Nintendo's new Switch with updated Tegra X1 SoC delivers significantly improved battery...

Shawn Knight

TechSpot Staff
Staff member

Digital Foundry recently put the new Switch through the paces and found that in The Legend of Zelda: Breath of the Wild, the old Switch lasted for three hours and five minutes at 50 percent screen brightness. With the new Switch, they were able to game for five hours and two minutes – an increase of 63 percent.

When boosting the screen brightness to 100 percent, the results were even more pronounced. The original Switch shut down after two hours and 25 minutes but the new model persevered for more than four hours and 18 minutes. That’s an improvement of 78 percent.

The revised Switch also ran a bit cooler. The publication observed a maximum vent temperature of 54C on the old Switch when docked but just 50C on the newer model. When undocked, the gap between the two was just 2C (from 48C to 46C).

Nintendo owes its efficiency gains to an upgraded version of Nvidia’s Tegra X1. The original SoC was built on a 20nm process but teardowns of the new Switch suggest we’re now dealing with a 16nm chip. Updated LPDDR4X memory chips are also more efficient than what was used in 2017, iFixit notes.

The easiest way to differentiate a new Switch from the old model is the product box. The newer Switch ships in retail packaging with an all-red background that lacks a hand on the cover. You can also check the serial number on the box – the new variant starts with XKW. And if all else fails, have a look at the model number on the back of the console itself. The new model number is HAC-001(-01) – the old model doesn’t include the (-01).

Permalink to story.

 
"When boosting the screen brightness to 100 percent, the results were even more pronounced. The original Switch shut down after two hours and 25 minutes but the new model persevered for more than four hours and 18 minutes. That’s an improvement of 78 percent."

So they made the screen dimmer? I don't see how you get lower relative power consumption at 100% than at 50% unless you reduce maximum brightness or the backlight LEDs have somehow been made more efficient.
 
R

retsxel

If wonder if the new Tegra chip is made on TSMC's 16FF+ or 16FFC process; the former uses half the power of their 20 nm process, whereas the latter is half as less again. Either one goes a long way to explaining where the increase in battery life comes from.
There is also TSMC's 14nm and 12nm process to consider. Until they declare what process is being used we can only wonder.
 

MaikuTech

TS Evangelist
Lets see how well this goes, when trying to trade our current switch systems with the newer ones at game stop.
Price gouging moments at its worst, with nintendo's hand in the mix.
 

Markoni35

TS Addict
Oh, wow. They dimmed it. I bet with a completely switched off screen it would last for 8 hours. While it may make watching the movies slightly less enjoyable, with the screen off you can watch 5 movies in a row on the same battery.
 

neeyik

TS Evangelist
Staff member
There is also TSMC's 14nm and 12nm process to consider. Until they declare what process is being used we can only wonder.
If it is on a different process, the Tegra X1 is more likely to be 16FF, as the new chip revision is only marginally smaller than the old one. TSMC's 20 and 16FF share the same back-end of the production process, so there's not much difference between the overall areas (comparing like-for-like), but power consumption is significantly less. TSMC don't have a 14nm line; their 12FF lines are the same as their 16FF ones, but with slightly narrow pitches (gap between parts), allowing for a raised transistor density and thus smaller chip.
 

neeyik

TS Evangelist
Staff member
Reread those articles carefully - they both use a DigiTimes piece as the source, which states an unconfirmed rumour that Intel would outsource production to TSMC. Neither company confirmed this and none of the reports addressed the point that TSMC does not have a process node equivalent to Intel’s 14nm; the nearest they have is their 16/12 node. If TSMC was being used, then Intel’s chips would have been on one of the 16nm variants closest to the required dimensions that Intel used in their respective 14nm.
 
R

retsxel

Reread those articles carefully - they both use a DigiTimes piece as the source, which states an unconfirmed rumour that Intel would outsource production to TSMC. Neither company confirmed this and none of the reports addressed the point that TSMC does not have a process node equivalent to Intel’s 14nm; the nearest they have is their 16/12 node. If TSMC was being used, then Intel’s chips would have been on one of the 16nm variants closest to the required dimensions that Intel used in their respective 14nm.
Ah, but how hard would it be for TSMC, with their vast experience, to adapt an existing process? Still I concede. Read the articles, didn't follow them up.
 

neeyik

TS Evangelist
Staff member
Well, it's worth noting that the numbers used (I.e. 16nm, 14nm) don't really mean anything - for example, TSMC's 20nm, 16nm and 12nm process nodes are all virtually the same 'size'; there's only marginal differences between the physical dimensions of the various parts of the transistors. It's how the wafers are fabricated that generates the revisions to the process nodes.

But let's assume for a moment that Intel had approached TSMC will a view to getting some of their 14nm products manufactured. We know that Intel have 3 revisions of their P1272 process node: 14, 14+, and 14++. There's some small changes in the pitches between the revisions, but generally speaking the dimensions are the same. Now TSMC's 16/12 process produces finFET transistors that are generally bigger and require more voltage (albeit by just a small amount) than Intel's 14 lines; the overall manufacturing process is the same at the front-end of the production line but there are big differences at the back end.

Could TSMC adapt this for Intel? Yes, certainly, and they're fully capable of doing some very specific or modified designs on the 16/12 line (see the recent Cerebras megachip launch). But the more the design falls away what TSMC can mass manufacture, the lower the yields and the lower the profits. This is why Intel wouldn't turn to another fab company to meet demands or at the very least, they certainly wouldn't publicly announce such a deal.

Another question to ask is could TSMC just manufacture the Intel design on the 16/12 process 'as is'? Again, yes they could, but the product would almost certainly not run in the same manner as if it was made by Intel, and then you'd have a bigger problem on your hands.

Finally, demand varies far quicker than supply does - from start to end (silicon to boxed product), a complex CPU takes well over a month to be made, so it's not like TSMC could have just picked up the order, modified a few things, and then churned out a large volume of goods for Intel to meet their demands. Instead, Intel would have looked at their production schedules, moved things around, changed a few work patterns, etc to increase the manufacturing rates for the products they were short on.