• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD VEGA confirmed for 2017 H1

Status
Not open for further replies.
benjii said:
Rumour has it that the Polaris 10 refresh will hit very soon. On that note, Bethesda and AMD have done a fair bit of advertising as to how Prey is optimised for Vega. Prey is releasing 5/5/17, it would be pretty odd if Vega wasn't available at the same time.
Click to expand...

OMG that stupid reference cooler again...

N19h7m4r3 said:
Hmm, the last die shot only shows two HBM modules; and HBM2 is 4GB a module.

a065dcefdbe14c64996b15ed262a16eb.png
Click to expand...

as far as i know it depands on the Hi stack, HBM2 can have 4GB or 8GB per stack, depanding on on the Hi stack chosen of 4 or 8, so 2 stacks can be 8GB or 16GB, my guess reference Vega will be using 2 and 4-Hi stack for 8GB.
the 4GB version would be 1 stack of Hi-4
i dont know the freedom AMD would give AIBs, if they can put 2 stacks of Hi-8 that would give them 16B of HBM2, but my guess 8-Hi stack must be much more expensive, so the premium could be significant if any AIB thinks it's worth it, if Vega is oriented value perf/$, i doubt any AIB would risk it, but if by any miracle it takes performance crown over 1080Ti, a premium 16GB card becomes more likely to happen.
or go 8GB HBM2 + 8GB GDDR5X, since AMD already said that Vega was a hybrid GPU compatible with both memory, although sounds less appealing than full 16GB HBM2
 
Last edited:
AlamoX said:
as far as i know it depands on the Hi stack, HBM2 can have 4GB or 8GB per stack, depanding on on the Hi stack chosen of 4 or 8, so 2 stacks can be 8GB or 16GB, my guess reference Vega will be using 2 and 4-Hi stack for 8GB.
i dont know the freedom AMD wuld give AIBs, if they can put 2 stacks of Hi-8 that would give them 16B of HBM2, or go 8GB HBM2 + 8GB GDDR5X, since AMD already said that Vega was a hybrid GPU ompatible with both memory.
Click to expand...

I doubt AIB's would have any way of adding more VRAM. The HBM is on the main solution AMD will be supplying them, so if there are 16GB versions, they'll be rather expensive and rare I believe.

opethdisciple said:
You can see it referenced in the spec table here. Also mentions it here.
Click to expand...

Sadly none of those are official spec tables from AMD; and best not to trust anything drawn up by WCCFtech. They're well known for lying to get clicks.

Let's not forget them convincing most of the web that Polaris will be Fury or GTX 1070 speeds on launch, with easy overlocks on reference to 1500Mhz.

Best to ignore those sites, and just sit tight and wait for official word.
 
N19h7m4r3 said:
I doubt AIB's would have any way of adding more VRAM. The HBM is on the main solution AMD will be supplying them, so if there are 16GB versions, they'll be rather expensive and rare I believe.
Click to expand...
well AMD's choice of 4-Hi stacks is mainly the cost, i don't think there is any technical difficulty for 8-Hi stack, so technicaly if the AIB is confident he could sell the same GPU with ~200$ premium over the 8GB cards, AMD would be able "technicaly" to make it happen.
 
Rroff said:
Wasn't an invention of wccf - AMD published (or they were leaked) slides with stats for 0.85v which if referenced against Samsung 14nm at ~1.2v suggested 1500MHz was possible.
Click to expand...

WCCFT stated they had working 1500Mhz cards, and that Factory OC'd ones will be at 1400Mhz without issue., and they even went so far as to say AMD was trying to "bribe" them with hardware to remove those "leaks".

Also it's a bit much to extrapolate 1500Mhz from 1.2v, as sure they could, but the heat output and power requirement wouldn't make it the RX 480 that we have.
There's a difference between stating "yeah we have Fury/Nano to GTX 1070 performance here on reference and AIB air cooling at 1500Mhz", and stating, that the process at 1.2v could allow up to 1500Mhz potentially.

The AMD Radeon RX 480 Is Capable Of ~1.4Ghz And Can Hit 1.5Ghz+ On Air
Now, all of these are just meaningless numbers without a performance comparison. So to put it all into perspective, an overclocked RX 480 will actually perform within striking distance of AMD’s current R9 Fury X flagship. At less than one third of the cost. If that’s not a good enough point to end on, I don’t know what is. It certainly looks like AMD may have hit a home run with its 14nm Polaris graphics chips.
Click to expand...
 
Last edited:
JediFragger said:
My only fear is that they'll again push Vega close to its absolute core clock limit and leave us tweakers with nothing to play with/gain :(
Click to expand...

I agree, it's pretty likely from what we've seen so far. GloFo's process at least appears to be a bit pants compared to TSMC and Samsung this generation, so AMD's been pushing everything to the limit at stock.
 
AlamoX said:
this is not the first time iv read seekingalpha articles about some cpu/gpu stuff, i am pretty convinced that who ever write this stuff, doesn't know what he is talking about.
Click to expand...
They really dont have a clue.

Claiming that good VR performance isn't possible without 'high end' memory tech is incredibly ridiculous.
 
N19h7m4r3 said:
Hmm, the last die shot only shows two HBM modules; and HBM2 is 4GB a module.

a065dcefdbe14c64996b15ed262a16eb.png
Click to expand...
What's the limit on the amount of memory chips that can be stacked? I thought I read 8 was max, but I dont think that was an official source.

And what are the largest memory chips at right now? Have we progressed beyond 1GB yet?
 
Seanspeed said:
What's the limit on the amount of memory chips that can be stacked? I thought I read 8 was max, but I dont think that was an official source.

And what are the largest memory chips at right now? Have we progressed beyond 1GB yet?
Click to expand...

EXxx said:
8 stacks was the max for HBM2 when the specification paper went public (the Hynix memory one)
Click to expand...

Yup, HBM2 is 4GB and 8GB per die. Although from what I understand the 8GB modules are abhorrently expensive; which is most likely why NVIDIA only have HBM2 on their GP100 Teslas.

I would really like if AMD managed 16GB of HBM2 on top end Vega, but having them mention they're going to have a 4GB card makes me worried we'll be limited to 8GB max on the top card simply because of cost.
 
EXxx said:
8 stacks was the max for HBM2 when the specification paper went public (the Hynix memory one)
Click to expand...

N19h7m4r3 said:
Yup, HBM2 is 4GB and 8GB per die. Although from what I understand the 8GB modules are abhorrently expensive; which is most likely why NVIDIA only have HBM2 on their GP100 Teslas.

I would really like if AMD managed 16GB of HBM2 on top end Vega, but having them mention they're going to have a 4GB card makes me worried we'll be limited to 8GB max on the top card simply because of cost.
Click to expand...
Thanks.

The way the tech works, the costs should scale pretty linearly, which would suggest that even the 4GB stacks are very expensive on their own.

That said, it's still very possible that 8GB's of super high bandwidth memory could deal with pretty much anything thrown at it for the next 2-3 years, even at 4k.

I agree that 12-16GB would make people feel a lot more comfortable, though.
 
By the time 8GB isn't enough (especially if HBCC works as well as they showed it to) then 16GB should be significantly cheaper anyway.

We may well see 16GB cards in 6 months or so after prices have come down a bit and Vega has sold a good amount.
 
Status
Not open for further replies.
Back
Top Bottom