New Polaris 30 Respin at 12nm to be released? (RadeonRX 590 or 680?)
Click here to post a comment for New Polaris 30 Respin at 12nm to be released? (RadeonRX 590 or 680?) on our message forum
Jumbik
vbetts
Moderator
JonasBeckman
Vega lost a few features due to well I guess we don't really know the full circumstances (Primitive shader stages seem to be coming in with Navi again though for recent Vulkan code commits for AMD's Linux patch submissions ranging from Vega 12, Vega 20 and Navi support and other stuff.) but overall up to 30% faster than Fury and a bit more via tuning although it seems to stabilize more from having a higher power limit or possibly undervolting due to AMD squeezing out those last "up to" percentage gains (3 - 5% or so.) via the default settings. (And that overdrive clock profile should just be tossed out really.)
I can't really see it as a failure but it couldn't quite close the gap to NVIDIA's lead and the pricing at it's worst made the 1080Ti *cheaper* while around having 15 - 20% or above performance lead. (Up to 30% in a couple of games actually.)
Remains to be seen what Polaris 600 could be or what the GPU will be called, if it closes in or matches Vega 56 to 64 that would be interesting too although without going for a even higher clock speed or GDDR6 I wonder how much room there is without majorly overhauling the GPU core and features which then wouldn't really be Polaris.
Would slot in nicely with the Vega cards still in low availability if it can get close though even if it might not outmatch the 1070/1070ti or 1080 from NVIDIA's side assuming pricing isn't a issue due to the newness of the 12nm fabrication and whatever the initial availability there will be. (Probably not the most complex chip compared to the upcoming Navi GPU but still going to be a new process though I guess it also tests this fabrication node out a bit before full production with the new GPU arch.)
Don't really have too much knowledge about this sort of thing myself though but I'm curious how it'll end up though without more info it's mostly just speculation for now and what might be a good position for AMD. 🙂
Price against performance and whether it'll match Vega or not, I'm thinking it might get close but we'll see since it's different from the 300 series and Fiji this time with Vega now having similar or additional architecture improvements.
(Even if it sticks close to the standard GCN setup for cores, ROPS and such.)
EDIT: I do find it ridiculous how Vega can shave of almost 200mv of voltage or more depending on GPU quality / binning too. My own settings for Vega 56 use 850Mhz for the (Hynix) memory and 1450 to 1500 Mhz for the GPU core at 1.0v instead of 1.2v for the GPU core instead of 1500 - 1590 but that reduction barely registers though the power draw and heat reduction certainly helps. Also helps with the HBM and it's temp sensitivity. 😀
Though I would expect AMD to be a bit less aggressive with Polaris stock settings although anything goes for Navi if AMD is trying to bridge the gap to the 1080Ti or above and positioning their higher-end models of that around NVIDIA's 2000 series lineup but that's something we'll see next year, maybe. Ha ha.
(And the die shrink should reduce power usage further plus maybe improvements to HBM2 could help with availability and it's power and clock speed efficiency.)
tunejunky
i agree with the analysis by AlmondMan.
spot on.
imho, refreshed Polaris is going to be a real thing...at the low end.
7nm is not "far away" it is here.
the AMD lineup will look like this - refreshed Polaris (at higher profitability as well due to die shrink and proven architecture/drivers) at the entry level with the elimination of all models not fully 1080p/60fps and full Polaris being 1440p with a price drop.
Navi will destroy the mid-market with 4k performance at the lowest price we've seen ($350-450) and then a wait for the refreshed and tweaked Vega 7nm.
i get this from AMD's prospectus, road maps, and earnings calls.
and the 2080/ti will still be comfortable...barely.
AMD is far more profitable (at the fab) with higher yields than either Intel or Nvidia. and to the point of Nvidia, be very afraid when the fabs can accommodate more contract production...they most definitely have something waiting...they're just behind the 8 ball atm.
Fox2232
Silva
At the right price, I'd get one right now.
I couldn't get an RX570/580 because of miners and I'm currently with a tiny RX560 (that performs the same as my old R9 270X).
I can't afford cards over 250€ so this refresh is what I've been waiting since the beginning of the year.
RzrTrek
Fox2232
Venix
@PrMinisterGR & @vbetts
Well pimpineasy so far all his latest posts follow the same stracture , presenting arguments in a confusing way! I mean literally i was really trying to make sense out of it .
Also he follows a golden rule ...if the word amd is on his post then trash or garbage is always there ! And vice versa!
vbetts
Moderator
So this topic isn't to argue on power usage and what not, nor to down talk anyone or anything here so...Let's hop back on topic boys and girls!
SniperX
vbetts
Moderator
JamesSneed
I don't think this news will pan out.
icedman
I always wondered what polaris would have been capable of if it had been able to clock as high as Nvidias cards, It's not good that AMD has to sell larger die GPU's to compete in terms of profitability for them.
-Tj-
Oh common AMD you can do better then that,..
I wonder why they kept 7nm exclusive to professional ecosystem.. I would like to see a 96 ROP and 6144 core vega96 @ 7nm , that would do some serious shizzle 😀
Vega64: 4096:256:64
"Vega96": 6144:384:96,
that looks doable from 16nm to 7nm no? or might that be Navi?
I'm pretty sure this imaginary Vega96 would rival 2080rtx and 2080ti np 😀
Mitch 74
JonasBeckman
Memory accounts for a part but there's something else. Hopefully whatever upcoming design AMD has planned will resolve some of these issues whatever the core cause is, although it might also be part of the driver software limiting things a bit for DX11 compared to VLK and DX12 where the GPU's start really working to their utmost. 🙂
(OC'ing memory does give results though but from the tests I've been reading up on at a 30% overclock from 800Mhz up to 1040 Mhz and a slight boost to 1700 Mhz for core clocks up from 1598 Mhz then there should have been more than a 8% total performance increase although memory alone accounts for about 5% so it's certainly a boost although not that big of one compared to the clock speed increase.)
Hmm and I see now, AMD is using up the 12nm they have for this and then it's all focused on 7nm after that I suppose. I was thinking about 7nm entirely but that's not the case at all. 🙂
And I guess Polaris gets GDDR5 or maybe GDDR5X unless AMD spends the time and money on a GDDR6 design but I suppose that's a bit less likely although it would probably bump up performance a bit more and might give a nice total bandwidth even it falls short in other areas compared to HBM, though it should be less costly. 🙂
(Hopefully that also goes for the retail product ha ha.)
EDIT: Hmm wonder if a 580 with a die shrink and some tweaks would be what to expect realistically, guess it depends on what 12nm can do and what AMD will do with Polaris30 here if anything beyond shrinking it down to a smaller node.
Just have to wait and see I suppose.
Andrew LB
AMD is doing this because they have nothing new for the holiday season while nVIdia will have brand new, top to bottom cards out by then. The best part is seeing the reaction from those who spent so much time criticizing nVidia for the RTX which at least is a new product, while saying this is a good move by AMD. I guess they had to so something considering Navi wont be around till 2H, 2019
Killian38
Seikon
They really should not bring out a other refresh with a 5 to 10% increase in performance , what they need to do is bring out a GFX card that is as fast as a gtx1080 but under 300$(and that should be the price in stores) while not eating more power then a vega56 , that would sell like hoot cake.