Fury X owners thread

SoCalMX70 said:
Still haven't seen concrete evidence of actual availability date though. Wouldn't be surprised if we don't see stock until February or March really...

Also, how about that Asetek lawsuit? I get that the responsibility and burden likely fall with Cooler Master, but is there a possibility that AMD re-thinks the cooler that they launch the X2 with? If so, we could see some serious delay. It's pretty late in the game at this point...
Click to expand...


If it's that late to the market, might as well not bother releasing it as both AMD with Greenland and Nvidia with Pascal, have those high end GPU's past their tape out phase, so both are undergoing revisions to work out the production kinks ( might take 2 ~ 3 revisions ), and then mass production is started.


With both using the 16nm process and possibly going north of 15 billion transistors in a single die, it's basically the same ballpark performance of the Fury X2 but using just a single GPU and less power and cheaper because it is just a single GPU card.



They may be released by the middle of next year, so if the Fury X2 comes out on the market in march, it'll hardly sell enough to offset the development costs of a 1000$+ Dual GPU card, which the market for those is pretty small to begin with......It's too rich for many users budgets.
 
shadow001 said:
If it's that late to the market, might as well not bother releasing it as both AMD with Greenland and Nvidia with Pascal, have those high end GPU's past their tape out phase, so both are undergoing revisions to work out the production kinks ( might take 2 ~ 3 revisions ), and then mass production is started.


With both using the 16nm process and possibly going north of 15 billion transistors in a single die, it's basically the same ballpark performance of the Fury X2 but using just a single GPU and less power and cheaper because it is just a single GPU card.



They may be released by the middle of next year, so if the Fury X2 comes out on the market in march, it'll hardly sell enough to offset the development costs of a 1000$+ Dual GPU card, which the market for those is pretty small to begin with......It's too rich for many users budgets.
Click to expand...

I guess the fury x2 will be pricey as hell.
within 9 months we as you noted might have single card reaching fury x2 performance and then its dead in water.
I assume the x2 is for the VR crowd.
 
i doubt it, im sure yields on 28nm are pretty good at this point.

I expect a price drop on single fury cards, and a decent price on the x2.
 
flopper said:
I guess the fury x2 will be pricey as hell.
within 9 months we as you noted might have single card reaching fury x2 performance and then its dead in water.
I assume the x2 is for the VR crowd.
Click to expand...


The largest performance gains have been once there's a change in the fab process node, and the change from 28nm right to 16nm + finfets ( not traditional transistors laid horizontally), pretty much allows to double the transistor budget for a given die size, as long as the overall power consumption also improves by the same amount (uses 1/2 the power of 28nm).


So we have nearly 9 billion with fury at 28nm and 600mm^, while still being PCI-e compliant power wise......Should the highest end Greenland also be a 600mm^ part, then it's not impossible that the part could have up to 18 billion transistors.



The thing to remember is that double precision floating point got nerfed in the fury cards as there simply wasn't enough transistor budget to allow larger gains in gaming AND large gains in HPC, so both AMD and Nvidia had to make a choice with this latest generation and they chose gaming....


Both Greenland and Pascal, potentially having an additional 8~9 billion transistors over Fiji and Maxwell, don't have to make that choice........Both get large increases on all fronts.
 
to a point.

neither wants to poach their workstation card sales..

AMD has always been significantly ahead of nvidia in DP anyway...

Which doesnt mean a whole lot to gamers...
 
The Mac said:
to a point.

neither wants to poach their workstation card sales..

AMD has always been significantly ahead of nvidia in DP anyway...

Which doesnt mean a whole lot to gamers...
Click to expand...


Yup as it's way more profitable on a per unit sold compared to what is essentially the same hardware packaged as a gaming card.


I don't get worked up as much on new releases as I used to, given the huge amount of GPU power that we already have right now, especially in multi GPU configurations, and new games that really push the limits take 4~5 years or more, with budgets that rival a major motion picture and involve 300~400 people in the project.



So imagine doing it the way I do, and going full 4 way Greenland's with each GPU packing 15+ billion transistors and 8GB of ram for each one (conservatively speaking, as HBM 2.0 can go to 32 GB capacities using 4 stacks), with 1 TB/sec of memory bandwidth, twice as fast as Fiji and exactly how long will it really take to create a graphics workload that would actually be challenging to something like that.



It's becoming pretty academic.....:bleh: :nuts:
 
xfire is a pretty tiny segment.

single card is still the biggest market.

obviously there will people that will do that, but as you say its pretty pointless other than for epeen...
 
The Mac said:
xfire is a pretty tiny segment.

single card is still the biggest market.

obviously there will people that will do that, but as you say its pretty pointless other than for epeen...
Click to expand...



I look at it long term, as go single GPU and you'll have to upgrade it that much sooner, or go multi GPU and one can sit on that same combo for a few years until software really has a chance to catch up, and one doesn't have to mess with the hardware during that entire time.



Ideally, i'd want to only change to something at least twice as fast as what I currently own ( R290x's with 8GB in each card), and Fiji isn't it (about 40% faster), so that leaves either Greenland or Pascal which looks like they might be released only in the second half of 2016.



Even then I still have serious doubts that my current cards, which will be 2 1/2 years old by then, will be struggling in performance terms, even at the pretty demanding settings I play at......
 
The Mac said:
if you have fiji crossfire, you are good to go for a quite a while at this point.
Click to expand...


Indeed.......Only thing that gives me pause is the amount of memory on the cards, and how it plays out in the long run as 4 Fiji's working together is a crazy amount of GPU power combined and will last for years, but the amount of memory on each is still 4GB and we're going to see 4K being common and maybe even 5k, as well as developers also using ever higher quality assets in their games which may use more memory too, so combine that with high resolutions and the use of AA and the amount of GPU power will handle it fine, but it's the memory amount that gives me doubts in the longer run (within the next 3+ years).



Given that 2016 will start in not much over 2 weeks from now, it would be a hell of a start if fury X2 comes into this world with 8GB of ram for each GPU (So 16GB per card), courtesy of HBM 2.0........I'd be all over a pair of cards like a cheap suit, as it also frees up a couple of PCI-e slots that I can use for other stuff, so it's not just because of graphics power that motivates this option in my case.



It's still strange that AMD's mid range performance cards like the R390 and R390X, are significantly cheaper than any version of Fiji and pack twice the memory onboard.
 
bill dennison said:
maybe to mid to the end of next year at 4k depending on how the dx12 games pan out
Click to expand...


The craziest thing would be that R390x's being significantly cheaper and packing twice the memory onboard, end up having more long term endurance in situations where the amount of memory used plays a significant role in overall performance.....



That could be the current Fury's only weakness as it's obviously quite a bit faster in raw GPU power, but that hardly matters if it doesn't have enough onboard ram now does it?......



I'm pretty sure that even before Greenland appears, and considering that the available production volume of 16nm is no where near what there is on 28nm, given how many years it's been available, that refresh parts of the existing Fiji 28nm GPU but packing 8GB of ram are going to happen.....



I don't see that many Greenland GPU's or Pascal GPU's, as long as TSMC's main customer is Apple and it soaks up nearly everything that TSMC can crank out.......Both AMD and Nvidia are small time players by comparison and get just left over capacity in the big scheme of things.
 
shadow001 said:
The craziest thing would be that R390x's being significantly cheaper and packing twice the memory onboard, end up having more long term endurance in situations where the amount of memory used plays a significant role in overall performance.....



That could be the current Fury's only weakness as it's obviously quite a bit faster in raw GPU power, but that hardly matters if it doesn't have enough onboard ram now does it?......



I'm pretty sure that even before Greenland appears, and considering that the available production volume of 16nm is no where near what there is on 28nm, given how many years it's been available, that refresh parts of the existing Fiji 28nm GPU but packing 8GB of ram are going to happen.....



I don't see that many Greenland GPU's or Pascal GPU's, as long as TSMC's main customer is Apple and it soaks up nearly everything that TSMC can crank out.......Both AMD and Nvidia are small time players by comparison and get just left over capacity in the big scheme of things.
Click to expand...

I suppose it's horses for courses really. I game at 1440p and my Fury Pro gives me 60+ FPS in most games and I don't think the 4gb ram will ever be a bottleneck provided I don't go mad with AA. Generally I only use 2 x AA and tbh I don't really notice jaggies.

I've always presumed that Arctic Islands would be out next but now you've mentioned it, it could make sense to release another Fiji on 28nm but with 8gb HBM2. It would certainly help recoup the development costs given the short time span since launch :sherlock:
 
I never go below 60FPS @ 1920x1080 + 8xQSS, although these days I play Deus Ex: Revision (Yeah, got a Fury X to play retro! :P)
 
LordHawkwind said:
I suppose it's horses for courses really. I game at 1440p and my Fury Pro gives me 60+ FPS in most games and I don't think the 4gb ram will ever be a bottleneck provided I don't go mad with AA. Generally I only use 2 x AA and tbh I don't really notice jaggies.

I've always presumed that Arctic Islands would be out next but now you've mentioned it, it could make sense to release another Fiji on 28nm but with 8gb HBM2. It would certainly help recoup the development costs given the short time span since launch :sherlock:
Click to expand...

theres no space to add 4 more HBM stacks to make 8GB unless they make the interposer bigger.

You think fiji is expensive now? a larger interposer, 4 more stacks, redesigned power phases, and a new cooling solution would probably push it over $800
 
The Mac said:
theres no space to add 4 more HBM stacks to make 8GB unless they make the interposer bigger.

You think fiji is expensive now? a larger interposer, 4 more stacks, redesigned power phases, and a new cooling solution would probably push it over $800
Click to expand...



Unless i'm mistaken, one of the big changes with HBM 2.0 is the capability to handle up to 32 GB capacities using 4 stacks of 8 chips high.......32, 1 GB ram chips in total......Obviously this requires a fab process shrink for the memory, as I assume that the current HBM 1.0 process for the memory is a pretty old one since each separate ram module is basically 256MB in capacity.....16 of these and you get 4GB total.


Right now, HBM 1.0 has a maximum of 4 stacks of 4 chips in each stack, to get a maximum capacity of 4 GB, so with this exact same configuration and interposer production costs, HBM 2.0 would allow configurations with 16GB onboard.......All I'm asking are cards with 8 GB.


HBM's 2.0 maximum of 32GB would be using 4 stacks surrounding the GPU die and each of those would have 8 chips, one on top of the other but that's serious overkill for a gaming card, even by my crazy standards.....:lol:



So an 8GB HBM 2.0 equipped high end Fury costs no more to produce, as each stack doesn't need to use more memory chips, and also doesn't need any additional stacks either, so there's no change to the interposer involved to begin with.......Since it's so simple ( one of the arguments of going from HBM 1.0 to 2.0), what the heck is taking so long?



I'm not asking for 8GB of ram still using HBM 1.0, and 2.0 solves it with a lot less external technical changes that involve changes to cooling or the interposer itself.....It's got noting to do with either, and the main change is a smaller fabrication process for the memory itself, to allow higher densities that's all.
 
I suppose they could release one with HBM2

They would need to change the interposer however, as the bus is wider.
 
I don't think they will unless 14/16nm is delayed again to like 2017
only then do I think you may see a 8gb or 16gb fury x on 28nm with hbm2


not that I would buy one just for more vram
 
The Mac said:
I suppose they could release one with HBM2

They would need to change the interposer however, as the bus is wider.
Click to expand...


The Bus is the same, as what gives HBM 2.0 it's extra Memory bandwidth is because the memory runs at 1 Ghz actual clock, while HBM 1.0 is at 500 Mhz actual clock......Both use quad signaling like GDDR5 memory, so 4 bits per clock travel across the bus.



I don't think they will unless 14/16nm is delayed again to like 2017
only then do I think you may see a 8gb or 16gb fury x on 28nm with hbm2


not that I would buy one just for more vram
Click to expand...


Fair enough as it's up to each one to decide, but given that the ramp up on 14/16nm will take a couple of years to reach the production volumes that 28nm can reach right now, precisely because it's been available for the last 4 years now, so I don't expect huge volumes on the new process as everyone and their uncle wants to use the initial production volume capacity for their next generation products, whatever they may be and it's not just AMD and Nvidia wanting to use it, but bigger players too.



This sets up the stage for said products being relatively hard to get and having price markups right from the start, and Fiji has plenty of grunt left to give as a viable gaming card for a long time to come, so it's only the ram issue that gives me pause, and leveraging HBM 2.0 in an updated version perhaps with higher clocks in a similar fashion as to what happened with the R290x becoming the R390x and packing twice the memory onboard, and I wouldn't call it a bad product in the least.


Heck price those around the same ~500$ range and it's awesome bang for the buck value, even if Greenland is twice as fast but hits your wallet for 800$+......Main thing is that revenue from the development cost of HBM 2.0 comes in from more than the very highest end cards that only 5% of consumers buy such expensive cards to begin with.
 
Guess that's that for the fury X2 card.......It won't likely be announced this year at this point, as I've never seen a launch (even a paper launch), happen this close to Xmas....:/
 
You must log in or register to reply here.
Back
Top