13th gen Intel mobile processors can use either DDR4 or DDR5, though you can’t mix both in the same system. Framework’s previous 13" systems with 13th gen processors used DDR4, but the FW12 will use DDR5.
3 Likes
Not quite, they doubled the channels but halved the bus width, the whole bandwidth advantage is coming from the increased frequency.
In a weird twist intel is dominating the low-end now as amd has mostly abandoned that segment.
Single dimm AND ddr4 would be pretty brutal.
Personally I think they should have gone with a hybrid model like the t480s where you have 4/8gb soldered and the rest in a sodimm slot. still gives you a bit of dual channel for the igpu but you also still only need space for one dimm slot. Don’t think the cost of 4/8gb of ram would make a huge difference especially since they could sell a base version with no sodimm.
3 Likes
Yeah, pressure sensitivity determines whether I even buy it or not. It would be my main reason for wanting it at all.
4 Likes
Please, no soldered-in RAM. Failing DIMMs are enough of a hassle. I have not so long ago retired both - a laptop with soldered-in RAM and a DDR5 DIMM.
6 Likes
Adding a way to disable the built in portion would mitigate those issues and still have the benefit of a bit of dual channel as long as the built in ram survives.
This is likely way more difficult than you think. Including any kind of switch (either physical or electronic) would probably have a very large impact on the very sensitive signal.
Of course some operating systems can be instructed to ignore certain sections of broken RAM in software, but that also has its limitations.
Thing is the soldered ram is on it’s own channel so you’d likely just need to disable that (so just cutting the connection to the spd would probably already do the trick). If you did have them on the same channel it would be a whole different story but that would also be a psychotic thing to do.
That is more of a last resort.
2 Likes
I have been waiting a very very long time for a product approaching this description. Color me extremely interested, though I wish it were possible to know more about the specifications in advance of the pre-order release drop so that we could make a more informed decision at that time.
1 Like
That is what I had to do with the laptop with soldered-in RAM, in the end, but that really was the last resort.
I would have much less of an issue with it if it was an ECC RAM, but I am yet to see a laptop supporting that for less than a price of a (not so small) car.
Ecc only goes so far and ddr5 already has on die ecc to cover up some defects so if the errors manage to get out of the chip there is enough damage for full blown ecc to also not help anymore.
1 Like
Then there is yet another consideration - a board with soldered-in RAM is a liability due to the RAM unreliability. You will have more (potentially costly) returns. It would have to be more expensive for this.
I am not an expert. I just don’t want an unreliable non-exchangeable component in my laptop. One of the reasons I went for Framework 
1 Like
You can easily mitigate that risk by making the soldered memory disableable so failing soldered ram can^'t take the whole board down with itself.
Also ram fails extremely rarely compared to other parts of the board so that concern is pretty small, you may have been extremely unlucky.
I am all for not soldering stuff if possible but having some (important just some and not all) ram soldered is a lot better than just leaving the memory channel completely unused.
2 Likes
It is not a matter of whether you can disable it, it is a matter of warranty. I cannot see anyone agreeing that something broken in your device is not a problem, because you can just disable it.
Yeah, maybe. I’m happy not to test my luck too much.
I beg to differ. I’d rather have an unused feature in a reliable machine, than used one in an unreliable machine.
But that’s me.
Well during warranty you are getting it replaced anyway but after warranty you are not screwed, you just loose the one memory channel you aren’t even getting in the current configuration.
If it actually was especially unreliable I’d agree but it really isn’t.
Well you are getting what you want here I suppose.
1 Like
I do not want to be especially contradictory, but
8% of DIMM modules per year. I don’t know about you, but to me it does not sound too reliable.
That’s a different topic, we are talking about the things failing not just having regular bit errors which is part of the ecc debate that applies to sodimms just as much as soldered memory. The study is also ancient and does not include ddr5s on board ecc and newer signaling standards, so it could have gotten better or worse. Though arguably soldered ram should perform slightly better in that regard as it has better signaling quality due to there not being connectors.
If this was a debate of should we put ecc on everything then I’d be all the way with you but unfortunately we or framework don’t have much of an influence there. It is likely a question of time though as making things faster seems to almost require it.
6 Likes
Actually, it isn’t. In the first column on the first page of the report I referred, it says:
We provide strong evidence that memory errors are dominated by hard errors, rather than soft errors, which previous work suspects to be the dominant error mode.
Sounds like a persistent hardware error to me.
“Should perform” is not a hard data, though, is it?
You do have newer figures backing up your “it really isn’t” claim, don’t you? How about sharing? Otherwise it is just hand-waving.
Also - you do understand that the cost of replacing a mainboard during warranty due to a failed RAM is bigger than the cost of replacing a failed module?
We are talking about two different thing here, one is bit errors which are a thing and are mostly hardware based but they aren’t failed as in “go replace that stick of memory” failed they just happen. The other is ram is broken need to replace it or computer won’t work anymore, you will get warranty replacement for the second. There is a difference between a few bits a year and failing memtest in a couple rounds.
Do you really need proof that shorter hard wired connections have better signal integrity than much longer connections that have to go through a connector?
Obviously but you need to divide that cost by how often that happens. With how many reports of faulty ram slots I have seen which also require a board replacement I don’t think that would make a significant difference but it is of course a concern.
1 Like
Which part of “hard errors” you did not understand? These are ones that continue after a re-write. Had you actually read the report, they say what these are - stuck bits. I daresay they will fail every round of memtest.
So on the “I have data to back my statements” that’s a “no”, or?
And again you do not back your statements with any data. I do not know what you’ve seen. It might have been a 
for what I know.
This discussion is going nowhere if you keep relying on hand-waving, ignoring data, and shifting the goalposts. Unless you can commit to stick to facts, this is my last reply to you.
The part where hard error is the ecc term which means not correctable by ecc and not physically broken. Just cause google had the policy of throwing the dimms when those happen doesn’t mean the stick is actually broken. I did have to re read it though cause I started thinking I was crazy for a bit there.
I daresay the bulk of those dimms will pass a lot of rounds of memtest.
This discussion is indeed going nowhere and I am certainly hand waving but I am not ignoring relevant data and absolutely not moving the goalpost XD.
Anyway it has been a pleasure mud wrestling and ultimately the product is following your preferences so there is that.
3 Likes