[abound]

The soldered RAM is surprising for Framework, and doubly surprising for being so in a form-factor that usually doesn't have soldered RAM.

Similar to what other commenters have expressed, it just seems like they shouldn't have built this product if they couldn't figure out the soldered RAM bit.

[danielEM]

That is not a framework choice, that is an AMD architecture that doesn't use regular RAM modules as it requires wider data bus

[zamadatix]

The statement above is that one would expect Framework to have chosen a platform which does not require soldered RAM, not that Framework kept such an option by choosing this AMD part.

All that aside, I absolutely can't wait for desktops to decide to go the same route of having 4 memory channels instead of 2. Right now the only way to have >2 channels is to buy workstation/server class stuff or an APU.

[sliken]

Sadly there's no alternative if you want 256 bits wide at the moment. Well a previous generation threadripper with a huge increase in power consumption and requiring a discrete GPU.

BTW, careful with "channels". Todays normal desktops and laptops (besides the strix halo and apple pro/max/ultra) are 128 bits wide. But in the DDR4 -> DDR5 transition they doubled the number of channels to 4. However the strix halo biggest advantage is the extra memory bandwidth going from 128 bits wide to 256 bits wide.

Finally the technology of the ps5 and XboxX comes to laptops, tablets, and small desktops.

[akritrime]

From their LTT video, it looked like they chose to do Desktop because of the AMD platform (and not the either way around, where they planned a desktop product and then chose AMD strix halo). Apparently setting up the manufacturing pipeline for laptops built on strix halo is expensive as of now and there are only two laptops in the whole market using it. So Frame.work choose to go the desktop route to save on cost while still making the platform available for everyone.

[elxr]

But think of it this way, anybody who specifically wants this chip (Ryzen strix halo) now has a way to buy it in a somewhat modular platform.

[soerxpso]

The architecture was their choice. It's completely possible to build a PC with that form factor which does not have soldered RAM, so it's strange that they went with it. Their brand with laptops is supposed to be 'more repairable' but they've chosen to make a desktop PC that's less repairable than their laptops are? It doesn't make any sense with the rest of their lineup. The person who would want this product is the opposite of the person who would want their other products.

[kllrnohj]

That market, "repairable modular desktops", is extremely saturated. This only exists because nobody was putting the Ryzen AI Max+ 395 in anything, so they got gifted the entire market of "GPUs with more than 32gb of RAM that aren't Apple." And then a desktop because fitting this in a laptop is hard (hence why nobody else is doing it, either)

[zamadatix]

> This only exists because nobody was putting the Ryzen AI Max+ 395 in anything... And then a desktop because fitting this in a laptop is hard (hence why nobody else is doing it, either)

The Ryzen AI Max+ 395 was paper launched a month ago. What Framework announced here is just another pre-order (even batch 1 won't ship until Q3). There are dozens of other systems, including mini-PCs, announced months ago which are set to launch before this. Small laptops were actually the first 395+ products to launch, hence this month's benchmark reviews all used them. E.g. https://www.tomsguide.com/computing/laptops/gaming-laptops/a...

I would not be surprised if one will be able to get a full 128 GB 395+ mini PC for less than the cost of the Framework Desktop baseboard before the Framework Desktop actually starts arriving in hands. What you're buying here is a premium to be able to replace the shell and front USB ports.

[ezschemi]

Where can I find those mini-PCs? The HP Z2 Mini G1a is "Coming Soon". If this is to be believed: "2025年5月以降販売開始予定", which translate to "Sales scheduled to start after May 2025" That is Q2 2025 - at the earliest.

I haven't found any other with a committed release/ship date.

This Reddit thread tends to agree: https://www.reddit.com/r/MiniPCs/comments/1ieiate/when_are_r...

Even the Asus Z13 laptops with the 395+ have no shipping date yet.

Looking at the two GMKtec systems with their rattling fans I have here, I would rather take the Framework system where I can easily swap the fan. Cooling those 100-120W will take space and air flow.

[zamadatix]

They were announced months ago, not shipping months ago. Framework announce Q1 and is shipping Q3. Again, keep in mind the CPU itself only launched last month. This is the real reason the 395 had not been put in anything yet.

Plenty of folks on Reddit have already received shipment (not order) confirmation of the Z13. It looks like even more as of this morning. I doubt if any have actually fully received said shipments yet as it has only been a couple days since launch. It'll likely remain pretty out of stock unless you managed to get an order immediately at launch, again due to the paper launch of the 395 last month and low 395 availability until Q2.

I didn't know HP was making one too! That'll be exciting as it's easier to order that at my work than e.g. an ASUS one or what have you.

[kcb]

This almost definitely did not happen.

[sliken]

Yes it's possible, but at the cost of 2-3x less memory bandwidth, which is a key feature to provide GPU and large language model performance.

[baobabKoodaa]

> The person who would want this product is the opposite of the person who would want their other products.

That's me! I'm that person! \o/

[abdullahkhalids]

I don't know anything about RAMs or their bus size. Is this something that will be "fixed" in the future, idk, with DDR6? Meaning we can have replacable RAM with such bus.

[sliken]

Problem is there is no socket available for thin/light/low power CPUs to have a 256 bit wide bus.

So to ensure 256 bits @ 8000 MHz works well AMD did it in the same package. In theory they could A) ship a halo without ram B) design a new socket, C) allow motherboard makers to pick 2 x CUDIMMs or 4x DDR5 dimms. Not sure that would buy them much market though.

As apple and others have proved, not many care about replaceable dimms, especially if it gives them 2-3x the performance and/or better perf/watt.

[neRok]

Apparently not. In Linus Tech Tips latest video, the Framework CEO says they talked to AMD about it and that one of their engineers ran some simulations and found the stability would degrade too far for it to be possible.

[pests]

This is getting down to speed of light and the distance to the ram. The RAM needs to be very close to the chip.

This is also being seen in the X3D CPU's where the memory is stacked on top to reduce latency and as density was maxed in 2d space.

[SSLy]

There isn't any standard to put that kind of RAM on DIMM-like slots.

[TheAmazingRace]

Honestly... I wonder why they didn't consider CAMM2? The bandwidth and signal integrity rivals that of soldiered memory. They could have had the best of both worlds, really.

[rangestransform]

AFAIK you would need 4 CAMM modules for that kind of bandwidth, good luck stuffing that onto a board at all, let alone for a reasonable price

[anticensor]

CAMM, unlike DIMM, is designed to be amenable to stacking from the get go.

[chipsa]

A single LPCAMM2 module is supposed to be all the RAM for a laptop, so it’s 128 bit wide. You’d need two LPCAMM for this, not four. And they are currently available in 32G and 64G varieties ($180/$330). So they aren’t absurdly priced, if they were in stock.

[piskov]

They went for local LLM route and for that high-bandwidth memory is a must.

Consider it a low-price alternative to mac mini or nvidia’s box.

This can also be chained though not as effectively as macs for example (those have thunderbolt for interconnect)

[awiesenhofer]

> low-price alternative to mac mini

This starts at 1099$ per the article, so the mac mini is the low-price alternative here.

[kllrnohj]

This starts with 32GB of RAM, so no the mac mini isn't the low priced alternative at that point.

Also fully spec'd at 128gb is $2000, which only gets you a 64gb mac mini

[TiredOfLife]

The max config with 128gb ram is half the price of mac and 8tb ssd is 800 and not 2000

[MindSpunk]

If you can find me a Mac Mini with 32GB of RAM I'd love to see it.

[zamadatix]

https://www.apple.com/shop/buy-mac/mac-mini/apple-m4-chip-wi...

Select 32

It's not exactly an out of the park win though as it's only $100-$200 cheaper and it trends poorly the farther from comparing base models you go (particularly after 64 GB, for which you need to leave the Mini family for the Studio). By the time you get to 128 GB, which you'd want for the 70B class AI models, you end up back at the original statement.

[MindSpunk]

Huh. Ate my words then. I briefly checked Apple's page and didn't see them.

However those are the base M4 chip and not the M4 Pro. You need the M4 Pro to get competitive in GPU compute numbers for a more like to like comparison. The M4 Pro mini comes out at $1800 with the 48GB option, or $2000 for 64GB. For the same price the Framework machine gets you 128GB.

[zamadatix]

I think you're probably right comparing to the M4 Pro would make more sense but keep in mind you don't really need much compute, it's just that the M4 Pro has memory bandwidth more similar to the AI Max 395+ while the "normal" M4 doesn't.

Every large AI model is heavy memory bandwidth constrained to the point my 9800X3D (with the extra L3 cache) and 128 GB/s memory attached is only 60% utilized running a 32 GB model in CPU only mode (no NPU, iGPU, or GPU offload enabled). Really small AI models can start to be compute bound but, at that point, you don't really need the 32 GB of memory anymore and probably just want a normal GPU.

[rokweom]

Apparently this is due to signal integrity. AMD says swappable RAM is not possible. Source: LinusTechTips video.

[MostlyStable]

I saw that, and I'm curious if that's a death-knell for LPCAMM ram. My understanding was that the entire point of that new standard was to allow for the higher ram speeds and lower latencies etc that you would normally get with soldered ram, but in a modular, swappable package.

If LPCAMM already can't keep up with requirements when it is barely even out, then my guess is it won't fare well going forward.

So 1 of 2 things is probably true:

AMD is not being completely truthful with their statements that LPCAMM wasn't able to work (maybe it was just more difficult/complicated than they were willing to do, but it could work or

latency/speed requirements have already outpaced what LPCAMM can provide and soldered ram is the future.

I really hope it's the former, but it wouldn't be the first time something like the second has occurred. Apparently cache also used to be a separate, swappable component before it became integrated into the die. RAM might end up going the same way.

[AshamedCaptain]

The non-MAX Ryzen laptops also announced today actually use socketed RAM.

I guess they'd claim it is only the MAX AMD procs which force soldered RAM, but since they could as well have used a non-MAX chip (and correspondingly reduce the price) this just shows how much of this is an arbitrary, and therefore questionable, decision from Framework rather than any restriction AMD sets.

[sliken]

Yes, those are the ones with a 128 bit memory bus that can reuse designs from previous generations. Nearly every laptop and desktop has has 128 bit memory for the last few decades, the strix halo is the first with 256 bit wide x86 targeted at tablets, laptops, and SFFs. Much like the m1/m2/m4 pro. The M3 pro for some reason decided on 192 bits wide.

[SSLy]

Yes, and they have less BW than this solution.

[braiamp]

Yeah, in LTT video they are talking with the CEO https://youtu.be/-lErGZZgUbY?t=447

[aitchnyu]

Apparently LPCAMM2 (brand new upgradable RAM) modules wont support 256 bits width soon.