AMD Polaris architecture – GCN 4.0

[Stanners]

it was same event he released the titan-x, pretty sure he did! i remember him looking sxy in his leather jacket

....maybe i daydreamed it

Would this not be more at home in the Pascal thread

 

[Mei]

im scared of being bullied

 

[shankly1985]

I would like to think AMD are able to keep the TDP down to a level were they don't need a pump on anything else other then a dual chip cards.

The 14nm cards really ought to bring massive leaps in performance, Nvidia were able to crack out a ton of extra FPS with the move the Maxwell whilst keeping on the same process so AMD with effectively 2 and bit worth of node drops plus a new architecture (which they have had years to work on) really should bring some amazing numbers.

Since adding a GH10 on my 290 I'll be using a water cooling solution either with GPU or adding myself.

Cooler and more quite is enough for me.

 

[eddyr]

The point being, that two 596mm^ chips on 28nm this late in the day, are going to seem pretty archaic compared to all the new, tiny, super efficient, 14/16nm chips that are supposedly also going to be shown.

Archaic, but seems they will be the peak performing cards for a while regardless of early 14/16nm gpus and their architectural improvements. If not for VR then I think it would be mad to purchase with the intent to run in xfire (just don't like dual gpu as it is atm) with scaling, dx 11, support issues and where eeked out performance might be desirable.

 

[Stanners]

im scared of being bullied

We shall protect you

 

[JediFragger]

didnt jhensun hold up pascal at last years event?
wow these things take a long time to come out

It was a bag of woodscrews in disguise

 

[Mtom]

didnt jhensun hold up pascal at last years event?
wow these things take a long time to come out

It was a mockup

 

[drunkenmaster]

The small Pascal compute card thing they showed the previous year was dodgy as hell. The focus wasn't on how fake it was as much as they attempted to make it look like they were the first to do 3d memory.

AMD worked with a memory company to make HBM, they were one of the companies pushing for interposers and working with the entire supply chain to get HBM/interposers going and it took 6-7 years to get it done... and before your first card comes out your biggest rival comes out, having contributed absolutely nothing to the development and starts showing off a fake card and talking up 3d memory so people associate it with them.

AMD had interposer + chip + memory samples for APU as far back as 2011, this is the time frame it takes to get these things up and running, then Nvidia jump out and talk about as if it's a breakthrough from them.

 

[humbug]

The small Pascal compute card thing they showed the previous year was dodgy as hell. The focus wasn't on how fake it was as much as they attempted to make it look like they were the first to do 3d memory.

AMD worked with a memory company to make HBM, they were one of the companies pushing for interposers and working with the entire supply chain to get HBM/interposers going and it took 6-7 years to get it done... and before your first card comes out your biggest rival comes out, having contributed absolutely nothing to the development and starts showing off a fake card and talking up 3d memory so people associate it with them.

AMD had interposer + chip + memory samples for APU as far back as 2011, this is the time frame it takes to get these things up and running, then Nvidia jump out and talk about as if it's a breakthrough from them.

Well, this is where AMD's marketing falls flat on its back.

Nvidia just run around insinuating they invented everything.

AMD say things like "First with HBM" what they really should be saying is "We designed and built the interposer to make HBM 3D Satcking a reality"
And if Nvidia are using those designs i think AMD should say so "Nvidia are using our interposer designs for their Pascal GPU's"
What are Nvidia going to do about that? i'd have them stick an AMD logo on it....

 

[D.P.]

Archaic, but seems they will be the peak performing cards for a while regardless of early 14/16nm gpus and their architectural improvements. If not for VR then I think it would be mad to purchase with the intent to run in xfire (just don't like dual gpu as it is atm) with scaling, dx 11, support issues and where eeked out performance might be desirable.

The question is do you want:
*) Fury X2, nano spec cores with 4GB per GPU, likely a high price for niche product. That 4Gb does become an issue at 4K xfire.
*) 2X 390X with 8GB per GPU, cheaper than the FuryX2 and no VRAM issue
*) 2X 980Ti with 6GB per GPU comparable price to FuryX2, faster, more VRAM for 4K SLI.


Add to the fact that big Polaris and Pascal wont be long why would you buy a FuryX2? I'm pretty sure the big next gen chips will get very close to a FuryX2 performance for less money, more vram (perhaps much more like 16GB vs 4GB) and less heat/power.

 

[Mtom]

They will bring the new set of drivers with Vulkan API support, etc. to GTC

Lets hope we will get some more Polaris info as well.

 

[Mtom]

The question is do you want:
*) Fury X2, nano spec cores with 4GB per GPU, likely a high price for niche product. That 4Gb does become an issue at 4K xfire.
*) 2X 390X with 8GB per GPU, cheaper than the FuryX2 and no VRAM issue
*) 2X 980Ti with 6GB per GPU comparable price to FuryX2, faster, more VRAM for 4K SLI.


Add to the fact that big Polaris and Pascal wont be long why would you buy a FuryX2? I'm pretty sure the big next gen chips will get very close to a FuryX2 performance for less money, more vram (perhaps much more like 16GB vs 4GB) and less heat/power.

I just thought about that...why they pushing back the X2 from late autumn to spring...if i were AMD i'd stick HBM2 on it (2x8GB), yes it is a again take some money, but considering how far the next X2 cards are (i assume no sooner than late 2017 if even) they can make the Fury X2 live longer on the market with 2x8GB. Well worth the cost of switching to HBM2.

 

[D.P.]

I just thought about that...why they pushing back the X2 from late autumn to spring...if i were AMD i'd stick HBM2 on it (2x8GB), yes it is a again take some money, but considering how far the next X2 cards are (i assume no sooner than late 2017 if even) they can make the Fury X2 live longer on the market with 2x8GB. Well worth the cost of switching to HBM2.

It is not a simple replacement, they would need to design a new memory controller and spin out a new chip. Hardly worth it unless big Polaris is really late.

There is nos ingle good reason why the X2 is so late but the FuryX launch was late, stock arrived ate and then in a feeble dribble, the Nano was late AFAIK but the X2 is really late. It might still be HBM and/or interposer problems/supply. AMD like prefers to sell 2 single Fury cards than a single FuryX with 2 GPUs.

 

[JediFragger]

The small Pascal compute card thing they showed the previous year was dodgy as hell. The focus wasn't on how fake it was as much as they attempted to make it look like they were the first to do 3d memory.

AMD worked with a memory company to make HBM, they were one of the companies pushing for interposers and working with the entire supply chain to get HBM/interposers going and it took 6-7 years to get it done... and before your first card comes out your biggest rival comes out, having contributed absolutely nothing to the development and starts showing off a fake card and talking up 3d memory so people associate it with them.

AMD had interposer + chip + memory samples for APU as far back as 2011, this is the time frame it takes to get these things up and running, then Nvidia jump out and talk about as if it's a breakthrough from them.

Can't blame nVidia for what they do tbh, let AMD do the heavy lifting then take advantage! Proper business thinking

 

[Dave2150]

Can't blame nVidia for what they do tbh, let AMD do the heavy lifting then take advantage! Proper business thinking

I guess we can all look forward to the day that AMD is no more, and NVIDIA have no-one to do the innovation/heavy lifting for them......

981ti 8GB, 982ti 12GB, 983ti 16GB, each released 6 months apart, and they'll still sell by the thousands, thanks to loyal fans

 

[JediFragger]

I guess we can all look forward to the day that AMD is no more, and NVIDIA have no-one to do the innovation/heavy lifting for them......

981ti 8GB, 982ti 12GB, 983ti 16GB, each released 6 months apart, and they'll still sell by the thousands, thanks to loyal fans

Don't forget that driver support will end for any older gpu's the very *day* that a new one is released

We're almost there with that one though

 

[Overlag]

981ti 8GB, 982ti 12GB, 983ti 16GB, each released 6 months apart, and they'll still sell by the thousands, thanks to loyal fans

 

[Mtom]

Don't forget that driver support will end for any older gpu's the very *day* that a new one is released

We're almost there with that one though

I have a feeling NV will flop somehow this pascal release/lineup. AMD showed the chips in december closed doors, and demoed them in january while nv had nothing so far.
Its just a feeling could be false.

AMD could use some time to build momentum and get back some market share.

 

[flopper]

I have a feeling NV will flop somehow this pascal release/lineup. AMD showed the chips in december closed doors, and demoed them in january while nv had nothing so far.
Its just a feeling could be false.

AMD could use some time to build momentum and get back some market share.

Nvidia already flopped with Pascal.
No card, no demo and no hope for Pascal.
When you see them have cards check that there is a card in the hand that holds it instead of woodscrews so your not screwed over.

Polaris is brighter than ever shining light upon those with faith.

 

[Simon Belmont]

Nvidia already flopped with Pascal.
No card, no demo and no hope for Pascal.
When you see them have cards check that there is a card in the hand that holds it instead of woodscrews so your not screwed over.

Polaris is brighter than ever shining light upon those with faith.

Biggest fanboy comment I've read this year.