Printable View
They're not just a rebrand though. It's not like they've taken some leftover stock and tweaked the firmware to report a different model name. The chips are now being produced at a different fab (GloFo rather than TSMC iirc), who have a better 28nm process. Whilst the architecture is the same they're now more efficient, meaning they clock higher.Quote:
Originally Posted by GrimMachine
Up to, Up to, Up to...No benchmarks. Let me when you figure out what your products REALLY are. I want to hear things like "it has X", it's up to something. More like, "it runs at X speed ALWAYS", not up to again...flashbacks to 290x running at 700mhz again...LOL.
I expected a REAL launch instead of paper today. No reviews. No benchmarks...blah...I was all excited today until I saw NO CARDS reviewed. Instead another delay, which to me means they have not cleared the shelves of old cards yet, which just means another terrible quarter for AMD shortly, or worse they can't make enough cards for a launch? Not sure which is worse, but today is just bad.
Do we know when the NDA for benchmarks will be lifted?
Damned if they do and damned if they don't!
Sure it could.Quote:
Originally Posted by abaxas
WARNING! NVIDIA 196.75 drivers can kill your graphics card
Nvidia 320.18 WHQL Display Driver is Damaging GPUs
deleted
Hardly the products of a New Era of PC Gaming though are they, basically 2 year old architecture refreshed which is almost a rebrand.Quote:
Originally Posted by Bagnaj97
Surely they should have put that tag line on the Fury X.
Presumably on the official launch dates, so tomorrow for the 300-series cards, 24th June for the Fury X, 14th July for the Fury. Currently unknown for the Nano (I note that AMD's slides call it just the R9 Nano, not the R9 Fury Nano...).Quote:
Originally Posted by Biscuit
Won't surprise anyone to know that I'm most interested in the Nano from this lot: wonder how much they've cut it to get the TDP down. If they are using the same GloFo 28nm process that Carrizo uses, they could potentially keep quite a lot of the performance whilst lowering TDP significantly....
Not going to jump on the cynicism bandwagon here, I'm still excited. After what seems like years of re-brands and refreshes, I personally think it's brilliant to see the imminent launch of what feels like a genuinely new and exciting GPU range (Fury cards obviously. 300-series = snore).
Especially interested to find out more about the Quantum PC, that looks like a lot of fun :mrgreen:
I'm also most interested in the nano. Improved perf/watt suggests a lower end part, pricing will be interesting.Quote:
Originally Posted by scaryjim
2x perf watt of the 290X.. to me that sounds like roughly same performance at half the peak power.
Is there a gap in the market? It is exactly where I want a card to be.
R7 360 $109 (£70)
R7 370 $149 (£95)
R9 380 $199 (£126)
?????
R9 390 $329 (£209)
R9 390X $429 (£272)
£ is today's rate and a direct exchange not taking into account VAT or anything else.
In the US one of the cheapest 970 is a Zotac priced at $300 (£190). In the UK the same model is £250. So there'll be another £60 minimum on that R9 390 price in the UK. A 4Gb 390 or a 380X would fit the bill.
Rumours are that the Tonga GPU that the R9 285/R9 380 are based on is a cut-down version of the full architecture, and that the full-fat chip (2048 shaders and a 384-bit memory bus) will be released as the R9 380X in due course. That would be the card to fill your gap in the market, if it exists.Quote:
Originally Posted by iranu
That would make sense. Joe Macri mentioned AMD plan to use HBM across the range so maybe thats why the card wasn't in the line up.Quote:
Originally Posted by scaryjim
I don't think we'll see the R 300 series getting HBM across the range - that'd need some significant reworking of the silicon. For this generation we'll get HBM-bearing Fury cards and GDDR5 300-series cards.Quote:
Originally Posted by jigger
I suspect when we see HBM trickled further down the stack will depend entirely on when HBM2 is ready to ship - not because of bandwidth, but because of capacity: HBM1 is limited to 1GB per stack. AMD are talking about 4GB of RAM for the R9 380; doing that on HBM1 means putting 4 memory controllers on the GPU die: that's a lot of silicon generating a completely unnecessary amount of bandwidth just to meet the 4GB capacity target. I doubt the economics on it add up at all.
The second gen HBM will have a minimum of 4GB per stack, and a single stack (requiring only a single controller on the GPU die) will provide roughly the same memory bandwidth as the R9 280X's 384 bit GDDR5 interface. That's the point where it starts looking realistic to trickle it down to the lower cards: a single controller and HBM stack means less GPU silicon and less overall memory cost, and also it'll require a simpler interposer. I'd expect to see a full range of GPUs housing between 1 and 4 stacks of HBM2 depending on market segment.
A 2GB cheaper variant would be very interesting if they could slot it into that price bracket. I presume there will be some die harvest chips that don't make the full 4096 shaders grade.Quote:
Originally Posted by jigger
What would you call a scaled down Fury, a Tantrum? Perhaps "Strop" :D
I was just think that Tonga + a 384 bit memory bus respin might be as much work as adding HBM.