GTX 1070 power pin question.

kol12

Honorable
Jan 26, 2015
2,109
0
11,810
Am I correct in thinking that the 1070 doesn't come with anything more than an 8 pin connector? I've just brought an EVGA 1070 SC which has an 8 pin. How does the 1070 consume roughly the same amount of power as the 970 with 4 pins less? Is this an efficiency improvement?

I'm seeing a nice reduction from 0.875v to 0.650v and 14.8w to 8w with the 1070 vs the 970 at idle which may result in slightly lower idle temps. The video and memory clock idle speeds are a tad higher on the 1070 at 240Hhz/202Mhz vs 135Mhz/162Mhz on the 970.
 
Solution
Some GTX 1070's have 1x 8-pin, some 6-pin + 8-pin and some 2x 8-pin. This is because the different board manufacturers are customizing 5 distinct GPU variables that determine the power envelope: (1) The type of memory, (2) Memory and core clock speeds, (3) Voltage regulator module parameters, (4) number of fans, and (5) additional lighting effects.

If you take all these variables out of contention and you consider the bare Pascal vs Maxwell chips, then you will find that the microarchitectures are quite different. Maxwell was first introduced in 2014 with the GTX 750 and was based on 28nm technology, meaning that they could ultimately fit 8000 million transistors on the 601 mm2 die, resulting in 96 GP/s and 192 GT/s respectively with...

cilliers

Honorable
Jul 13, 2012
825
0
11,360
Some GTX 1070's have 1x 8-pin, some 6-pin + 8-pin and some 2x 8-pin. This is because the different board manufacturers are customizing 5 distinct GPU variables that determine the power envelope: (1) The type of memory, (2) Memory and core clock speeds, (3) Voltage regulator module parameters, (4) number of fans, and (5) additional lighting effects.

If you take all these variables out of contention and you consider the bare Pascal vs Maxwell chips, then you will find that the microarchitectures are quite different. Maxwell was first introduced in 2014 with the GTX 750 and was based on 28nm technology, meaning that they could ultimately fit 8000 million transistors on the 601 mm2 die, resulting in 96 GP/s and 192 GT/s respectively with the Maxwell Titan X. This all was done within a 250W power envelope.

When considering the GTX 970, the card in discussion, the results were 54.6 GP/s and 109 GT/s on a 145W envelope, with a die size of 398 mm2, including 5200 million transistors. It resulted in 3494 GFLOPS single precision processing power and 109 GFLOPS double precision processing power.

With Pascal on the other hand, the microarchitecture is based on 16nm technology. Meaning that with the same amount of real estate, engineers are able to fit almost twice as many transistors onto a die. As a result, the GTX 1070 chip has 7200 million transistors on a mere 314 mm2 die. It pumps out 96.4 GP/s and 180.7 GT/s on a 150W power envelope. This results in 5783 GFLOPS single precision processing power and 181 GFLOPS double precision processing power.

So in terms of effiency gain, the theoretical comparison would be in terms of GFLOPS per Watt or GT/s per Watt or GP/s per Watt. Take your pick. As with the microarchitecture, that is almost twice as small on an even smaller die size, these ratios will probably be similar.

In layman's terms: A simple analogy for me has always been to think of a chip as a factory with 10 workers, making x-amount of product while the same factory will make 2x-amount, given that the resource inputs and outputs will now also double. Its all about real estate, until they start implementing 3D transistors on large scale.

Ref: http://thepcenthusiast.com/geforce-gtx-1070-compared-asus-evga-zotac-msi-gigabyte/
Ref: https://en.wikipedia.org/wiki/GeForce_900_series
Ref: https://en.wikipedia.org/wiki/GeForce_10_series
 
Solution