• 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 Polaris architecture – GCN 4.0

fs123 said:
AMD seems to be showing the next gpu's while nothing seems to be coming out from Nvidia when Pascal is supposed to be here first. Am I missing something?
Click to expand...

Nvidia was never supposed to be here first, people saw a supposed tape out and used how long the tape out process was on different processes to determine that Pascal was only 6 months away. They were completely wrong, not least because tape out to market time increases with process complexity. What took 6 months on 40/28nm could take over a year on 14nm. Production troubles, TSMC falling further behind. For the entire past 2 years everyone in the industry thought Samsung was ahead of TSMC for finfet.

Ignoring this some people on forums decided Pascal might come as early as Q1 this year but no later than end of Q2.
 
FoxEye said:
Hmm, a "console class GPU" is something a bit better than a 7870, iirc. Not as good as a 7950 by a long way.

That is one of the announced GPUs, intended mainly for notebooks, and very low-end desktop (equal to the 270/370, roughly...)

That means one of two things, then. They have one new GPU to power everything from the new Fury class GPU, then the new 490 class, then the new 480 class.

Or they are going to have to re-use some of the current range. Like the Fury and the 390.

Otherwise it would be like nV having to use the same chip (GM100) for Titan, the 980ti, the 980 AND the 970. Could they actually do this? Seems unlikely.

So... we should expect some re-brands in the next-gen lineup?
Click to expand...

It depends on the pricing. Look at the Nvidia GTX900 series - everything from £100 to £450 is served by two GPUs.

The GM204 is in products from £250 to £450 and the GM206 is in products between £100 to £200.

It could be that Big Polaris will have the full part at around £350 to £450 and the smaller part at £250 to £350,and the small Polaris part is under £200.

OFC,all speculation on my part.
 
BTW,regarding the process node used. The GF/Samsung 14NM LPE process is meant to be around 10% denser than the TSMC 16NM process but with worse power consumption,but it appears the GF/Samsung 14NM LPP process is entering production too which improves on power consumption. It might mean AMD has a density advantage this generation.

Having said,that I hope the RTG can actually become more focused like ATI/AMD of old and not botch up the launch of the Polaris 10 and Polaris 11.
 
CAT-THE-FIFTH said:
The GM204 is in products from £250 to £450 and the GM206 is in products between £100 to £200.

It could be that Big Polaris will have the full part at around £350 to £450 and the smaller part at £250 to £350,and the small Polaris part is under £200.
Click to expand...

Reading that makes me sad :(

Remember the time when nV's 2nd top GPU was £250?

Now the 3rd and 4th top are £400 and £250 respectively :/
 
drunkenmaster said:
Then an enthusiast card, probably 300-400mm^2 somewhere. It will be aiming for probably 15-20% beyond Fury performance, depending on architecture it may be more or less. It depends heavily on how dense they can make the gpu on the new process compared to yields.
Click to expand...

I'd be very surprised if a 300-400mm^2 card on this new process would only be 20% faster.

This new process is supposed to be roughly a 1.5 generation jump from 28nm, and TSMC are supposedly claiming 3x performance/watt for their process.

So surely a 7970/680 level card should be pushing twice the speed of a Fury.
 
AllBodies said:
I'd be very surprised if a 300-400mm^2 card on this new process would only be 20% faster.

This new process is supposed to be roughly a 1.5 generation jump from 28nm, and TSMC are supposedly claiming 3x performance/watt for their process.

So surely a 7970/680 level card should be pushing twice the speed of a Fury.
Click to expand...


Going from a high end 600mm^2 core like Fury to a midrange 300-400mm^2 core isn't going to offer double the performance, it won't even come close to double, especially if it was closer to the 300 than 400mm^2.

Fury is huge because it could come so late on the process that yields were still decent at that size.

Fury and Titan X were huge huge chips primarily because 28nm was around so long. I'd be near enough shocked if we got anything over 400mm^2 within the first 6 months of 14nm gpus being out, let alone 600mm^2.

TSMC can say what they want, 28nm > 14/16nm is about one node jump. It uses a 20nm metal layer. The 28nm > 20nm pretty much gained nothing compared to a normal node, half the usual node jump at best and 14/16nm is just smaller transistor ontop of the same bigger base layers of the chip, it is also barely a half node difference.

In reality 20nm brought with it about 1.8transistor density, the 14nm brings with it about 50-60% power saving and another 0.1 gain in transistor density over 20nm. Both together are close enough to what a normal single node drop has been. There is no reasonable expectation that something half the size will perform twice as faster. A 600mm^2 14nm, sure, twice as fast as Fury hopefully, 300mm^2, not a chance, hopefully with architecture gains 20%, if the architecture is a huge improvement then maybe 30%.
 
So high end Polaris won't really be a worthwhile change from a 980Ti @ 1080p or 1440p then? Seems as if it will just catch or slightly surpass a factory OCed 980Ti.

Oh wait you weren't talking about the next Fury card, not expected any time soon?
 
2017 for da big bad boys, 2016 for low and mid range is how i see it.

Don't be fooled into getting a mid range like nVidia did with people last time 970/980 (Sorry mid to high end)
Just wait for the big cards.
 
drunkenmaster said:
Going from a high end 600mm^2 core like Fury to a midrange 300-400mm^2 core isn't going to offer double the performance, it won't even come close to double, especially if it was closer to the 300 than 400mm^2.

Fury is huge because it could come so late on the process that yields were still decent at that size.

Fury and Titan X were huge huge chips primarily because 28nm was around so long. I'd be near enough shocked if we got anything over 400mm^2 within the first 6 months of 14nm gpus being out, let alone 600mm^2.

TSMC can say what they want, 28nm > 14/16nm is about one node jump. It uses a 20nm metal layer. The 28nm > 20nm pretty much gained nothing compared to a normal node, half the usual node jump at best and 14/16nm is just smaller transistor ontop of the same bigger base layers of the chip, it is also barely a half node difference.

In reality 20nm brought with it about 1.8transistor density, the 14nm brings with it about 50-60% power saving and another 0.1 gain in transistor density over 20nm. Both together are close enough to what a normal single node drop has been. There is no reasonable expectation that something half the size will perform twice as faster. A 600mm^2 14nm, sure, twice as fast as Fury hopefully, 300mm^2, not a chance, hopefully with architecture gains 20%, if the architecture is a huge improvement then maybe 30%.
Click to expand...

This I agree with^^^

There is something else that could hinder Small Pascal when compared to Big Maxwell. Part of the reason Maxwell got a performance boost is all the compute functions were removed. What is going to happen when the compute functions goes back in with Small Pascal and how much is it going to hinder gaming performance.

People looking for an upgrade from a GTX 980 Ti could be disappointed with Small Pascal as I am guessing it will be barely faster.
 
r7slayer said:
2017 for da big bad boys, 2016 for low and mid range is how i see it.

Don't be fooled into getting a mid range like nVidia did with people last time 970/980 (Sorry mid to high end)
Just wait for the big cards.
Click to expand...
with that attitude you could be waiting forever ;)
 
Kaapstad said:
This I agree with^^^

There is something else that could hinder Small Pascal when compared to Big Maxwell. Part of the reason Maxwell got a performance boost is all the compute functions were removed. What is going to happen when the compute functions goes back in with Small Pascal and how much is it going to hinder gaming performance.

People looking for an upgrade from a GTX 980 Ti could be disappointed with Small Pascal as I am guessing it will be barely faster.
Click to expand...

They were touting something called mixed precision, facts a little light on the ground though.

https://techreport.com/news/27978/nvidia-pascal-to-feature-mixed-precision-mode-up-to-32gb-of-ram

This also showed up in the search results, looks interesting:

https://www.eecs.berkeley.edu/Pubs/TechRpts/2015/EECS-2015-265.pdf
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom