Despite the benefits of LPDDR(X) RAMs, they are typically available only in soldered form. This is against the design philosophy of Framework with respect to upgradability and repairability. It seems to me that we have not seen LPDDR(X) RAMs in a standard socketed form because they were primarily developed targeting the smartphone industry and so far, there is no major commercial demand for it. There is also engineering challenges to achieve this but hardly anyone ever tried that we know of.
Recently, I encountered these LPDDR sockets available commercially. It seems they develop solder-less sockets for other BGA mounts as well. Can anyone knowledgeable in this matter enlighten me on this? I am not sure about the cost but that will reduce as the adoption increases I hope. Does this really indicate the possibility of socketed LPDDR(X) RAMs in near future?
@Abhisek one challenge with supporting LPDDR chips directly is that any system using it hard codes the timing information for the chips into the bios/firmware. For SODIMMs there is an eeprom on the PCB that saves the timing information (SPD).
Imho, LPDDR(X) is designed to make Apple, MS and Dell corporate rats happy in their attempt to stop freedom and 3rd party intrusion into their laptops + charge extra hundred(s) dollar if user wants to have more than 8GB of RAM. I don’t think open standards should ever follow this technology.
So, if someone wants to replace a LPDDR chip he/she needs an exact same replacement (with identical specs)? If that is the case, it certainly limits the flexibility in terms in terms of upgrade but theoretically does not affect repairability?
Also how hard do you think it would it be to provide an option in the bios to set these timing information if someone decides to go with a RAM chip with slightly different spec? I can guess, it won’t be something targeted to the average users, but a lot of potentially dangerous overclocking options these days are available in the BIOS of a decent gaming motherboard.
I don’t know how much these companies have influenced JEDEC in the development of LPDDR standards but I agree that these companies benefit the most out of it. The soldered RAM chip certainly helped them extracting more money from their customers in the name of low power consumption and better speed while killing repairability in the background.
Despite the drawbacks, lets not deny the fact that LPDDR(X) does have some important advantages over normal DDR. Why not thrive to innovate so that we can have best of both worlds rather than outright discarding a piece of technology? Until now we couldn’t imagine thin and light but highly upgradable and repairable like Framework is practically possible. But here we are, that is because people at Framework decided to innovate and change things which were really hard to pull off.
LPDDR has a lot more pins than normal DDR4 for the same memory capacity. DDR4 SODIMM’s already have 260 pins which are pretty tightly packed. Not everything is a conspiracy by big laptop to hurt reparability.
LPDDR4 is designed to run down to 1.1V, while LPDDR4x can run down to 0.6V. DDR4 is specced at 1.2V, while DDR5 is specced at 1.1V. Things are improving, but companies can’t produce new versions of DDR every year for a 5% reduction in power.
These are test sockets that are made for practically every chip, BGA or not. They cost $100’s-$1000’s per unit. Here is what a full socket looks like.
.
LPDDRX has way more pins and uses a different type of subsystem as mentioned here. However it doesn’t mean an open standard couldn’t be made to make these chips socketable in some way. It would be great to not have soldered memory.
Would you consider having an LPDDR4x/5 option in the future? I know it breaks upgradability and repairability for this one part, but as there is no GPU with dedicated memory, having higher RAM bandwidth could be beneficial for running (simple) games. I think RAM is the one part where breaking upgradability has actual real benefits.
Of course this would be in addition to the SO-DIMM version, so I guess it would require you to design two different motherboards.
Well that would be ideal, but would require creating a new standard, not sure Framework would be up for that. Also, not sure if it’s even technically possible, as I understand it LPDDR4x/5 signals have such tight tolerances that SO-DIMM simply wouldn’t work. Would love to be proven wrong though.
@ajp_anton Framework could develop a new form factor using mezzanine style connectors which increase pin density. It would take more money to develop which framework should put elsewhere but it would be very neat
Yea LPDDR4X would be nice, but it is soldered only. A modular version of this LPDDR4 or LPDDR5 when it comes out would be awesome. I am sure some other laptop mfg’s would hop on this if it could be made.
@RandomUser But DDR5 isn’t supported Tiger Lake. LPDDR4x is faster than DDR4. And when DDR5 support gets here, LPDDR5 will still be faster. Soldered RAM has its benefits. Plus it has low power benefits.
What other options are you talking about that have easily replaceable wifi, battery, keyboard, screen, bezels and IO? If you think RAM is the only thing Framework does differently, there are already options for that as well.
Personally, I regard soldered RAM as a drawback when considering buying a laptop, even with 32+ GB of RAM. As someone working in Desktop Support, I’ve lost count of the number of old desktops that my team has repaired by replacing the RAM.
Don’t forget about the benefit of repairability! It can prevent a computer from becoming e-waste long past the end of its warranty. Also, a benefit specific to this product: if/when any motherboard upgrades are offered that use the same generation of RAM as the previous motherboard, RAM could be re-used.
Edit: or what if you need to replace a damaged motherboard? There are still many parts on it that can fail. It’s great to not be forced to throw away expensive memory chips.
Also note that capacity is not the only possible upgrade, though this is irrelevant for the vast majority of users who already have the highest supported clock speeds, CAS latency and subtimings.