[Reuters]Nvidia to use Samsung's 14nm process?

[jpiniero]

Sounds interesting. From what I understand with Apple chips TSMC's 16nm versions used less power than Samsung's 14nm chips so it might be due to lower costs.

Samsung's fab is denser but the design costs to port it (I would think) would more than offset that.

I have to think (if this is true) that it is because they won't be able to get enough wafers to support a Pascal Refresh, presumably because the 2017 iPhone is rumored to be a sales monster, especially since the 2016 iPhone is expected to not sell that well and so there will be extra pent up demand.

The added density would still be nice, maybe they will throw in some extra cores instead of just a respin with faster memory. I wouldn't expect the die to get much bigger (relatively speaking) unless they really want to jack prices up even more.

 

[dark zero]

Now that Intel and ARM ended into a deal, is easier to nVIDIA to enter to the 10 nm ARM chips.

 

[AtenRa]

A market replacement is not the same as a successor. Pitcairn wasn't the successor to Cayman, and Polaris isn't the successor to Hawaii.

You want to talk about successors ??

https://www.techpowerup.com/reviews/AMD/HD_7850_HD_7870/26.html

HD6870 @ 40nm
Die size 255mm2, 1700 M Transistors , 151W TDP

HD7870 @ 28nm
Die size 212mm2, 2800M Transistors , 175W TDP

HD7870 vs HD 6870

Transistors = 65% more
Die size = 20% smaller
TDP = 16% more
Power Consumption = 103W-119W = 13,5% less
Performance = 39% faster

---------------------------

https://www.techpowerup.com/reviews/AMD/RX_480/24.html

HD7870 (R9 270X) @ 28nm
Die size 212mm2, 2800M Transistors , 175W TDP

RX 480 @ 14nm
Die size 232mm2, 5700M Transistors , 150W TDP

RX 480 vs HD7870

Transistors = 103% more (more than double)
Die size = 9,5% bigger
TDP = 14% lower
Power Consumption = 163W-119W = 37% more (Im sure they dont measure on the same game version or same drivers)
Performance = 85% faster

Its a huge performance increase, the biggest we have ever seen for a successor the last 3-4 years.

You're comparing a 512-bit 440mm2 former flagship chip to a 256-bit 230mm2 mid-range chip. PLEAAAASE. RX 480 didn't even match Hawaii in performance.

Did you bother to take a look at the technical specs of Polaris 10 vs Hawaii ??? Polaris 10 has 8% less transistors and its only 3% slower than Hawaii with 45.5% lower TDP.

Also, as much as HD7870 could be compared to last gen Flagship Chip HD6970 ( according to Transistor count), same can be done with RX 480 vs R9 390X ( according to Transistor count).
But again in both situations, the numbers tell a different story than your "common sense" missing targets that you have no idea what they really were.

 

[tviceman]

You want to talk about successors ??

https://www.techpowerup.com/reviews/AMD/HD_7850_HD_7870/26.html

HD6870 @ 40nm
Die size 255mm2, 1700 M Transistors , 151W TDP

HD7870 @ 28nm
Die size 212mm2, 2800M Transistors , 175W TDP

HD7870 vs HD 6870

Transistors = 65% more
Die size = 20% smaller
TDP = 16% more
Power Consumption = 103W-119W = 13,5% less
Performance = 39% faster

Now you're doing better! 6870 was a 2nd generation 40nm chip. Unfortunately, there was no chip from AMD on 40nm prior to Barts that was around the 200-250mm2 mark. Everything was bigger or smaller than that. But I like that you're actually making valid comparisons now! Good for you.
---------------------------

https://www.techpowerup.com/reviews/AMD/RX_480/24.html

HD7870 (R9 270X) @ 28nm
Die size 212mm2, 2800M Transistors , 175W TDP

RX 480 @ 14nm
Die size 232mm2, 5700M Transistors , 150W TDP

RX 480 vs HD7870

Transistors = 103% more (more than double)
Die size = 9,5% bigger
TDP = 14% lower
Power Consumption = 163W-119W = 37% more (Im sure they dont measure on the same game version or same drivers)
Performance = 85% faster

Its a huge performance increase, the biggest we have ever seen for a successor the last 3-4 years.

It's a pretty terrible jump, actually. It took AMD 4 years to squeeze out 85% better performance than Pitcairn at a similar die size, but with higher power draw. It took Nvidia 18 months, less than half the time, to nearly double the performance of GM206 with a smaller die size and similar power draw. Also In the same 4 years time, compared to GK106, Nvidia has gained a factor of 3x or greater in performance without increasing die size or power consumption.

Did you bother to take a look at the technical specs of Polaris 10 vs Hawaii ??? Polaris 10 has 8% less transistors and its only 3% slower than Hawaii with 45.5% lower TDP.

I did, thanks for asking! As I said, you're comparing Polaris 10 and Hawaii at a resolution that Hawaii sucks at. When you increase the resolution to Hawaii's more suited target market, Polaris is 10% slower. But since you want to make apples to oranges comparisons, look at what Nvidia did with GP104 vs. GM200....10% less transistors, 30% faster, and 30% lower TDP! Nvidia decreased their transistors by more, had a significant jump in performance, and also a lower rated TDP! Nvidia increased their performance, while AMD decreased performance. Nvidia increased their performance per watt by 75%, while AMD increased it by a lesser 50%. You tell me which is more impressive.... 30% faster AND 75% more efficient or 10% slower and 50% more efficient?

Also, as much as HD7870 could be compared to last gen Flagship Chip HD6970 ( according to Transistor count), same can be done with RX 480 vs R9 390X ( according to Transistor count).
But again in both situations, the numbers tell a different story than your "common sense" missing targets that you have no idea what they really were.

You keep wanting to make these apples to oranges comparisons.... so lets do it.

How fast was 7870 vs. 6970 at 7870's launch? TPU says it was 10% faster. How fast is RX 480 compared to Hawaii? If you go by Hawaii-gimping 1080p resolution, 3% slower. If you go by Hawaii appropriate 1440p resolution, 10% slower. Either way, Polaris 10 comes up short AND power consumption is noticeably higher than Pitcairn. Targets = missed. Common sense prevails.

 

[beginner99]

HD7870 vs HD6970

Transistors = 6% more
Die size = 45,5% smaller
TDP = 30% less
Power Consumption = 103W-184W = 44% less
Performance = 10% faster


RX 480 vs R9 390X

Transistors = 8% less
Die size = 47% smaller
TDP = 45,5% less
Power Consumption = 163W-264W = 38% less
Performance = 3% slower

Pretty much the same as going from 40nm to 28nm, they managed to have almost the same perf with 8% less transistors at 45% lower TDP and 38% lower power consumption when HD7870 had 6% more transistors than HD6970. It is clear the targets for Polaris 10 were , smaller die as possible with lower TDP as possible at the same performance as R9 390X.
Also, TPU use way too much GameWorks games and they measure power consumption in an old 2012 DX-11 game. perf/watt in DX-12 games between RX 480 vs R9 390X will be even higher.

But that doesn't change the fact that you had no idea what the official targets were for Polaris 10 and "common sense" is not an evidence of "missing" official performance and perf/watt targets.

Edit: And one last and extremely important aspect, we dont know what the R&D target was for Polaris.

Problem with that comparison is that 28 to 14 nm is the bigger jump than 40 nm to 28 nm was and 14 nm adds FinFet with is an additional advantage. If you look at Power consumption and Performance only it looks pretty bad for RX 480 especially considering that 390x had rather bad performance/watt to begin with (probably the worst in the entire stack).

Whether issues are due to the chip or the process is up for debate.

 

[Arachnotronic]

Problem with that comparison is that 28 to 14 nm is the bigger jump than 40 nm to 28 nm was and 14 nm adds FinFet with is an additional advantage. If you look at Power consumption and Performance only it looks pretty bad for RX 480 especially considering that 390x had rather bad performance/watt to begin with (probably the worst in the entire stack).

Whether issues are due to the chip or the process is up for debate.

28nm brought the use of high-K metal gates over the polysilicon gates in 40nm, so I would say that it's actually a similar jump process wise.