Intel lunar lake more environmentally friendly despite soldered RAM

General Topics
1

I can’t wait for LP-CAMM as much as the next person, but the reality is that we aren’t there yet, and it will probably be at least a year before we see it on the framework. Unfortunately I need a laptop soon, and I wish that Framework would make an Intel lunar lake version of the 13 inch. I’m worried that nobody will produce an x86 chip this efficient soon since:

  • Intel probably won’t use the more efficient TMSC node again and it will take a long time for Intel’s foundries to catch up (see stock price lol)
  • AMD always combines their desktop and mobile architectures leading to lower efficiency for light workloads
  • Soldered RAM helps, and I can’t see Framework doing LP-CAMM soon enough

The following details how from an environmental perspective, the tradeoff of soldered RAM for battery life is probably worth it this generation. We use global warming potential (GWP) as the key metric and draw data from frameworks life cycle report 2022.

It’s still early days in the review cycle, but it seems that intel’s new lunar lake products have at least 20% longer battery life in light workloads than previous generation / AMD in framework laptops (source). Therefore under perfect conditions that the battery lifespan will be ~2x longer (see Table 4 in source). In this analysis we will assume that the battery lifespan for the new Intel lunar lake products is a more conservative 50% longer. Hence if our base configuration has a battery life of 5 years then the Intel lunar lake version will have a battery lifespan of 7.5 years. This gives us two scenarios:

Hypothetical Intel Lunar lake config: (lifespan 7.5 years)

  • 32GB of soldered RAM: 18.4420.8 = 29.5 kg CO2e GWP [source: Table 5-3, Figure 5-5 Framework-Life-Cycle-Report]
  • Battery: 8.46 kg CO2e GWP [source: Table 5-3 Framework-Life-Cycle-Report]

Total: 38.0 kg CO2e GWP

Average per annum: 5.20 kg CO2e GWP

Baseline Config: (lifespan: 10 years)

  • 32GB of SODIMM RAM: 18.44*2 = 36.9 kg CO2e GWP [source: Table 5-3 Framework-Life-Cycle-Report]
  • Battery: 8.46 kg CO2e GWP [source: Table 5-3 Framework-Life-Cycle-Report]
  • First Battery replacement: 8.46 kg CO2e GWP [source: Table 5-3 Framework-Life-Cycle-Report]

Total: 53.82 kg CO2e GWP

Average per annum: 5.38 kg CO2e GWP

Hence, the global warming potential (GWP) of the batteries and RAM on a new Intel lake model is about 3% lower per year than the best current models from Framework. Note that we are being generous to the baseline config here by assuming that we continue using the device for another 2.5 years with the same RAM. A better analysis would include consideration of upgrades etc. The point I’m trying to make is that efficiency is really important.

Additional assumptions:

  • We only consider framework 13 model
  • We ignore lower power consumption with the Intel lunar lake model
  • The customer can accurately predict how much RAM they need in the next 7.5 years (32GB)
  • RAM failure rates are vanishingly low (post testing - see bathtub curve)
  • The baseline config has dual channel RAM (apples to apples for Intel lunar lake). Hence after 10 years at least one would need to be replaced (or a whole new mainboard)
  • A users’ charging behaviour doesn’t change based on their laptop’s battery life
  • We ignore differences in manufacturing nodes and die sizes of CPUs
1 Like
2

Unfortunately, a Lunar Lake version of the Framework is basically impossible as it would be a complete 180 from Framework’s mission statement.

Lunar Lake uses on package memory, so the memory is physically on the CPU package. Unlike even the current LPDDR5, this on package memory 100% has no upgradeable version.

LPCAMM2 is a replacement for soldered memory on the motherboard, but there is no replacement for on package memory which is what ships with Lunar Lake. This is why the CPU model now determines the amount of memory your computer will have when using the new Lunar Lake laptops.

Framework’s main goal for reducing carbon emissions and helping the environment is to reduce e-waste, not reduce electricity consumption. By having upgradeable components, they reduce the number of parts that have to be replaced whenever something breaks reducing e-waste.

In the end, reducing power consumption is a nice positive, but it isn’t on the list of top priorities for Framework. From their about page, you can see that it is mainly mentioning helping the environment by reducing e-waste, rather than power consumption.

1 Like
4

Lock screens at 250 nits max to be environmentally friendly too.

5

I think you may have missed the point of the post. In the GWP calculation I actually ignore the reduced electricity consumption from the grid (see additional assumptions).

The main points are:

  • That soldered/on package RAM takes ~20% less emissions to produce compared to SODIMM due to no gold being used in contacts
  • Lunar lake is significantly more efficient which means that batteries have to be replaced significantly less often leading to less E-Waste. The battery is the component of the laptop which fails first by far (assuming laptop has good structural build quality).
  • Intel thoroughly tests their RAM (first part of bathtub curve) and as a result RAM is extremely unlikely to die during use
  • After 10 years of use (see baseline case) the user will probably want to upgrade their mainboard which probably comes with a new generation of RAM anyway (eg. DDR4 to DDR5). Or upgrade their in which case they still need to remove one of the 16GB sticks in the above scenario.
6

32Gb being enough in 7.5 years is a very bold assumption. It’s probably not going to be unusable but enough is a bit of a stretch. It doesn’t take a lot of mainboards ending up in e waste instead of getting new ram to evaporate your advantage.

Not sure that’s fair, the multi die packaging process may easily have a bigger impact than the little bit of gold in the socketed ram.

Discharge cycles is only one of the sources of wear on a battery so the impact there may be a lot les significant than you think.

Given recent event I’d take reports of intel thoroughly testing anything with a grain of salt XD.

4 Likes
7

32Gb being enough in 7.5 years is a very bold assumption. It’s probably not going to be unusable but enough is a bit of a stretch.

Depends who you are, I’m confident for my use cases. In 7.5 years I’m also confident that the base spec of most laptops will be 32GB or lower. Maybe LLM’s will change things, but I have no current desire to run them locally.

It doesn’t take a lot of mainboards ending up in e waste instead of getting new ram to evaporate your advantage.

Maybe less than you think. On package RAM savings + one less battery replacement is 45% of the GWP of a mainboard according to Framework (see table 5-3).

Not sure that’s fair, the multi die packaging process may easily have a bigger impact than the little bit of gold in the socketed ram.

I’m not sure what part of the multi die packaging process would have significant GWP? But true, there are a lot of variables here (eg. smaller dies of lunar lake), feel free to contribute your own analysis :slight_smile:

Discharge cycles is only one of the sources of wear on a battery so the impact there may be a lot les significant than you think.

Yes, but cycles are by far the largest one assuming the user looks after their battery (ie. not let it get too warm - easier with lunar lake). If you have stats to prove otherwise, please share!

1 Like
8

I am personally not that sure predicting my needs that far ahead but you may well be right.

It’s a very complicated process that can fail lowering yields, not sure if rejects get calculated into the gwp of an individual one but the resources were used regardless. Then again it is on a tsmc node so the better yields from that may cancel it out.

Point is it’s hard to know and the info you’d need is porbably not publicly available.

That highly depends on the usage profile, if you use the thing mostly plugged in the wear comes primarily from other sources than cycling, if you constantly charge discharge it the wear comes from that.

Also lunar lake only really has better efficiency at very low loads which also produce little mileage, amd still has more efficient cores when actually doing something so the actual difference there heavily depends on how it’s used, so the difference may be smaller in real world use.

There is also the point of the secondary market, it’s a lot easier to resell something with socketed ram which would keep it out of the trash longer but that’s a whole other can of worms.

10

They will be using TSMC’s nodes on Arrow Lake, which is the direct successor to Meteor Lake and does maintain support for replaceable memory and replaceable Wi-Fi.

Furthermore Lunar Lake uses TSMC’s N3B node. TSMC’s N3B node has struggled with high defect rates so when considering environmental impact you should also consider the waste caused by the defective processors. Around a year ago it was reported that the Apple A17 chip was encountering ~45% defect rate, whereas Lunar Lake is ~40% larger which likely would mean over 50% defect rate, which is well over double the normal levels.

AMD’s desktop architectures have lower efficiency in light workloads, however that is primarily due to their use of the CPU being two (or three) separate chips working together and the link between those chips struggling with efficiency under light load.

AMD’s laptop architectures on the other hand are a single chip, eliminating that link between chips and improving efficiency for light workloads.

Other differences between AMD’s desktop and laptop architectures include often using a newer TSMC node (ex. Zen4 was 5-6 nm on desktop and 4 nm on laptop), having less cache (cache is less important since the memory controller and CPU cores are on the same chip on AMD laptop CPUs, removing cache reduces defect rate), use of efficiency optimized cores (starting with low-end SKUs of Phoenix/Hawk point and now all Strix Point SKUs), and some other differences.

AMD also even had the mobile specific Zen3+ which brought massive efficiency improvements.

I’d say that they’re on a similar level of combined-ness to Intel.

In my experience ram failure rates are low, however I’ve had a concerning number Wi-Fi cards fail on me (mainly Intel Wi-Fi 5 cards, not yet had any Wi-Fi 6 cards fail on me so maybe things have improved) so I’m concerned by those also being integrated on Lunar Lake.

Also, even assuming 100% reliable ram and Wi-Fi, if something else were to fail on the mainboard in a way that isn’t worth repairing (motherboards drop in value quickly over time, increasing the likelihood that one won’t be worth paying a repair shop to fix) having the ram and Wi-Fi integrated means that people would be forced to replace those even they’re fully functional (creating more e-waste).

 

There are also other disadvantages to Lunar Lake aside from the soldered ram and integrated Wi-Fi.

For example Lunar Lake targets a lower performance tier than Meteor Lake (what Framework currently offers) or Arrow Lake (the successor to Lunar Lake and big sibling of Lunar Lake) due to having half the core count.

Ever since the AMD Framework Laptops were announced I have seen a lot of people talking about how one of the big advantages of the Intel models has been that there are no limitations on functionality with the modular IO (whereas on AMD models only the rear ports are fully capable). Personally I don’t mind those limitations, however clearly some people do. However Lunar Lake only supports three Thunderbolt 4 ports, meaning that there would then be restrictions.

I also believe Framework is likely interested in expanding their Intel offerings to the Framework Laptop 16 aswell. With the AMD models they use the same CPU family on both the 13 and 16, allowing them to share a chunk of the engineering work and firmware. I suspect Framework would like to do that on the Intel side as well, however Lunar Lake is a non-starter for the 16 due to its limited PCIe lane count (even AMD’s new Strix Point poses challenges due to reduce PCIe lane count and it has twice as many lanes as Lunar Lake).

In an world where Framework has unlimited engineering capacity and budget I would love to see both a Lunar Lake and Arrow Lake model, however I think that realistically Framework will have to pick between them and will pick Arrow Lake.

12

In addition to carbon footprint, I think what’s more important for most end users of 13-inch laptop is battery life, which is why Apple’s M-series have gained so much market despite its poor repairability. I bought Framework 13 because it offers comparable performance, aesthetics and power efficiency to other Windows laptops, without compromising on repairability. This won’t be true anymore if Framework does not offer any Lunar Lake options.

Actually, given the modularity of Framework 13’s design, why not offering both Lunar Lake and Arrow Lake-H to users, who can choose based on their use cases? If one’s use case changes, he/she can always exchange a mainboard for either higher performance, more RAM or better power efficiency.

13

I would also say that, for me at least, not only is day to day battery life important, long term battery life is as well. With the Framework, I like that if my battery starts to drop in performance after a few years, I can easily source and easily replace it. At least, in theory. Assuming the company doesn’t fail in that time, or something.

I have some older devices that have gotten to the point where they only work when plugged in. And replacement batteries are few and far between, questionable quality, and expensive.

Everyone’s needs are different. But for me personally, I’m happy with the 6-8 hours I get out of my FW 13. I’m happy to plug it in a bit more often, if it means I can easily swap out the battery if/when that 6-8 hours starts to plummet, due to a failing battery.

15

Or maybe Lunar Lake only for 13, and Arrow Lake-H for 16.

A Lunar Lake MB will lose memory upgradeability/repairability, while retaining all other awesome repairability benefits of 13’s design. Meanwhile, the real estate saved from SO-DIMM slots can host either better heatsink or another M2 2230 slot, for example.

16

Glad that the current design works for you, but I’m afraid most mainstream Windows users would choose a much improved battery life of Lunar Lake. According to reviews of Lunar Lake that I’ve seen, it also performs as strong on battery versus plugged in, much like Apple’s M-series. This would be a much better user experience than previous Windows laptops.

17

I’m waiting for this…or newer from Framework. Initially placed an order for the 165H…but performance appeared rather weak.

19

Same here…and / or swap RAM if RAM dies.

1 Like
20

Also having the option to upgrade the RAM after the time of purchase is a nice addition that Lunar Lake doesn’t offer. It’s either 16GB or 32GB and if you find yourself needing more RAM than previously expected, that would mean a whole new mainboard.

1 Like
21

" It’s still early days in the review cycle, but it seems that intel’s new lunar lake products have at least 20% longer battery life"
->> “200% longer battery life” FIFY
Let me help you with the math: from 5h to 15h that is 200% additional battery life, and 300% longer in total.

22

where are the 5h coming from?

23

From your current laptop batter life, which I assume is not 12.5h.

24

Depends on the workload, it surely isn’t just 5h in whatever workload lunar lake does 15 though.

With my current test numbers it’s about:

  • 17h idle sitting on a desktop with 2 terminals open and nothing else (3.5W)
  • 8.5h 720p30 windowed youtube playback in firefox (7.17W)
  • 7h 4k 60 fullscreen playback in kodi (8.7W)

That testing is done at 20% brightness, wifi connected, bt on but not connected, kb backlight off and webcam and mic switch off (running relatively current arch and sway). Testing is measuring power consumption over a 10minute window.

The video playback numbers could be better on windows if you can get it to not doo too much background crap, windows still makes better use of the hardware decoder but linux has improved a lot since I got this thing.

1 Like
25

GTFO, you’re not getting even half that… use google and stop being funny.