Differences between TSMC 16nm FF+ (Pascal) and GloFo 14nm LPP (Polaris)?

[DA CPU WIZARD]

Essentially, are there any major differences between the two processes that will result in any notable advantages / disadvantages for either Nvidia or AMD once their next generation cards arrive this coming Summer?

 

[Silverforce11]

There's differences in SOCs built on the two nodes, but whether we can extrapolate or draw much from that to GPUs, remains unproven. Any claims would be pure speculation.

 

[flash-gordon]

At this point there's just one difference, the first is nowhere to be found and the second is already working.

 

[Mondozei]

Samsung recently announced their 2nd gen 14 nm node for their E8890. So they do seem ahead of TSMC generally speaking. This may also explain why Pascal faces delays and AMD does not.

After all, GloFo is tightly co-operating with Samsung these days.

Will it mean higher performance? Who knows, but peak theoretical performance matters just as much, if not less, as intelligent implementation.

 

[DA CPU WIZARD]

Are they both the same size, with each just using different measurement methods and branding, or is the GloFo process truly 2nm smaller than the TSMC one?

 

[Vesku]

Samsung 14FF is physically smaller, however TSMC choice in size may lead to better performance characteristics at higher clocks.

 

[el etro]

TBH we will really know the differences about them after we see a Pascal vs. Polaris showdown with real products. Until then we don't have the right idea about performance/size characteristics of the two process.

 

[MrTeal]

TBH we will really know the differences about them after we see a Pascal vs. Polaris showdown with real products. Until then we don't have the right idea about performance/size characteristics of the two process.

How would seeing a Pascal vs Polaris showdown tell us anything useful about the two processes? Maybe, if AMD does do big Polaris on 16FF+ it would at least be a little useful, but even then other differences could likely swamp the process differences. Trying to compare the performance of two wildly different GPUs in order to glean information about how good the process itself is will tell us nothing.

Unless someone dual sources the same GPU die and tunes it at both fabs, no one here who's not under NDA will be able to really compare the processes. Even looking at the A9 isn't perfect, since a CPU running at 2.2GHz is not the same as a GPU running at a lower clock speed.

 

[flopper]

Polaris is ready today, Pascal isnt.

 

[Azix]

14nm is going to be better I think. The only problem would be if AMD or Samsung/Glofo fail and do not make the most of it. given the resources samsung and glofo have I expect it will mature nicely.

Nvidia can just make bigger chips though. Even if you can fit more into the same space, the competition can just fit more into more space.

 

[NTMBK]

 

[boozzer]

nice picture, had to quote.

 

[Good_fella]

Polaris is ready today, Pascal isnt.

It so ready for today that Polaris will be released July/September.

 

[ShintaiDK]

It so ready for today that Polaris will be released July/September.

They are just waiting on drivers

/sarcasm

 

[Glo.]

http://news.samsung.com/global/sams...-14-nanometer-finfet-logic-process-technology

Samsung’s new 14nm LPP process delivers up to 15 percent higher speed and 15 percent less power consumption over the previous 14nm LPE process through improvements in transistor structure and process optimization. In addition, use of fully-depleted FinFET transistors brings enhanced manufacturing capabilities to overcome scaling limitations.

Fully Depleted Silicon on Insulator. IBM technology. It looks like not only Samsung licensed the tech to GloFo, but GloFo could license it to Samsung also.

And that would say why the pre-production, working mule of silicon was able to work on 0.8V at 850MHz in AMD presentation.

So in all it looks like it is denser, has better power. I don't know why, but I do feel like AMD could have really good silicon this time.

 

[NTMBK]

http://news.samsung.com/global/sams...-14-nanometer-finfet-logic-process-technology
Fully Depleted Silicon on Insulator. IBM technology. It looks like not only Samsung licensed the tech to GloFo, but GloFo could license it to Samsung also.

And that would say why the pre-production, working mule of silicon was able to work on 0.8V at 850MHz in AMD presentation.

So in all it looks like it is denser, has better power. I don't know why, but I do feel like AMD could have really good silicon this time.

Fully depleted!=SOI.

 

[tviceman]

Polaris is ready today, Pascal isnt.

Where can we buy it?

 

[Azix]

This could end up a really big deal in defining this generation. Unless nvidia jumps to samsung later on which is maybe not likely. There is now a practical difference in manufacturing between the two companies. They are not using the same TSMC anymore. It could turn out to be a major advantage for either.

Its exciting because 7970 to Fury x is a huge improvement on the same process. A new process brings massive improvements in itself, matured and improved architectures on 14nm would be ridiculous.

4K 144hz freesync full range :biggrin:

like i could afford it.

 

[ShintaiDK]

This could end up a really big deal in defining this generation. Unless nvidia jumps to samsung later on which is maybe not likely. There is now a practical difference in manufacturing between the two companies. They are not using the same TSMC anymore. It could turn out to be a major advantage for either.

Didn't AMD say they would use TSMC as well. That pretty much says it all.

 

[Azix]

Didn't AMD say they would use TSMC as well. That pretty much says it all.

if they are refreshing or still releasing older products, yeah they would say that (TSMC 28nm). But I'd like to see a source so I can better understand. There were a lot of rumours. All we have now for sure is 14nm finfet straight from AMD's mouths.

edit: its unlikely they will use it for new GPUs and x86 CPUs when they've invested in 14nm. Maybe ARM CPUs

 

[Udgnim]

Samsung stole TMSC's head process engineer

Samsung has got a lot more money to R&D with than TMSC

my bet is that Samsung's 14nm will be better than TMSC's 16nm

Nvidia might still end up designing & engineering a better card than AMD like Maxwell vs Fury

 

[Azix]

Samsung stole TMSC's head process engineer

Samsung has got a lot more money to R&D with than TMSC

my bet is that Samsung's 14nm will be better than TMSC's 16nm

Nvidia might still end up designing & engineering a better card than AMD like Maxwell vs Fury

I always find this a shaky thing to say. because they went in somewhat different directions. Nvidia's advantage there is probably being able to redesign heavily enough to cut out power consuming features which lead to more overclockability.

Whether its actually better design and engineering... I don't think that's so easy to say. Maybe later on we can tell, eg kepler vs GCN 1/2.

 

[guskline]

WOW! Sounds like an exciting 6 to 9 months coming up for high end GPUs.

 

[ShintaiDK]

WOW! Sounds like an exciting 6 to 9 months coming up for high end GPUs.

No, it sounds like a long 6-9 months to wait before we get them

 

[3DVagabond]

Where can we buy it?

Moving goalposts 101. There are lots of things that are ready today that you can't buy.