on a few compute units and you have a Big Polaris, there is no sense using a Polaris coding and a high end GPU code x90 is normally reserved for highest end part. Leading the deduction that it is a beefed up Polaris rather than a "Small Vega".

 


AMD have confirmed there were only ever 2 Polaris gpu dies produced. That means if it's Polaris based- it's going to be dual gpu and it *won't* have HBM (the memory controllers are totally different, you can't run HBM off of the GDDR5 based controlers already in Polaris).

If it's got HBM then it is either a new Fiji based gpu (hopefully with HBM 2 for a decent capacity this time), or it's going to be Vega.

Whilst this might not make sense at first- it actually follows with what AMD have being doing recently- they release a range of cards in stages, with different gpu's being added and other gpu's being re-positioned accordingly. Just look at their previous releases: Hawaii was added to the top of the R9 200 series, the rest of the range being based on earlier GCN gpu's. Then Fiji was released above Hawaii.

Now they have replaced the entire lower end of the stack with Polrais- it would follow previous convention for them to slot Vega in above Polaris rather than outright replace it. I mean if you look at AMD's product stack vs nVidia they have 2 gpu's to nVidias 4 so there's definitely room to slot a couple of Vega parts in there.
 


I think the idea is there would be two Vega GPUs of different size and performance level. The same way there were two Polaris GPUs.

Then AMD would have 4 modern GPUs to compete with Nvidia's GP102, GP104, GP106, and GP107.
 


Exactly- the other thing all the articles keeps getting wrong is they have heard there is Vega 10 and Vega 11- and that Vega 11 is small because Polaris 11 was the smaller of the two gpus- AMD however already explained that the Polaris numbering was based on *the date when the gpu's were designed* and NOT a convention relating to size.

On that basis, Vega 10 would be the *first designed Vega part*- which means we have no idea of the specifications of Vega 11.

So far Polaris 11 -> GP107, Polaris 10 -> GP106, so in theory (assuming the benchmarks are true) Vega 10 -> GP104 and Vega 11 -> GP102. There was also mention of a 'Vega 20' although that doesn't make much sense based on AMD's date basis for the numbering (unless '20' refers to gen 2?).
 
Funny thing is, wasn't Vega just a re-work of the Fiji GPU?

So I would imagine the "Fiji 2.0" and "brand new GPU" are not that un-related. Case in point: Polaris is as new as Tonga was to Tahiti, I'd say. Well, it's still a bit unfair to say, but most of the improvements were thanks to the die shrink IMO. Polaris is not a revolutionary re-work of GCN in the slightest. I'm really not expecting Vega to be worlds apart from Fiji, TBH.

Cheers!
 


Thanks, although you're a little late to the party: That's the benchmark we've been discussion- the concern is that it's sitting below a 1080 (although when you look at how close all the cards are to each other it looks as if the benchmark in questions is platform bound so it's not really a good comparison anyhow).
 


If it's below a 1080 then it'll cost less than a 1080.
 
All i know about vega is that is is GCN 2.0 official, not GCN rehashed from Tahiti to Polaris, it is a brand new architecture hence my opinion that that is likely a rehash of something whether a Fury X 2.0
 


Polaris is GCN 4, Vega was originally expected to be GCN 4 as well, but some rumors have indicated it's actually GCN 5.
 


As I understand it the Vega GPU is truly a redefined architecture like nothing AMD has had before, I would term Polaris as GCN 1.4 as the architecture is the exact same one from the Tahiti days, much like Fermi and Kepler was the exact same architecture just more refined.

 
This is getting weird thanks to marketing terms, lol.

What we know: current Polaris is GCN revision 4 (implying a new GPU underneath). Differences from GCN revision 3 (Fiji?) are larger than differences from GCN revision 2 (Tonga?), right? That implying it's a part "tweaked" on top of a Tonga design than a Fiji design. At least, I remember that was the impression when reading the release notes in all sites.

That being said, whatever Vega ends up being, if the marketing team decides to slap "2.0" to it, for me (and you guys as well IMO) it will be "GCN revision 5" and it all points to be a closer revision/re-work of Fiji than a beefed up Polaris; as in, closer to GCN revision 3 than GCN revision 4.

And you can exchange "revision" for "generation" or "version" or "candy polar bear". I actually like "Candy Polar Bear".

Cheers!
 
I have to put now a list to remember it fully:

7970Ghz/280X - Tahiti -> GCN v1 big
7870/270X/370X - CapeVerde -> GCN v1 small
290X/390X - Hawaii -> GCN v2 big only?
285/380X - Tonga -> GCN v3 small
Fury X - Fiji -> GCN v3 big
RX480 - Polaris 10 -> GCN v4 big
RX460 - Polaris 11 -> GCN v4 small
RX490 - Vega 10 -> GCN v4 bigger OR GCN v5?

That's how it is, then?

EDIT: So for Vega, would those options be a better "concept" to follow forward? 😛

Cheers!
 


The 7870 etc. used Pitcairn, which was the mid-sized GCN 1 chip. Cape Verde was the small one used in the 7770 etc.

GCN 2 added two GPUs, Hawaii the big one and Bonaire the small one, used in the 7790/260X.
 


Ah! Forgot about those 2. Thanks for that.

So if GCN v1 is of any indication, they can perfectly go "small, medium, large and larger" with this iteration.

Full list then is:

7970Ghz/280X - Tahiti -> GCN v1 big
7870/270X/370X - Pitcarin -> GCN v1 medium
7700 - CapeVerde -> GCN v1 small
7790/260X - Bonaire -> GCN v2 small
290X/390X - Hawaii -> GCN v2 big
285/380X - Tonga -> GCN v3 small
Fury X - Fiji -> GCN v3 big
RX480 - Polaris 10 -> GCN v4 [strike]big[/strike] medium
RX460 - Polaris 11 -> GCN v4 small
RX490 - Vega 10 -> GCN v4 big?

EDIT: Stroke incorrectly 😛

Cheers!
 


It is a money spinner for AMD, not only does the contract pay up front, Intel also pay royalties per chip sold so this is good money. Intel have never really challenged AMD with iGPU's, Iris does but does so at a huge cost, I mean a Haswell Iris based i5 in my country costs what a i7 4790 costs, for lower clock speed. I can just buy a RX460 and get a i5 4560 for less and deliver about 60% better returns.

 




will this going to benefit AMD? because that licensing deal is only part of the story. another part i heard was RTG themselves want to break away completely from AMD (take it what you will. the old part of ATI not liking AMD management is not a new story). i think kyle even suggesting RTG wants to be part of intel and not simply licensing those graphic tech to intel.

in any case intel licensing deal with nvidia will expire early next year. and when looking at nvidia and AMD right now nvidia is more of a threat to intel than amd especially on HPC and professional space. so intel probably did not want to keep pumping more money towards nvidia. and the licensing deal part of it just an a made up excuse to settle the issue with nvidia back then and to forever seal nvidia from getting x86 license. and intel graphic engineer once mention that the licensing with nvidia is just so they can make their own gpu with out infringing nvidia patent illegally. they did not use nvidia developed tech in their gpu. so this licensing with AMD could be the same deal. some people suggesting AMD probably offering better deal than nvidia.
 
Hey people...

Can't help wondring if that 10xt2 is a dual GPU Polaris an thats what we seen in AOTS, an Vega aint happening as soon as were hoping... :-(

Just a taught... hopefully not the case... but it is possible I guess....

Someone please convince me otherwize.

Jay

PS (edit) I noticed over the last while a lot of 'crossfire' fixes an updates an such bein released... is this to pave the way for a dual GPU card, namely a Polaris one (10 being 'Polaris 10' xt2 meaning dual GPU) ??

PPS Ok scratch that think Vega is ready lads !!! it does kinda come in round the end of the year on their own roadmap, actually it would be bout spot on timing wize, by the roadmap that is... put me on the VIP list for the party wup wup ..
 
AMD could very well be ready with Vega design. but they probably waiting for HBM2 mass production. also i also thinking that AMD did not want to replace the original Fury X only after a year hitting the market. Vega most likely AMD flagship gpu for 2017-2018 time frame.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom