New AMD roadmap shows Vega and Navi Architectures

That looks like a interesting road map. If all goes well I might buy that Vega thing.

 

So, without HBM2 does that mean only 4gb on Polaris based cards? I mean it could make sense, if they don't plan to increase performance much, at lower resolutions, but just reduce powerconsumption and heatoutput.

 

Couldn't they do two stacks of 4GB HBM1 on a card? Honestly, not sure if they need HBM2 for this year's releases.

 

Why not GDDR5X like Nvidia are using?

If HBM1 still has limitations of 4GB why not just go for GDDR5X for the extra bandwidth?

Its reported to still perform around 2-3x of current GDDR5 so GDDR5 basically matches HBM1 for performance since that only reach 1Gbps and HBM2 is rumoured to be around the 2Gbps range as it uses larger denser memory stacks.

 

Polaris not even release and already "showcasing" the two future generations?

Something's not right. I had(have?) high hopes for Polaris but this seems a bit discouraging.

 

Well imho GDDR5X isn't bad. It's overclockable, cheaper than HBM, and gives enough performance. I just thought they were able to stick two of those 4GB modules on a PCB and connect them both to the GPU, but maybe I imagined that to be too easy.

Why so? They just try to bring back confidence and trust in their release schedules I think

 

If they use GDDR5X with Polaris it will be fine. HBM2 would be better but 5X will do the trick for middle to high end cards. Vega then with HBM2 would be the "big pascal" equivalent.

 

Will GDDR5X even be available in time and in volume?

EDIT: was referring to Polaris

 

Very little chance of GDDR5X. It's HBM, hopefully v2 HBM, double capacity.

OR...
it's GDDR5, which is not that far fetched considering perf/W numbers.

2.5x perf/Watt of R9 285(GCN1.2, Sep.2014) means 40-50% better than GTX 980, which is not particularly high target considering 28nm->14nm,
but is line with that another roadmap perf/W jump in 2017 with HBM2, ie. Vega.

 

I read yesterday on the topic of the Geforce 1080 rumor that the company responsible for GDDR5X would be in mass-production in spring 2016.

 

Looks like HBM2 is not quite ready in general.

 

I wonder if it's HBM2 not being ready, or big die sizes worthy of HBM2 not being ready.

I mean they showed Polaris 10 running Hitman at roughly the same performance as a 980Ti. And it's not like slapping HBM2 on a 980Ti is going to do much other then increase cost and slightly reduce power. If Polaris 10 is AMD's largest card for this year, they can easily just throw GDDR5x on it instead of HBM and hit nearly every performance target while slightly reducing power/cost.

I'm sure Nvidia is in the same boat, hence why rumors about the GP104 using GDD5x as well.

Why go through all the pain of using an interposer with HBM2 if the the manufacturing is limiting you to midsized chips?

 

So, based on the release date and the use of HBM2, Vega would be the new fury?

 

Define "fury"

Polaris 10 has roughly the same performance as Fury and probably has HBM1 on it. So I'd argue that Polaris 10 is the new Fury.

But if you mean "Fury" as in card that's pushing technology, then yeah, Vega probably first cards to use HBM2.

 

I think we are all guilty of rushing to make assumption. The more we here it, the more we expect it. Patience is not allows one of our virtues.

 

I don't think we are going to have high end Polaris I think it will be lower end low power notebook/midrange I don't even think they will have hbm on them maybe next year for what we want?
I know I know but....that would be crazy for amd to put out a high end card in the next few months(besides the frankin card crap),they have to work on the 3xx and lower replacements like why have them power piggy cards in the line ups if you are boosting about lower power use?(that all they have said about Polaris so far tbh).
Samsung can do exactly that for them small die chip that has performance.

just saying of coarse no one really knows no matter what amd says anyway

 

What about Navi and "Nexgen memory"? What could that be? And its not like its far away, less than two years time.

 

^^ Probably HBM3 or something, efficiency improvements maybe. Can't be anything else really can it? I mean HBM is designed to beat a bottleneck that doesn't even exist yet so unless AMD are planning some top tier magic ****, memory won't be a bottleneck again for ages.